The amount of RAM isn't a big price difference, but it does suggest which GPU is being used. There is a pattern that one can notice about AMD desktop and FirePro GPUs. GPUs don't support a wide range of memory configurations, in part because everything is done 'wide', a GPU is highly parallel compared to a CPU. But desktop GPU cards are sometimes available in 2 RAM configurations, i.e. 1GB or 2GB. With FirePro cards, they always come in one configuration, presumably the maximum the GPU supports. The GPU and memory are BGA (Ball Grid Array) soldered onto the card. So, no, I don't think memory will be upgradable.
So, I expect 2 W9000s will be an option. The idea that Apple will get some kind of huge break on the price or have a custom chip seems very unlikely to me, given the probable low volume, especially of the high-end $10k configuration with W9000s. I think the final retail price to the end user will be just as high as the PCIe cards (over $3k each).
The other part of 'up to' is the option of a lesser FirePro with its lower standard amount of memory. Now moving more to speculation than hard fact: In the new FirePro line, the W7000 is the minimum card that has 4 displayport connections, which I think is required for the 7 total display outputs. This card is $709 and has 4GB of GDDR5. You could do 1 W5000 + 1W7000, but that seems unlikely. You could do Xeon with built-in GPU and 2 W5000s, but this is even more unlikely as it would require different routing of one display channel.
The amount of RAM isn't a big price difference, but it does suggest which GPU is being used. There is a pattern that one can notice about AMD desktop and FirePro GPUs. GPUs don't support a wide range of memory configurations, in part because everything is done 'wide', a GPU is highly parallel compared to a CPU. But desktop GPU cards are sometimes available in 2 RAM configurations, i.e. 1GB or 2GB. With FirePro cards, they always come in one configuration, presumably the maximum the GPU supports. The GPU and memory are BGA (Ball Grid Array) soldered onto the card. So, no, I don't think memory will be upgradable.
So, I expect 2 W9000s will be an option. The idea that Apple will get some kind of huge break on the price or have a custom chip seems very unlikely to me, given the probable low volume, especially of the high-end $10k configuration with W9000s. I think the final retail price to the end user will be just as high as the PCIe cards (over $3k each).
The other part of 'up to' is the option of a lesser FirePro with its lower standard amount of memory. Now moving more to speculation than hard fact: In the new FirePro line, the W7000 is the minimum card that has 4 displayport connections, which I think is required for the 7 total display outputs. This card is $709 and has 4GB of GDDR5. You could do 1 W5000 + 1W7000, but that seems unlikely. You could do Xeon with built-in GPU and 2 W5000s, but this is even more unlikely as it would require different routing of one display channel.
It is clear that Apple intends these workstations to be dual GPUs, of the same chipset…
The apparent GPU choices then are all in the FirePro WX000 line…
(2) FirePro W5000 GPUs w/2GB GDDR5 RAM
(2) FirePro W7000 GPUs w/4GB GDDR5 RAM
(2) FirePro W8000 GPUs w/4GB GDDR5 ECC RAM
(2) FirePro W9000 GPUs w/6GB GDDR5 ECC RAM
BTO will allow the end-user to choose one of the above listed GPU sets…
Volume purchasing & exclusivity to the Mac Pro will allow Apple to leverage the pricing on these chips…
Between the threads on the new Mac Pro here, and including the one about the new production facility here in the States for said new Mac Pro; I think we can expect Apple to push at least a million Mac Pros a year. That is two million GPUs of various speeds. Again, large volume & 'claim to fame' of being in the new Mac Pros will leverage AMD into providing these chips at a lower than expected cost to Apple…
It is clear that Apple intends these workstations to be dual GPUs, of the same chipset…
The apparent GPU choices then are all in the FirePro WX000 line…
(2) FirePro W5000 GPUs w/2GB GDDR5 RAM
(2) FirePro W7000 GPUs w/4GB GDDR5 RAM
(2) FirePro W8000 GPUs w/4GB GDDR5 ECC RAM
(2) FirePro W9000 GPUs w/6GB GDDR5 ECC RAM
BTO will allow the end-user to choose one of the above listed GPU sets…
Volume purchasing & exclusivity to the Mac Pro will allow Apple to leverage the pricing on these chips…
Between the threads on the new Mac Pro here, and including the one about the new production facility here in the States for said new Mac Pro; I think we can expect Apple to push at least a million Mac Pros a year. That is two million GPUs of various speeds. Again, large volume & 'claim to fame' of being in the new Mac Pros will leverage AMD into providing these chips at a lower than expected cost to Apple…
I really hope that there are nVidia BTO options. And Radeon for that matter. Why would you want a FirePro W5000 ($450) over a 7970 ($400) when on OSX the drivers are the same quality? 2048 GCN processors on the 7970 vs 768 GCN processors on the 5000.
