Apple files hint at re-engineered iMac and Mac Pro models, potentially without optical drives

17891012

Comments

  • Reply 221 of 257
    hmmhmm Posts: 3,405member

    Quote:

    Originally Posted by Marvin View Post





    Yeah, people will bring that up again but if they laminate the glass, it should have the same effect as the rMBP. There's a massive reduction in glare to the point that it shouldn't be an issue any more.





    http://www.anandtech.com/Show/Index/6023?cPage=2&all=False&sort=0&page=4&slug=the-nextgen-macbook-pro-with-retina-display-review

    IMO, the first image shows they've got the best of both. The old-style displays on the left were like mirrors and you can't see a thing in certain conditions. On the right, the anti-glare shows all the content ok but it's faded. In the middle image, you can see the content in vibrant colours with small amounts of distracting glare.

    In normal conditions, eliminating glare should at least be a lot easier. Hopefully they will allow running the display at 1080p too using scaling to bring text sizes up a bit.


     


    I've seen the new ones. The glare is nowhere near as bad. If we're talking about anyone using a similar display for content creation or editing, it should be doable assuming some basic control over the lighting in their work area. If you've got a huge open window, it is bad no matter what on any display.

  • Reply 222 of 257
    nhtnht Posts: 4,522member

    Quote:

    Originally Posted by Marvin View Post



    How does increasing performance by an order of magnitude, reducing the price, making parallel computing simple and doubling available expansion ports **** up the Mac Pro? Ah you mean because it's not exactly the same design as last year's model with a fractional improvement in performance. To me, that would be ****ing up the Mac Pro. If you want the Mac Pro to die out, fine, keep hoping for that same design and a 40% performance jump after 3 years.


     


    Because your move to TB vs slots reduces the available bandwidth from 16 lanes to 4 lanes for high speed uses (GPUs, dedicated video transcoders, etc).


     


    There's no reason to drop slots except that you want a smaller xMac which again, no one else gives a shit about except that it would be nice if the Mac Pro was rack mountable.


     


    Everything you want can be done with the mini.  Small footprint, HPC computing modules connected via TB, etc.

  • Reply 223 of 257
    hmmhmm Posts: 3,405member

    Quote:

    Originally Posted by nht View Post


     


    Because your move to TB vs slots reduces the available bandwidth from 16 lanes to 4 lanes for high speed uses (GPUs, dedicated video transcoders, etc).


     


    There's no reason to drop slots except that you want a smaller xMac which again, no one else gives a shit about except that it would be nice if the Mac Pro was rack mountable.


     


    Everything you want can be done with the mini.  Small footprint, HPC computing modules connected via TB, etc.



    It doesn't just reduce slots. You need enough PCI lanes for whatever will be implemented. They're located in the cpu package, so you have 40 lanes in a single package model. PCI 3 does give you more per lane, but cards are still designed for the same lane counts. x16 PCI 2 is still designed as x16 PCI 3. The others seem to follow the same basic rule. I don't think heavy use of thunderbolt would do anything pleasant to the cost either, given that each of these chips is quite expensive and apple doesn't like to give anything out when it comes to the mac pro.

  • Reply 224 of 257
    MarvinMarvin Posts: 15,360moderator
    nht wrote:
    Because your move to TB vs slots reduces the available bandwidth from 16 lanes to 4 lanes for high speed uses (GPUs, dedicated video transcoders, etc).

    Edit: The current Mac Pro has one x16 and two x4 (non-configurable). If the GPU internally uses 16 lanes, on a single CPU model, you get 6 x 4 lanes left, which is enough for 6 x4 speed Thunderbolt ports. The only thing you lose vs the current Mac Pro is one x16 slot but you get more available expansion ports.

    Running a GPU on the outside isn't going to be common just like putting multiple GPUs in a Mac Pro isn't so we're talking about peripheral cards and adaptors like fibre channel, which will work just fine over 20Gbps TB ports.

    I don't think they will leave Thunderbolt out of the Mac Pro so if they retain expansion slots, they will need a setup like the following:



    The GPU will need software to route the framebuffer through the IGP and in the above scanerio, they are virtualizing the GPU:

    http://www.lucidlogix.com/technology-virtual-graphics.html

    Apple can of course do this and obviously they already allow their consumer lineup to be able to switch between dedicated and integrated but they'd need to go further with a machine where you could have multiple dedicated GPUs and need to route the display output in multiple ways.

    I think a simpler route is to get rid of the slots, have an internal GPU that can be upgraded like the iMac's GPU but doesn't have external video outputs. I think it can easily be a mobile GPU. In 2013, an 8970M will be pretty fast. Even though higher-end desktop cards can exceed their compute power, if they add an Intel MIC, that's a better option for compute anyway.

    With either 4 or 6 Thunderbolt ports, people will connect to RAID, cameras and displays. Mostly transcoding and audio processing will be done natively on the fast internal hardware but the TB ports will allow people to connect external cards via a PCI box if that situation arises.
  • Reply 225 of 257
    nhtnht Posts: 4,522member

    Quote:

    Originally Posted by Marvin View Post



    You don't have x16 available in the current Mac Pro. You get one x16 for the GPU, one x8 and two x4. If the GPU internally uses 16 lanes, on a single CPU model, you get 6 x 4 lanes left, which is enough for 6 x4 speed Thunderbolt ports. The only thing you lose vs the current Mac Pro is an x8 slot but you get more available expansion ports.


     


    You don't have to use the x16 for the GPU if you don't want to.  In any case x8 > x4 and x8 cards are common in the video world.  You also have the flexibility of configuring the Mac Pro as 3 x8 slots and one x1 slot.


     


     


    Quote:

    I think a simpler route is to get rid of the slots, have an internal GPU that can be upgraded like the iMac's GPU but doesn't have external video outputs. I think it can easily be a mobile GPU. In 2013, an 8970M will be pretty fast. Even though higher-end desktop cards can exceed their compute power, if they add an Intel MIC, that's a better option for compute anyway.


     


    Upgrades that largely don't exist.  And instead of a driver and a firmware flash on a normal pro card GPU manufactures have to design for a much smaller footprint in a non-standard mezzanine format.  Meaning it wont happen and the number of GPU options will be even lower.


     


    Again, this is nothing that cannot be done in a Mini.  Adding a mezzanine based GPU and a MIC (if that's actually worthwhile) is more than possible in a mini the size of the older Minis.  Or simply attaching a MIC compute unit via TB if you think TB actually provides sufficient bandwidth...which it doesn't to get full speed but you seem to think it will.


     


    Have you coded for a MIC?  Do any tools support a MIC?  Would it not be more sane to simply slot in a MIC in the Mac Pro as opposed to **** everyone with existing cards like the Rocket or Tesla cards while the transistion occurs?  Especially since there exists Knights Ferry PCIe cards and there will be Knights Corner PCIe cards?


     


    http://blogs.intel.com/technology/2012/06/intel-xeon-phi-coprocessors-accelerate-discovery-and-innovation/


     


    The "simpler route" my ass.

  • Reply 226 of 257
    hmmhmm Posts: 3,405member

    Quote:

    Originally Posted by Marvin View Post





    You don't have x16 available in the current Mac Pro. You get one x16 for the GPU, one x8 and two x4. If the GPU internally uses 16 lanes, on a single CPU model, you get 6 x 4 lanes left, which is enough for 6 x4 speed Thunderbolt ports. The only thing you lose vs the current Mac Pro is an x8 slot but you get more available expansion ports.

    Running a GPU on the outside isn't going to be common just like putting multiple GPUs in a Mac Pro isn't so we're talking about peripheral cards and adaptors like fibre channel, which will work just fine over 20Gbps TB ports.

    I don't think they will leave Thunderbolt out of the Mac Pro so if they retain expansion slots, they will need a setup like the following:



    The GPU will need software to route the framebuffer through the IGP and in the above scanerio, they are virtualizing the GPU:

    http://www.lucidlogix.com/technology-virtual-graphics.html

    Apple can of course do this and obviously they already allow their consumer lineup to be able to switch between dedicated and integrated but they'd need to go further with a machine where you could have multiple dedicated GPUs and need to route the display output in multiple ways.

    I think a simpler route is to get rid of the slots, have an internal GPU that can be upgraded like the iMac's GPU but doesn't have external video outputs. I think it can easily be a mobile GPU. In 2013, an 8970M will be pretty fast. Even though higher-end desktop cards can exceed their compute power, if they add an Intel MIC, that's a better option for compute anyway.

    With either 4 or 6 Thunderbolt ports, people will connect to RAID, cameras and displays. Mostly transcoding and audio processing will be done natively on the fast internal hardware but the TB ports will allow people to connect external cards via a PCI box if that situation arises.




    The Xeon class they've used has no IGP. there's no room in the cpu package. You can use an embedded gpu, but once you go to LGA2011, those chips have no integrated graphics included. Some of the cheaper Xeons are based on a different socket shared with the mainstream i5s. This is how HP offers an integrated option in their Z1. Beyond a certain point they don't exist. You would have to go with embedded graphics, and you do pick up restrictions in board design due to this. Multiple gpus are a bit problematic outside of dual socket boards. The PCI lanes are on the cpus. With a single package you get 40 of them that are already tapped out in the current design given the lanes that must be dedicated to existing ports. USB3 would be there. Firewire is still likely to be there. As much as you may hate it, I will be surprised if ethernet goes anywhere.


     


    Also the PCI box is a very poor idea. They've been extremely limited so far in terms of what they will support, and you spend way more for an inferior product.  Thunderbolt lacks the adoption rates and options. The chips are more expensive, yet it's one of the least used connection types. Trying to push that on a line that has become somewhat of a niche product while ruling out some of the extreme bandwidth options that were previously available such as mini SAS or cheaper (6GB eSATA) is just going to further damage such a platform, as Apple has no way of motivating third party development to the necessary levels. Notice how last year everyone was asking when more things would become available. They're available today, yet similar solutions are much more costly compared to what they were without pushing the available speeds any further. The reason most people on here wouldn't know that is because Apple was notorious for a lack of fast IO options. Right now on the consumer end it can't compete with usb3 on price. If we're getting into high bandwidth needs of must view uncompressed footage, better options existed prior to TB. The problem here is that even if such a package became better supported over time, the potential to hemorrhage off customers and defeat the line is pretty much there, especially as the other lines currently outpace it in overall growth. Windows lines still support their workstations more than Apple, but this is because they make large margins on them. In a lot of ways they're better at that kind of thing. They can get replacement parts out overnight. Workstation support is typically within the US rather than outsourced and if it's not 24/7, it at least runs for fairly long hours. Given some of these things, I don't really see Apple stealing a lot of marketshare on that end unless something is extremely desirable in terms of software. I still see quite a few mac pros around. It's just that most of them were purchased several years ago, as there hasn't been a lot of meaningful cpu growth in that area over the past several years unless you go to escalating price points. The reasoning there was split between intel and Apple, but I won't go into that. Sandy Bridge E would have been a moderate bump for some things. For others which actually take advantage of updated hardware instructions, it's huge. Unfortunately you're unlikely to use it to its full potential today.

  • Reply 227 of 257
    MarvinMarvin Posts: 15,360moderator
    nht wrote:
    In any case x8 > x4 and x8 cards are common in the video world.

    If they use more than 20Gbps of bandwidth, then they'd run slower but I doubt they would use that much bandwidth. That's enough for multiple uncompressed 4K streams and as I say, data has to come from somewhere and go somewhere else. The channel bandwidth only matters if the source and destination are fast to enough to cope.

    Give me an example of data that can be read and written at that data rate and you'd be describing a scenario that applies to almost nobody.
    nht wrote:
    You also have the flexibility of configuring the Mac Pro as 3 x8 slots and one x1 slot.

    It looks like only a couple of Mac Pros had configurable slots - the latest ones have two x16 and two x4 that are fixed.
    nht wrote:
    Upgrades that largely don't exist.

    They would be available when a new Mac pro arrived like the current setup.
    nht wrote:
    And instead of a driver and a firmware flash on a normal pro card GPU manufactures have to design for a much smaller footprint in a non-standard mezzanine format.  Meaning it wont happen and the number of GPU options will be even lower.