The OpenCL luxmark benchmarks aren't even close between the two. On windows there's a lot of app performance difference because of the drivers but on OSX that hasn't been as much of an issue because of Apple.
The OpenCL luxmark benchmarks aren't even close between the two. On windows there's a lot of app performance difference because of the drivers but on OSX that hasn't been as much of an issue because of Apple.
They could still differ in drivers. It could also be due to required ram. It could just be branding. It could be different firmware. The Windows cards are mostly the same at the hardware level. Workstation cards generally use ECC ram, but it's impossible to totally price these things out based on Windows counterparts when most configurations could just use 7970s with tweaked drivers.
BTO will allow the end-user to choose one of the above listed GPU sets…
Actually I suspect the only choices will be good, better, best as Apple has done in the past. The only thing that throws a wrench in the works is CPU offerings. I could see a base model with 2GB GPU cards with a six core Xeon.
Volume purchasing & exclusivity to the Mac Pro will allow Apple to leverage the pricing on these chips…
The question is by how much!
Between the threads on the new Mac Pro here, and including the one about the new production facility here in the States for said new Mac Pro; I think we can expect Apple to push at least a million Mac Pros a year.
The production facility in the USA won't help one bit, it has been shown repeatedly that people don't care.
Given that a Million a year is a massive bump. To that end Apple is going to need some very interesting configurations to reach a million a year. By interesting I'm looking at price and performance. I don't see a million a year happening unless they can introduce a viable model for under $2000.
That is two million GPUs of various speeds. Again, large volume & 'claim to fame' of being in the new Mac Pros will leverage AMD into providing these chips at a lower than expected cost to Apple…
Yes I do expect a lower cost to Apple. The question is by how much. The success of this new Mac Pro could be a huge win for AMD giving their hardware a very visible platform upon which to shine.
I doubt that you will see anything from NVidia. Such cards would require NVidia being willing to work with Apple on hardware design and to aggressively support OpenCL.
And Radeon for that matter. Why would you want a FirePro W5000 ($450) over a 7970 ($400) when on OSX the drivers are the same quality? 2048 GCN processors on the 7970 vs 768 GCN processors on the 5000.
This is where we need more info because the options for the new Mac Pro are not clear. As for retail pricing you can give up on anything related to retail cards being of value here. The cards are very non standard and Apple supplies the cooling solution. Combine that with high volume pricing and who knows what sort of costs Apple will be seeing.
The OpenCL luxmark benchmarks aren't even close between the two.
Well the difference in GCN cores ought to tell you why. However the Mac Pro comes with two cards.
As a side note a Mac Pro with one decent GPU card would come real close to what I imagine an XMac should be. Sell it for $1500 or less and you have a very acceptable machine that sits between the Mini and the Pro in the line up.
On windows there's a lot of app performance difference because of the drivers but on OSX that hasn't been as much of an issue because of Apple.
There are still a lot of unknowns here with respect to the new Mac Pros configurations. Until the unit ships we really don't know what pricing or configuration is.
I doubt that you will see anything from NVidia. Such cards would require NVidia being willing to work with Apple on hardware design and to aggressively support OpenCL.