    Ok but that doesn't bother Apple and how many people are flashing GPUs anyway? Are people really prepared to blow thousands on a high-end machine and shove in a GPU with custom firmware that isn't certified for the machine?
    nht wrote:
    Adding a mezzanine based GPU and a MIC (if that's actually worthwhile) is more than possible in a mini the size of the older Minis.

    Not with an 85W PSU and the heat would be way too much.
    nht wrote:
    Or simply attaching a MIC compute unit via TB if you think TB actually provides sufficient bandwidth...which it doesn't to get full speed but you seem to think it will.

    The bandwidth doesn't matter because it has local storage (8GB RAM) so it would be possible to do this but the Mini target audience is not likely to be looking for an MIC. No doubt someone will get a Knights Ferry card and put in a PCI box though.
    nht wrote:
    Would it not be more sane to simply slot in a MIC in the Mac Pro as opposed to **** everyone with existing cards like the Rocket or Tesla cards while the transistion occurs?

    I don't see how it affects people with existing cards. The Rocket can be connected externally and the MIC would outperform the Tesla card.

    Since when does Apple do transitions anyway? They didn't transition people to FCPX, they yank out the rug and people get back up all by themselves. The reason is that it's the only way to get people to move. If an MIC is optional, they won't easily convince people to code for it. People generally like having a comfort zone and will do anything to hold onto what's familiar so that nobody has to expend any effort making a change for the better.

    The reason it's a change for the better is because it will drive down the cost of high performance and it will make high speed peripheral expansion simple enough for anyone as well as universal so that it drives down the price there too.
  • Reply 228 of 257
    MarvinMarvin Posts: 15,360moderator
    hmm wrote:
    The Xeon class they've used has no IGP. there's no room in the cpu package.

    Well that settles it then, either GPUs can't connect externally or the MP won't support Thunderbolt. I honestly can't see Apple shipping a properly new MP without Thunderbolt support.
    hmm wrote:
    USB3 would be there. Firewire is still likely to be there. As much as you may hate it, I will be surprised if ethernet goes anywhere.

    I doubt FW800 will be on the next MP, iMac or Mini. Ethernet is useful for internal networking so I'd expect it on the desktops, I just think it doesn't make sense on the mobile products any more.
    hmm wrote:
    Also the PCI box is a very poor idea.

    You are coming at this from the point of view that PCI expansion is a fundamental requirement for the Mac Pro audience and would require a significant portion of that audience to invest in a poor solution. I don't think internal PCI expansion is a requirement any more. It's not going to be ideal for those who do have PCI cards and don't have a convenient Thunderbolt equivalent but it really doesn't matter any more - those users don't dictate Apple's product designs just like people who worked with tape didn't dictate the design of FCPX.
  • Reply 229 of 257
    hmmhmm Posts: 3,405member

    Quote:

    Originally Posted by Marvin View Post





    If they use more than 20Gbps of bandwidth, then they'd run slower but I doubt they would use that much bandwidth. That's enough for multiple uncompressed 4K streams and as I say, data has to come from somewhere and go somewhere else. The channel bandwidth only matters if the source and destination are fast to enough to cope.




     



    I explain this below, but your suggested 20Gb aren't all data. If you could really drive that much data to a single device over thunderbolt, it would be different. Only 2 channels can be used for normal data. I don't think the card can cannibalize the others for display bandwidth. If that was possible we might actually see 10 bit RGB support and displayport 1.2 compliance. Adobe's forums keep acquiring comments on it until they mentioned that Apple has decided not to support it (which annoys me, as it delivers nice shadow detail on Adobe RGB displays).


     


     


    Quote:


     


    It looks like only a couple of Mac Pros had configurable slots - the latest ones have two x16 and two x4 that are fixed.




    That's what they have on paper. The reality is that at least the quad and hex core are over subscribed. If you populate all your ports, something will throttle. It's not that uncommon. They do the same thing with macbook pros and chargers. If you run them hard enough your battery life will drain over time. I don't even have a problem with this. I just wish Apple would be more open about it rather than wait for users to uncover such details. You listed 40 lanes, yet the integrated ports that ship standard all take up PCI lanes. If I recall correctly that chipset has a total of 40. With Sandy if you're looking at the dual models as in 12-16 cores, you get 80 lanes instead.


     


     


     


    Quote:


    Originally Posted by Marvin View Post





    Well that settles it then, either GPUs can't connect externally or the MP won't support Thunderbolt. I honestly can't see Apple shipping a properly new MP without Thunderbolt support.


    I'm not sure what you mean there. I just said the chips used in the mac pro have their silicon allocated differently. In their "mainstream" (using wiki's term for it) they do allocate space for graphics, yet the Xeon chips that Apple has used to date are different. I know HP has an igp as part of their configuration matrix in the Z1, yet it's also offered with something like i3s at the low end. I'd have to look it up again, but I think they may be using the lighter Xeons based off the mainstream line. It would make sense as they'd fit the same socket as the i3s in their starting configuration.


     


    Quote:


    I doubt FW800 will be on the next MP, iMac or Mini. Ethernet is useful for internal networking so I'd expect it on the desktops, I just think it doesn't make sense on the mobile products any more.





    I don't see it as so much doubtful. I see it as a low priority. It's still in the thunderbolt display. It still exists in the cMBP. Remember they removed FW400 from the white macbook a few years ago without redesigning it. In this case I don't think they'll try to stamp it out so much as they are unlikely to design around FW800 as a requirement. I've got a couple random FW things around. Some are broken. The rest can also use usb. I looked for that as I'm never sure what will outlive what. Some stuff I replace quickly. Other things.... well... *stares over at 10 year old document scanner and outdated Epson wide format printer with pesky RIP dongle*. Sometimes something is expensive and it's gotten to a point where I don't use it enough to justify a costly update, yet it's used a few times a year so it remains. I have an old tape drive here too xD. I can't remember the cartridge type, but that's actually destined for e-recycling along with a few other things.


     


    Quote:


    You are coming at this from the point of view that PCI expansion is a fundamental requirement for the Mac Pro audience and would require a significant portion of that audience to invest in a poor solution. I don't think internal PCI expansion is a requirement any more. It's not going to be ideal for those who do have PCI cards and don't have a convenient Thunderbolt equivalent but it really doesn't matter any more - those users don't dictate Apple's product designs just like people who worked with tape didn't dictate the design of FCPX.