This arguably already happens. Either NVidia or PNY went back and added OpenCL support for OSX on the Quadro 4000 where it didn't exist initially. Their cards also support OpenCL under Windows. Some of the workstation cards aren't always tuned for it. In fact I've seen a lot of solutions that strapped additional cooling to gaming cards, simply because some calculations are very fast on them. Typically with workstation gpus you have certain driver features enabled and tuned for high precision OpenGL. Tuning them for computation as well has been more of a recent thing. So far NVidia in general has been much more stable when it comes to computation. I would look at more than benchmarks too. Marvin used this example earlier. I do have some familiarity with that software. The gpu leveraged functions are for draft modes, as many of their target customers would be restricted by ram and number of material nodes if everything was run on the gpu.
www.spot3d.com/vray/help/rt100/render_gpu.htm
Quote:
Supported hardware and drivers
V-Ray RT for GPU has two back-ends (or engines). One is based on OpenCL (see the references section below for more info on OpenCL) and the other one - on the nVidia CUDA platform.
The OpenCL engine should be able to run on any OpenCL-compatible hardware. However, as of the time of this writing (April 28th, 2012), only the nVidia implementation of OpenCL is sufficiently advanced to run it properly. For best results, a Fermi- or Kepler-based card with at least 2 GB of video RAM is recommended. Older cards will work, but performance will be significantly worse. Due to the large amount of RAM needed to compile the OpenCL code, currently it only works in 64-bit builds of V-Ray RT. It may be possible to run the OpenCL engine on software CPU implementations of OpenCL from AMD and Intel, however this has not been thoroughly tested.
The CUDA engine is supported only in 64-bit builds of V-Ray RT for Fermi- and Kepler-based nVidia cards. It is recommended to use the CUDA engine on nVidia GPUs.
That links to further gpu leveraged libraries. This is one of those things where I think if Apple pushes OpenCL libraries, both vendors will contribute low level work. As I think I mentioned somewhere, 7970 drivers for the Mac Pro have been showing up in random developer builds of Mountain Lion. It may have happened once in Lion too. These could have been under development prior to Kepler. After all these cards have been available retail for almost a year at this point. It's fully possible that they will include rebranded 7970s as an option with the Firepro moniker used to designate some of the driver and firmware options. I'm not sure what they will show up as in Bootcamp, but I haven't been using it lately.
This arguably already happens. Either NVidia or PNY went back and added OpenCL support for OSX on the Quadro 4000 where it didn't exist initially. Their cards also support OpenCL under Windows. Some of the workstation cards aren't always tuned for it. In fact I've seen a lot of solutions that strapped additional cooling to gaming cards, simply because some calculations are very fast on them. Typically with workstation gpus you have certain driver features enabled and tuned for high precision OpenGL. Tuning them for computation as well has been more of a recent thing. So far NVidia in general has been much more stable when it comes to computation. I would look at more than benchmarks too. Marvin used this example earlier. I do have some familiarity with that software. The gpu leveraged functions are for draft modes, as many of their target customers would be restricted by ram and number of material nodes if everything was run on the gpu.
I still have to wonder if NVidia would even be remotely interested in doing custom cards for Apple.
NVidia also has a nice page on OpenCL.
That links to further gpu leveraged libraries. This is one of those things where I think if Apple pushes OpenCL libraries, both vendors will contribute low level work. As I think I mentioned somewhere, 7970 drivers for the Mac Pro have been showing up in random developer builds of Mountain Lion.
I've seen mention of those drivers.
It may have happened once in Lion too. These could have been under development prior to Kepler. After all these cards have been available retail for almost a year at this point. It's fully possible that they will include rebranded 7970s as an option with the Firepro moniker used to designate some of the driver and firmware options. I'm not sure what they will show up as in Bootcamp, but I haven't been using it lately.
This whole thing with so called Pro GPUs can be confusing as often the only thing pro about the cards is some firmware tweaks and different drivers. However admittedly pro cards do offer some advantage these days such as ECC RAM. Obviously both AMD and Apple will have to agree on how these cards are branded.
I'm still wondering about a custom spin to meld well with advanced TB displays of greater than 4K resolution. I could see them tweaking the output ports for this and giving the chip a custom part number. A workstation for the 4K market would ideally be able to drive larger than 4K displays. We have heard little about new displays from Apple even though they are obviously needed. Maybe it is wishful thinking but I see a big splash coming with a complete system from Apple. In other words the Mac Pro is released with a new display option and a disk array option. The timing seems right for this to happen.
I still have to wonder if NVidia would even be remotely interested in doing custom cards for Apple.
I've seen mention of those drivers.
PNY put out an after market Quadro 4000 for the Mac Pro. Early drivers were absolutely terrible and caused kernel panics. Supposedly the later ones were good. It outperformed other applications in some software, but it varied. OpenCL 1.1 support was added later in spite of this being an after market card. The hardware would have been identical to the Windows version with whatever driver development. A lot of people have mentioned generic NVidia cards having received some level of support as of Mountain Lion, even if they won't display boot screens. I haven't tried it. The EVGA 680 for Mac is basically the same hardware again. I don't see any reason why NVidia would be opposed to doing a custom order for Apple. If this is a pricing thing, they would have less reason to undercut their other offerings on Quadro branded cards, given how much they depend on margins there. NVidia does have Titan at $1000 with 6GB of ram. As I mentioned the framebuffer size is a big deal with these if they're leveraged for computation, as they can't rely on virtual memory. The problem either runs or it doesn't.
Quote:
This whole thing with so called Pro GPUs can be confusing as often the only thing pro about the cards is some firmware tweaks and different drivers. However admittedly pro cards do offer some advantage these days such as ECC RAM. Obviously both AMD and Apple will have to agree on how these cards are branded.
That is very true. Even on Windows sometimes gaming cards do work. You just have to be careful as some are terrible in certain applications. The 5870 had a ton of Windows bugs in a lot of Autodesk apps, and it was never officially supported there or even recommended on their compatibility list. It'll say certified, recommended, and mention any known caveats found in their testing. Some of the other gaming cards were fine. The 5870 did okay in OSX for its time, but I wouldn't want to buy one today, especially not at the same price. It's normal to expect a faster card with an update, as software tends to become more aggressive in its features when the people using it are running newer hardware. I definitely think it's silly to interpret everything based on current market specs of hardware that has been available for at least a year at its current pricing. When I looked up the top 12 core mac pro option, the cpu + gpu costs were still significantly below what it would cost using the parts everyone seems to be projecting as long as the stratospheric pricing on those gpus remains. If low volume is the problem that everyone suggests with the current machine, I don't think they're going to push it even higher unless they're sure they have a market for it somewhere. If anything it needs to be significantly more competitive at the sub $3k configuration if it's going to maintain any kind of volume. This is especially true with the possible need to invest in additional das solutions when moving to the new one.
NVidia owns a lot of the workstation market, and it's always rumored that it makes up a small portion of AMD's overall revenue. If you look at what they market as their low end Firepro cards, they're based on chips of the 5870 era. I couldn't find any breakdown in their quarterly report though. I do think it could be partly a branding thing here but still focused on computation and professional graphics. The choice of branding may just be to signify the emphasis even if in terms of hardware we're looking at a Radeon with a larger framebuffer.
By the way, I did read somewhere that memory speeds were a big factor in bringing about heterogeneous computing. I suspect they want something that can maintain lockstep with the gpu portion of the APU.
Quote:
I'm still wondering about a custom spin to meld well with advanced TB displays of greater than 4K resolution. I could see them tweaking the output ports for this and giving the chip a custom part number. A workstation for the 4K market would ideally be able to drive larger than 4K displays. We have heard little about new displays from Apple even though they are obviously needed. Maybe it is wishful thinking but I see a big splash coming with a complete system from Apple. In other words the Mac Pro is released with a new display option and a disk array option. The timing seems right for this to happen.
You would need larger than 4K if Apple wanted to maintain the current size of their displays and apply the same doubling at the 27" level. That would bring you to 5120 x 1880. As I've mentioned thunderbolt was really designed with portable devices in mind more than anything else. I don't particularly care for it for a number of reasons, mainly it doesn't easily shoehorn into everything, it's often more expensive at equivalent performance levels, and it's tied to intel rather than something like PCI which has its own group to set specific standards. I also hate serialized chains of equipment. The availability of 6 ports is nice as it at least alleviates the issue of things that don't play well with others on the chain or placement concerns. Displays have to go at the end of the chain. With the often proposed solution of placing storage in another room, it couldn't go on the same port as a display. Displaysport displays supposedly cannot go on the end of a thunderbolt chain either, although displayport 1.2 support would allow for daisy chaining displays as long as it supports the full standard. The wikipedia article on displayport is actually very well sourced. Displayport 1.2 was finalized December 2009. It increased bandwidth to 17.28 Gb/s. Raw bandwidth is actually higher, but there is some overhead. A lot of its functionality overlaps with thunderbolt in terms of chaining displays and implementing usb hubs.