     






    I don't think it makes sense to sell an external chassis for $500~ or so that supports only some cards at a limited wattage range. Much of this should become easier as TB bandwidth increases assuming development becomes cheaper. Otherwise it can still be an okay solution if companies take control of the entire product and release a thunderbolt version, not X product from this vendor and a generic untested housing from another. There are many cheap do it yourself storage chassis solutions out there. A couple are tolerable, but they're mostly dreadful. Making PCI expansion into something like that isn't really a great solution at all. This is why I'm saying if it really goes that way, it makes more sense the way Black Magic does it. Right now thunderbolt options are still fairly poor though. USB3 and eSATA both offer storage that is nearly as fast in real world transfer speeds for less. Mini SAS remains a better solution if you're looking for extreme performance. The figures you quoted for TB actually assigned the bandwidth allocated for the display to data, so you can't really claim 20Gb if it won't reach that. 10 can be allocated to data, and you actually get less than that even under ideal circumstances. The bandwidth was basically displayport 1.2 with slightly lower processing overhead split into data and display data channels. Note that displayport 1.2 is around 16Gb, but that's due to overhead.


     


    Anyway the important thing here is that they design a computer which actually has a market going forward. This means that if it's based on thunderbolt, thunderbolt solutions need to be at least marginally competitive with existing standards. I don't see them being that great prior to 2015 or so. When bandwidth hits a point where they can easily support anything even with the inefficient restrictions it places on the use of bandwidth, which you ignored, and we see cables that won't go out of date on an annual basis due to bumped standards, it could see some decent adoption rates. Right now it's a very clunky solution that just looks pretty on paper. I'm not sure what intel will do. They haven't shown any plans to include it in chipsets. Even next year the ivy options the Mac Pro could use don't have it scheduled to be integrated with the chipset. The lighter ivy xeons based off the consumer line might include a displayport connection by default, but that's just building overlap with the imac in terms of computing power.

  • Reply 230 of 257
    nhtnht Posts: 4,522member

    Quote:

    Originally Posted by Marvin View Post





    If they use more than 20Gbps of bandwidth, then they'd run slower but I doubt they would use that much bandwidth.


     


    And 640K is plenty of memory.  20 Gbps TB wont be out until 2014 with Falcon Ridge...past the 2013 Mac Pro revamp.  So 10 Gpbs max.


     


     


    Quote:

    They would be available when a new Mac pro arrived like the current setup.

    Ok but that doesn't bother Apple and how many people are flashing GPUs anyway? Are people really prepared to blow thousands on a high-end machine and shove in a GPU with custom firmware that isn't certified for the machine?


     


     


    I'm not talking about users flashing a card but GPU manufacturers supporting a Mac Pro version that doesn't require creating a completely new form factor.


     


    So they wont be there when a new Mac pro comes out except for the very few that Apple pays for.


     


     


    Quote:

    Not with an 85W PSU and the heat would be way too much.

     


     


    Who says that the mini compute server has to stay with only a 85W PSU?  There is zero need to start with the Mac Pro and cripple it vs starting with the Mini and beefing it up.


     


     


    Quote:

    The bandwidth doesn't matter because it has local storage (8GB RAM) so it would be possible to do this but the Mini target audience is not likely to be looking for an MIC. No doubt someone will get a Knights Ferry card and put in a PCI box though.


     


    http://energy.gov/articles/high-school-students-build-their-own-supercomputer-almost


     


    If bandwidth doesn't matter then a TB based MIC or Tesla module can be used with the Mini, iMac or MBP.  No need for building it into the Mac Pro since not every Mac Pro user needs it.  The target audience for MICs is tiny.


     


     


    Quote:

    I don't see how it affects people with existing cards. The Rocket can be connected externally and the MIC would outperform the Tesla card.


     


     


    It affects people with existing cards because the Mac Pro ceases to exist as an upgrade option.  If they have to move to an external enclosure they might as well just use a MBA.


     


    And Intel has not released performance benchmarks vs the Tesla so there is no way for you to assert that it can outperform the Tesla.  In fact the probability is that the K20 (1.3-1.5 DP TFLOP estimated) crushes Phi for things that can be translated into CUDA.  The key selling point for Knights Ferry/Phi is that there exists a large body of scientific code that will never get ported to GPGPU code due to cost.  THESE can more economically be ported to MIC code than to CUDA/OpenCL code...at least if your problem is embarrassingly parallel and you can actually get the 512-bit wide SMID cranking.  


     


    Sure porting code to simply run will be more or less a recompile.  Making the code performant is a whole different ballgame.


     


    Quote:

    The reason is that it's the only way to get people to move. If an MIC is optional, they won't easily convince people to code for it. People generally like having a comfort zone and will do anything to hold onto what's familiar so that nobody has to expend any effort making a change for the better.


     


    The MIC is completely optional because it wont be on the iMac, MBP, MBA or Mini...only your crippled Mac Pro that no one is likely to buy because there will be no apps to take advantage of the MIC and there are no slots for expansion.  If TB is the only expansion option then you can just use the high end iMac.


     


     


    Quote:


    The reason it's a change for the better is because it will drive down the cost of high performance and it will make high speed peripheral expansion simple enough for anyone as well as universal so that it drives down the price there too.



     


    Except it doesn't drive down the cost for "high performance" because there will be zero volume.


     


    In the 2013 timeframe the Xeon Phi still largely lives on a PCIe card.  Intel doesn't have any near term plans to stick it on the QPI interconnect with access to system memory so pray tell what advantage will Apple gain by grafting it into every Mac Pro and vastly inflating the price for little to no gain?  The Phi wont be any cheaper than a Kepler so it's like adding $2500 (and a huge hit to the TDP) to the base cost of every Mac Pro.


     


    MIC will drive down pricing of HPC when Intel bakes a number of MIC cores into the Xeon die or makes an economical 20-30 core version that lives on the QPI removing the need for 8GB of DDR5 RAM.