I don't really care for closed standards in most cases. With NVidia as I mentioned the difference is that they do a lot of extra development that some of these developers are simply implementing in their own software. That is apparently how Adobe obtained their CUDA based raytracer. It's basically the same thing Apple does where OSX and certain software packages are tied to Macs. They add value to Macs and none to any machine that doesn't run Apple hardware. TB 2.0 introduces channel bonding just to get to 4K. I'm not entirely confident you'll see anything beyond that anytime soon, especially given the cost of custom development for what isn't one of the higher volume products (large displays). I was surprised just to see the amount of custom work on the imac. It was less surprising on portables given that their numbers.
The desktop display market in general has been pretty conservative for years. The manufacturers that made or at least designed extremely high quality panels, such as Hitachi, all pulled out of it. They may still be making specialized equipment. Speaking of Hitachi, they were the ones that pioneered IPS in the late 90s. In spite of its flaws it seems like it outships Samsung's PVA. PVA was nice in some ways. It just had gamma shift issues at some viewing angles. I will say that some of the non - LG branded panels had far fewer flaws, but basically everything uses LG now. That is why I kind of facedesk whenever someone says Apple should go elsewhere. Unless the elusive Sharp displays start to show up in large quantities, what are they going to use? Even 2003-2009 when others were available, Apple wasn't one of the adopters, probably due to cost and possibly required quantity.
Though 4k panels for an iMac and a Pro monitor...would get the price down..?
Desktop displays are often produced at 16:9 to allow for some reuse of parts initially designed for televisions. As of today those are mostly 1080 components. You would have a much greater component overlap if both went to 4K compared to today where a high quality 27" display is 2560x1440. Desktop displays in general tend to be a very conservative point in terms of panel technology relative to notebooks, phones, tablets, etc. If you start to see enough 4K televisions, they will definitely impact the price of 4K displays. I think overall sizes are good where they are. I like 24", but I can do 27". No matter how many pixels you cram on, a notebook display loses too much space to UI elements for me personally. It depends on the user, but 2-3 display setups are used by people outside of creative professions as a means of organizing multiple windows that are each designed to fill most of the screen. Someone brought up the topic of curved displays in another thread. I wouldn't care for the visual distortion, although I do angle secondary displays.
Comments
The amount of RAM isn't a big price difference, but it does suggest which GPU is being used. There is a pattern that one can notice about AMD desktop and FirePro GPUs. GPUs don't support a wide range of memory configurations, in part because everything is done 'wide', a GPU is highly parallel compared to a CPU. But desktop GPU cards are sometimes available in 2 RAM configurations, i.e. 1GB or 2GB. With FirePro cards, they always come in one configuration, presumably the maximum the GPU supports. The GPU and memory are BGA (Ball Grid Array) soldered onto the card. So, no, I don't think memory will be upgradable.
So, I expect 2 W9000s will be an option. The idea that Apple will get some kind of huge break on the price or have a custom chip seems very unlikely to me, given the probable low volume, especially of the high-end $10k configuration with W9000s. I think the final retail price to the end user will be just as high as the PCIe cards (over $3k each).
The other part of 'up to' is the option of a lesser FirePro with its lower standard amount of memory. Now moving more to speculation than hard fact: In the new FirePro line, the W7000 is the minimum card that has 4 displayport connections, which I think is required for the 7 total display outputs. This card is $709 and has 4GB of GDDR5. You could do 1 W5000 + 1W7000, but that seems unlikely. You could do Xeon with built-in GPU and 2 W5000s, but this is even more unlikely as it would require different routing of one display channel.