  • Reply 231 of 257
    MarvinMarvin Posts: 15,360moderator
    hmm wrote:
    I'm not sure what you mean there. I just said the chips used in the mac pro have their silicon allocated differently.

    In order for Thunderbolt to exist on the next Mac Pro, Apple has to route a displayport signal through it. They can only do this by having a motherboard GPU linked to it. The framebuffer then gets copied from a dedicated GPU into the IGP and pushed out the TB port. Without an IGP or fixed GPU, Apple can't take display data from a dedicated GPU and push it out over Thunderbolt. This leads to 3 possible outcomes:

    - the Mac Pro won't use Thunderbolt and can't use the Thunderbolt display
    - the Mac Pro will have Thunderbolt but won't have PCI slots
    - the Mac Pro will be discontinued
    hmm wrote:
    It's still in the thunderbolt display.

    That hasn't been updated though. I think we'll start seeing an extra Thunderbolt port replace FW800 on updated models.
    nht wrote:
    And 640K is plenty of memory.

    Ok, now replace 640K with 32GB and claim that the statement is just as ridiculous.
    nht wrote:
    20 Gbps TB wont be out until 2014 with Falcon Ridge...past the 2013 Mac Pro revamp. So 10 Gpbs max.

    Yeah they'd have to get an early implementation of the controller from Intel for late 2013.
    nht wrote:
    It affects people with existing cards because the Mac Pro ceases to exist as an upgrade option. If they have to move to an external enclosure they might as well just use a MBA.

    If you assume that the only value in a Mac Pro is the expansion slots then sure you can switch to a MBA.
    nht wrote:
    The Phi wont be any cheaper than a Kepler so it's like adding $2500 (and a huge hit to the TDP) to the base cost of every Mac Pro.

    If it's going to cost that much, it's not worth it. The costs listed here suggested it would be much less:

    http://www.hpcwire.com/hpcwire/2011-09-22/dell_to_build_10-petaflop_supercomputer_for_science.html

    "$2.5 million will go toward 8 petaflops worth of MIC coprocessors"

    We'll see in a couple of months what the price is per unit. Apple wouldn't be using the PCI card though and it would be a variant of the next version (Knights Landing) and if they integrated it, they'd be placing large orders.

    If we are to assume the next Mac Pro is what you guys are clearly driving at i.e:

    Ivy Bridge Xeon
    no Thunderbolt support
    PCIe 3 expansion slots
    no MIC, just a GPU like the Radeon 8970 (2013)
    same price points

    is that really what you want to see after a 3 year wait?
  • Reply 232 of 257
    nhtnht Posts: 4,522member

    Quote:

    Originally Posted by Marvin View Post



    If we are to assume the next Mac Pro is what you guys are clearly driving at i.e:

    Ivy Bridge Xeon

    no Thunderbolt support

    PCIe 3 expansion slots

    no MIC, just a GPU like the Radeon 8970 (2013)

    same price points

    is that really what you want to see after a 3 year wait?


     


    Regarding Thunderbolt there are several options:


     


    1) IGP on the motherboard


    2) external DP connection from a GPU to the chassis


    3) internal DP connection from a GPU to the motherboard


    4) data only TB


     


    If the next Mac Pro has Ivy Bridge Xeons, USB3, 4 PCIe 3 slots, updated GPUs and TB (in some form) in a package that is rack mount friendly at the same price points then I think most Pro users would be very happy.  


     


    Folks that want a MIC can simply slot one in.  


     


    The only viable reason to add a Mac Pro Lite to the line up is to provide a lower cost model to replace the current base model.  And that's ADD not replace the whole line up.


     


    Amusingly you could use the 45W 2.5 Ghz quad Xeon E3-1265L v2 in the Mini server and provide TB since it has a IGP in it (the Intel HD 2500) and actually do it in September 2012.  With 2 slots for ECC ram and it would be the smallest workstation available.


     


    Add this to the lineup for $1499 and folks would buy it in a heartbeat.  So MY wish list wouldn't be cripple the Mac Pro but to upgrade the Mini Server to this:


     


    2.5Ghz Quad Xeon E3-1265L w/Intel 2500HD


    2 slots for ECC RAM (4GB standard)


    2 slots for SSD sticks  (128GB SSD stick standard)


    1x USB 3


    3x USB 2


    2x TB


    1x 10GBase-T


    1x HDMI


    SDXC card slot


    Audio In/Out


    1x HDMI


     


    Fills the low end server and workstation niche.  


     


    For the light workstation users you do 10GbE to SAN, 1 TB dedicated to external GPU, 1 TB for external processing cards (rocket, tesla, xeon phi, etc).  If no SAN then USB3 or TB to local RAID.

  • Reply 233 of 257
    hmmhmm Posts: 3,405member

    Quote:

    Originally Posted by Marvin View Post





    In order for Thunderbolt to exist on the next Mac Pro, Apple has to route a displayport signal through it. They can only do this by having a motherboard GPU linked to it. The framebuffer then gets copied from a dedicated GPU into the IGP and pushed out the TB port. Without an IGP or fixed GPU, Apple can't take display data from a dedicated GPU and push it out over Thunderbolt. This leads to 3 possible outcomes:

    - the Mac Pro won't use Thunderbolt and can't use the Thunderbolt display

    - the Mac Pro will have Thunderbolt but won't have PCI slots

    - the Mac Pro will be discontinued

     


    I don't feel the thunderbolt display is a great match there, but some people would obviously buy it. Its design is more that of a docking station. For photography/videography (popular uses for the mac pro) I'd still regard it as a bit cold and defunct in some colors. It scores rather poorly on typical validations. If they did something similar to the treatment of the rMBP, it might be a better match in terms of color gamut for such fields. If it must go through integrated, that one still has an igp that isn't listed or used for anything else. I'm not kidding when I say it's an issue to place it on the current mac pro due to the socket type used. I don't think X79 can use embedded gpus anyway, even if that would work. They definitely don't include one, so unless you want to use the same cpus as those that are in the imac, you can't do this. Beyond that if you push the imac to a 6 core machine as you suggested in another post, assuming you wanted to do it next year, you'd lose the ability to run thunderbolt over an igp.


     


    Quote:


    That hasn't been updated though. I think we'll start seeing an extra Thunderbolt port replace FW800 on updated models.



     


    They put it on a new product in 2010 ish that's really a docking station. Like I said, I see it as more not a design priority that it must be included than a crusade against it at this point. It is probably the lowest priority port on the list.

     


     


    Quote:


    Ok, now replace 640K with 32GB and claim that the statement is just as ridiculous.



     


    Something that is interesting there is that many of us were bottlenecked on ram for years and forced to use scratch disk systems or heavy virtual memory for many applications. It hasn't moved forward much, yet it's possible to use real ram instead of virtualized ram.

     


     


    Quote:


    Yeah they'd have to get an early implementation of the controller from Intel for late 2013.



     


    I really don't see this happening. I agree they could push the mac pro as far as late 2013, although I'm not completely sure whether it will survive that long. The other concern would be if it can be made to work with the X79 chipset while still lacking an igp. It's just a really bad mismatch at the moment unless intel certifies a data only thunderbolt. The IO problems that exist on the others macs aren't really an issue for the mac pro anyway at this moment, so it's not so much fixing a problem. The only motivation would be a desire to unify how things are handled across the lineup.


     


    Quote:


    If you assume that the only value in a Mac Pro is the expansion slots then sure you can switch to a MBA.



     


    It's obvious the macbook aiir and updated macbook pro design are bigger design priorities for Apple, but this still makes little sense. The mac pro does solve certain problems that would be present on the notebook. There are certain areas where they're just a bad match. As an example, Apple likes to keep their power chargers light for portability reasons. If you run the computer to its limits on cpu, even with a moderate gpu + display load, it will have to pull from the battery. There are plenty of other examples. The Air still doesn't support OpenCL, and it definitely lacks a CUDA option. The Mac Pro could use better CUDA options too. While more things are going toward OpenCL, some still exist on CUDA simply because NVidia started first in this regard. Something that occasionally seems to be missed with a workstation class machine is that they're designed to be able accommodate a range of solutions and maintain a high level of up-time beyond the obvious implied requirement of greater processing power. That design isn't so much the question. The only question would be if Apple will choose to address that. If they don't, anyone who would have bought one will buy something else from Apple, or they'll move on. The other computers in the line have grown to a point where they can definitely satisfy some of these requirements, yet they aren't a perfect match as they're designed from a different perspective. Given Apple's size, the desire to chase further growth dictates that they need to consider large markets. This isn't to say that the other oems don't do well selling workstations. It's just that they likely satisfy many corporate accounts at both a respectable volume and high markup with these machines.

  • Reply 234 of 257
    MarvinMarvin Posts: 15,360moderator
    nht wrote:
    Regarding Thunderbolt there are several options:

    1) IGP on the motherboard
    2) external DP connection from a GPU to the chassis
    3) internal DP connection from a GPU to the motherboard
    4) data only TB

    1. They'd have to use an NVidia or AMD IGP assuming they are certified for use with that setup and ship a second dedicated GPU.
    2. Apple would never do an external connection
    3. Apple might do an internal connection but it means that without the cable, the Thunderbolt ports are not Thunderbolt ports, same with option 2. Plus it needs support on the GPU.
    4. Can't happen as they won't be allowed to call them Thunderbolt ports

    The only viable option I see for TB assuming PCI slots remain is if Ivy Bridge Xeons will have IGPs.
    nht wrote:
    If the next Mac Pro has Ivy Bridge Xeons, USB3, 4 PCIe 3 slots, updated GPUs and TB (in some form) in a package that is rack mount friendly at the same price points then I think most Pro users would be very happy.

    Even given that IB is only 15% faster than SB? SB was 40% at most over the last model so after 3 years, they jump 1.4 * 1.15 = 61%.

    Given that they aren't putting resources into an IGP, an IB Xeon could probably jump further but the biggest problem I have with this outcome is that it would mean there was no reason for avoiding Sandy Bridge Xeons. The refresh that just happened is pretty clear evidence that this isn't going to be a straight upgrade or they would have made one already.

    Why delay a proper refresh by another 1.5 years when people were already desperate for an update if it was just going to be minor?
    nht wrote:
    Mac Pro Lite / cripple the Mac Pro

    The entry model would have a 6-core Xeon, a fast GPU, up to 64GB RAM, possibly a co-processor and at least 4 TB ports. The co-processor would make up for the lack of a 2nd CPU. While it might seem crippled to people who buy dual-CPU models and fill up the PCI slots with cards, to most Mac pro buyers, it will offer better value for money. The few who will be disgruntled will find a way to make it work and I think they'd still be very happy with it.
  • Reply 235 of 257
    hmmhmm Posts: 3,405member

    Quote:

    Originally Posted by Marvin View Post





    1. They'd have to use an NVidia or AMD IGP assuming they are certified for use with that setup and ship a second dedicated GPU.

    2. Apple would never do an external connection

    3. Apple might do an internal connection but it means that without the cable, the Thunderbolt ports are not Thunderbolt ports, same with option 2. Plus it needs support on the GPU.

    4. Can't happen as they won't be allowed to call them Thunderbolt ports

     


     


    I still see thunderbolt at the moment as more nice to have than absolutely necessary. Intel doesn't really have a current strategy for integration into Xeon boards for this year or next year. Note that Ivy isn't coming until probably the second half, and they're still using the X79 chipset. it's likely that workstation boards will remain roughly the same just like they did with nehalem ---> westmere for stability + cost reasons pushing such a change in design to at least the Haswell version. I kind of doubt that will see a displayport connection on the chipset or igp. Eventually such a thing could be possible if they're making it more of a co-processor in the longer term, but unless it's a higher priority for intel, it may come after a later die shrink. The problem I see is that you're trying to create a problem set for it, mainly uniformity across the line. Assuming some kind of breakout box functionality is the issue, such a thing could be adapted to also fit a more mac pro friendly port type. I don't see external PCI chassis designs becoming viable as shells in which to stick off the shelf boards. That is just backwards design to compensate for lack of a better option from Apple. I've mentioned before that it makes more sense for it to be sold as a complete unit rather than two separate purchases. It alleviates concerns over whether a card will fit and whether the maximum power consumption falls within the stated power levels of the chassis. Typically on the supply side they're labeled by the peak wattage that can be supplied to underlying hardware, yet the tdp numbers of many of these cards don't necessarily indicate peak power consumption. The risk of a bad match due to overlap and an overall untested solution just makes this a very poor choice. In the typical tower scenario, it's possible to allocate quite a lot of fan power/ventilation back there compared to the restricted space of some of the products you've previously mentioned.


     


     


    Quote:


    The only viable option I see for TB assuming PCI slots remain is if Ivy Bridge Xeons will have IGPs.

    Even given that IB is only 15% faster than SB? SB was 40% at most over the last model so after 3 years, they jump 1.4 * 1.15 = 61%.

    Given that they aren't putting resources into an IGP, an IB Xeon could probably jump further but the biggest problem I have with this outcome is that it would mean there was no reason for avoiding Sandy Bridge Xeons. The refresh that just happened is pretty clear evidence that this isn't going to be a straight upgrade or they would have made one already.

    Why delay a proper refresh by another 1.5 years when people were already desperate for an update if it was just going to be minor?

    The entry model would have a 6-core Xeon, a fast GPU, up to 64GB RAM, possibly a co-processor and at least 4 TB ports. The co-processor would make up for the lack of a 2nd CPU. While it might seem crippled to people who buy dual-CPU models and fill up the PCI slots with cards, to most Mac pro buyers, it will offer better value for money. The few who will be disgruntled will find a way to make it work and I think they'd still be very happy with it.



     


    There is absolutely nothing to indicate this, and beyond that the chipsets and reference X79 board designs lack an on board displayport connection. You probably think I'm trolling you, but I'm just looking at what is actually manufactured and the trends in behavior. On the Xeon end once you're out of those that share parts with the mainstream desktop line, they tend to push board designs for at least 2 years. There's very little chance of seeing an IGP there in the near future. Eventually die shrinks may get to a point where Intel decides to chase something other than ever increasing core counts. Right now the space simply isn't allocated to an IGP in these designs, and I find it difficult to believe they'll add this on something using the same chipset, especially given the potential for such an addition to introduce bugs and push the release back even further. Note what happened with Sandy Bridge and Sandy Bridge E. It is possible Apple has a workaround if thunderbolt is an absolute priority, but I don't see this as something that could be populated in a logical manner with existing thunderbolt options. Regarding 4TB ports, you would be able to get that if slots were completely removed, yet peripherals are still lacking. A co-processor isn't really the same. You won't pick up PCI lanes from that, so you are working with a budget of 40 lanes. This must cover all usb, thunderbolt ports, graphics, ethernet options, etc. I still don't think they'd go with that many though. Apple is always stingy on ports. I always end up with pesky dongles and things to account for, so it may affect me more than some users.

  • Reply 236 of 257
    nhtnht Posts: 4,522member

    Quote:

    Originally Posted by Marvin View Post





    1. They'd have to use an NVidia or AMD IGP assuming they are certified for use with that setup and ship a second dedicated GPU.

    2. Apple would never do an external connection

    3. Apple might do an internal connection but it means that without the cable, the Thunderbolt ports are not Thunderbolt ports, same with option 2. Plus it needs support on the GPU.

    4. Can't happen as they won't be allowed to call them Thunderbolt ports


     


     


    The trade off between 2 and 3 is the ability to support any GPU vs ones designed for internal DP passthough to TB.


     


    4 is simply option 2 or 3 without the wire connected and could still be called Thunderbolt in as much as 2 is allowed with the option of not making the connection:


     


    "Asus is demonstrating at Computex 2012 the first add-in card to extend its motherboards with Thunderbolt support. Available in July for about $200, the ThunderboltEX PCIe card is connected to a graphics card or onboard graphics via DisplayPort. The loop-through enables users to combine Thunderbolt with a powerful graphics card. According to Asus, the Thunderbolt card also functions without the DP port connection; however, in this case, it will only be able to support regular data transfers to other Thunderbolt devices, but will not support a Thunderbolt display"


     


    http://www.tomshardware.com/news/asus-thunderbolt-Z77-H77-pcie,15925.html


     


    If there's no displayport data then there's no displayport data.  Pro users that use their own GPUs will figure this out pretty easily.


     


    As far as Apple never doing an external connection I could counter with Apple would never do a rats nest of 6 TB either.


     


    Quote:

    The only viable option I see for TB assuming PCI slots remain is if Ivy Bridge Xeons will have IGPs.


     


     


    Some of the E3 IB Xeons do but I don't believe that any of the E5s have them.  Possible though.


     


    But it's not the only viable option.  Internal DP passthrough would be just fine for GPUs that support it and if not then you simply don't get video over TB just like on the ASUS card.


     


    Quote:

    The refresh that just happened is pretty clear evidence that this isn't going to be a straight upgrade or they would have made one already.

    Why delay a proper refresh by another 1.5 years when people were already desperate for an update if it was just going to be minor?


     


     


    Doesn't mean they are going to remove all the slots on the MP in 2013.


     


    Quote:


    The entry model would have a 6-core Xeon, a fast GPU, up to 64GB RAM, possibly a co-processor and at least 4 TB ports. The co-processor would make up for the lack of a 2nd CPU. While it might seem crippled to people who buy dual-CPU models and fill up the PCI slots with cards, to most Mac pro buyers, it will offer better value for money. The few who will be disgruntled will find a way to make it work and I think they'd still be very happy with it.



     


    The "co-processor" is a ~$2500 item in LATE 2012 designed for the price insensitive HPC market.  Yields on a 50+ core chip isn't going to stellar and numbers produced will be quite small.  So by 2013 when the Mac Pro comes out the price will STILL be high meaning your "entry level" Mac Pro isn't $2,499 anymore but $4,499 with no slots and needs a $500 to $1000 expansion chassis to use any existing PCIe cards at greatly reduced speeds (since the x4 lane is shared among all the cards in the chassis).


     


    Nothing you propose to remove actually costs all that much money.  What you propose to add is expensive as hell.


     


    So you are proposing only replacing the somewhat affordable entry level model with your super expensive but gimped Mac Pro?  Fewer people would be pissed I guess.


     


    Far better to add a $1499 mini with a quad core Xeon and ECC ram.

  • Reply 237 of 257
    hmmhmm Posts: 3,405member

    Quote:

    Originally Posted by nht View Post


     


     


     


     


    Some of the E3 IB Xeons do but I don't believe that any of the E5s have them.  Possible though.


     


    But it's not the only viable option.  Internal DP passthrough would be just fine for GPUs that support it and if not then you simply don't get video over TB just like on the ASUS card.


     



    This sort of goes back to what I mentioned. The Xeons based on mainstream cpus which are single package only, limited in PCI lanes, and similar to what you see over the 21-27" imac lineup do have IGPs and built in displayport connections.


     


    http://www.tomshardware.com/reviews/xeon-e3-benchmark-review,3258.html


     


    Apple has never used these in the past. They are seen on some lighter/inexpensive workstations. I don't see where anyone got the idea that the price of a mac pro is due to lots of aluminum + PCI slots. Pricing is really driven by high markup on the mac pros. They used the daughter board + backplane configuration as it saves the cost of a dual package chipset for the single models. A few silly people on here suggested that it still has to use a dual board, which isn't really true due to that design. I doubt the volume is even as dismal as many would suggest aside from long un-updated periods. Much of the time when the notebooks / desktop sales graphs come out for reference, people fail to note that the desktops still move units, but they are merely outpaced on growth. In terms of getting a mac pro out, you need probably one person to work on hardware. Many board designs aren't produced by huge teams. You need one or two people allocated to drivers/firmware issues.


     


    Look at any other workstation on the market. Reference board designs see very few annual changes. The chipset will not change, and Intel will not want to risk hiccups. There's really nothing special coming with Ivy. At the top end it migrates to 20 core machines. Below that it's really quite uninteresting. Chipset is the same. Dell, HP, etc. will most likely take the same board and add new cpus with updated firmware/bios tweaks as necessary. It's possible that Apple just changed their minds on whether or not to drop it. The sales volume is most likely still available. The issue is more likely to be one that other options have outpaced it on growth. In terms of development resources, if they can't find a couple guys to work on it, that is an Apple problem in terms of recruiting engineers.

  • Reply 238 of 257
    nhtnht Posts: 4,522member

    Quote:

    Originally Posted by hmm View Post


    This sort of goes back to what I mentioned. The Xeons based on mainstream cpus which are single package only, limited in PCI lanes, and similar to what you see over the 21-27" imac lineup do have IGPs and built in displayport connections.


     


    http://www.tomshardware.com/reviews/xeon-e3-benchmark-review,3258.html


     



     


    I'm not suggesting an E3 in anything but a Mini.


     


    IBM has some E3 based Xeon "budget workstations" in the $400 range.

  • Reply 239 of 257
    MarvinMarvin Posts: 15,360moderator
    nht wrote:
    4 is simply option 2 or 3 without the wire connected and could still be called Thunderbolt in as much as 2 is allowed with the option of not making the connection:

    Asus is demonstrating at Computex 2012 the first add-in card to extend its motherboards with Thunderbolt support.

    If there's no displayport data then there's no displayport data.  Pro users that use their own GPUs will figure this out pretty easily.

    Internal DP passthrough would be just fine for GPUs that support it and if not then you simply don't get video over TB just like on the ASUS card.

    http://www.tonymacx86.com/86-asus-thunderbolt-add-card-asrock-s-thunderbolt-motherboard-doesn-t-pass-certification.html
    nht wrote:
    As far as Apple never doing an external connection I could counter with Apple would never do a rats nest of 6 TB either.

    That's like saying they would never do a rats nest of USB, FW, ethernet and Mini-DP ports. It's only a rats nest if you actually connect things up to them.
    nht wrote:
    Yields on a 50+ core chip isn't going to stellar and numbers produced will be quite small.  So by 2013 when the Mac Pro comes out the price will STILL be high

    I don't know where the $2500 price is coming from as they haven't even hinted at pricing yet but you can't tell that the 2013 implementation would be just as expensive. Plus, Apple wouldn't be using the card. The problem with shipping an optional PCI card is that it doesn't drive down the price.

    I'd expect Apple to use an implementation that works directly with OS X and GCD and possibly even uses system memory instead of external memory. It doesn't have to be 50-cores either. A 25-core Knights Landing chip would be plenty fast to boost software performance.
  • Reply 240 of 257
    hmmhmm Posts: 3,405member

    Quote:

    Originally Posted by Marvin View Post





    I don't know where the $2500 price is coming from as they haven't even hinted at pricing yet but you can't tell that the 2013 implementation would be just as expensive. Plus, Apple wouldn't be using the card. The problem with shipping an optional PCI card is that it doesn't drive down the price.

    I'd expect Apple to use an implementation that works directly with OS X and GCD and possibly even uses system memory instead of external memory. It doesn't have to be 50-cores either. A 25-core Knights Landing chip would be plenty fast to boost software performance.


     


    While it doesn't fully align with the quote here, I still have to ask, what makes you feel thunderbolt is so great? I mean you've stated before that your opinion is if existing solutions are removed things will be forced forward. My opinion would be if this is really is a niche market, it's unlikely that many of these companies would allocate the necessary development costs to completely reworking something. I still see thunderbolt as half baked. Yes it will change in a couple years. We'll need new cables. The peripherals will require new chips. Looking at it today, it solves a very specific problem with portable form factors and carries a limited number of peripherals. The only reason I feel at least 2 are required to be useful at all is that non thunderbolt devices break the chain. If you're running one as mini displayport or using a configuration that saturates a connection, it's limiting if only one port is available. I noted that a graphic regarding ivy bridge has been circulating again.


     



     


     


    A problem that I see with the mac pro is that performance growth has been relatively flat around its base configuration with the cost of entry still remaining somewhat high. When it remains flat that long, it's harder for other advantages to prop it up. If Ivy doesn't make it into LGA2011 workstations prior to September/October of next year, which I think is likely, it seems unlikely that the next mac pro would use this. I think if they keep it going you're just going to see a late Sandy Bridge E rollout. I'm not  sure why something exciting is always expected. Sometimes delays are just delays.

Sign In or Register to comment.