Quote:
Originally Posted by davida
The amount of RAM isn't a big price difference, but it does suggest which GPU is being used. There is a pattern that one can notice about AMD desktop and FirePro GPUs. GPUs don't support a wide range of memory configurations, in part because everything is done 'wide', a GPU is highly parallel compared to a CPU. But desktop GPU cards are sometimes available in 2 RAM configurations, i.e. 1GB or 2GB. With FirePro cards, they always come in one configuration, presumably the maximum the GPU supports. The GPU and memory are BGA (Ball Grid Array) soldered onto the card. So, no, I don't think memory will be upgradable.
So, I expect 2 W9000s will be an option. The idea that Apple will get some kind of huge break on the price or have a custom chip seems very unlikely to me, given the probable low volume, especially of the high-end $10k configuration with W9000s. I think the final retail price to the end user will be just as high as the PCIe cards (over $3k each).
The other part of 'up to' is the option of a lesser FirePro with its lower standard amount of memory. Now moving more to speculation than hard fact: In the new FirePro line, the W7000 is the minimum card that has 4 displayport connections, which I think is required for the 7 total display outputs. This card is $709 and has 4GB of GDDR5. You could do 1 W5000 + 1W7000, but that seems unlikely. You could do Xeon with built-in GPU and 2 W5000s, but this is even more unlikely as it would require different routing of one display channel.
It is clear that Apple intends these workstations to be dual GPUs, of the same chipset…
The apparent GPU choices then are all in the FirePro WX000 line…
(2) FirePro W5000 GPUs w/2GB GDDR5 RAM
(2) FirePro W7000 GPUs w/4GB GDDR5 RAM
(2) FirePro W8000 GPUs w/4GB GDDR5 ECC RAM
(2) FirePro W9000 GPUs w/6GB GDDR5 ECC RAM
BTO will allow the end-user to choose one of the above listed GPU sets…
Volume purchasing & exclusivity to the Mac Pro will allow Apple to leverage the pricing on these chips…
Between the threads on the new Mac Pro here, and including the one about the new production facility here in the States for said new Mac Pro; I think we can expect Apple to push at least a million Mac Pros a year. That is two million GPUs of various speeds. Again, large volume & 'claim to fame' of being in the new Mac Pros will leverage AMD into providing these chips at a lower than expected cost to Apple…
Quote:
Originally Posted by MacRonin
It is clear that Apple intends these workstations to be dual GPUs, of the same chipset…
The apparent GPU choices then are all in the FirePro WX000 line…
(2) FirePro W5000 GPUs w/2GB GDDR5 RAM
(2) FirePro W7000 GPUs w/4GB GDDR5 RAM
(2) FirePro W8000 GPUs w/4GB GDDR5 ECC RAM
(2) FirePro W9000 GPUs w/6GB GDDR5 ECC RAM
BTO will allow the end-user to choose one of the above listed GPU sets…
Volume purchasing & exclusivity to the Mac Pro will allow Apple to leverage the pricing on these chips…
Between the threads on the new Mac Pro here, and including the one about the new production facility here in the States for said new Mac Pro; I think we can expect Apple to push at least a million Mac Pros a year. That is two million GPUs of various speeds. Again, large volume & 'claim to fame' of being in the new Mac Pros will leverage AMD into providing these chips at a lower than expected cost to Apple…
I really hope that there are nVidia BTO options. And Radeon for that matter. Why would you want a FirePro W5000 ($450) over a 7970 ($400) when on OSX the drivers are the same quality? 2048 GCN processors on the 7970 vs 768 GCN processors on the 5000.
The OpenCL luxmark benchmarks aren't even close between the two. On windows there's a lot of app performance difference because of the drivers but on OSX that hasn't been as much of an issue because of Apple.
Quote:
Originally Posted by nht
The OpenCL luxmark benchmarks aren't even close between the two. On windows there's a lot of app performance difference because of the drivers but on OSX that hasn't been as much of an issue because of Apple.
They could still differ in drivers. It could also be due to required ram. It could just be branding. It could be different firmware. The Windows cards are mostly the same at the hardware level. Workstation cards generally use ECC ram, but it's impossible to totally price these things out based on Windows counterparts when most configurations could just use 7970s with tweaked drivers.
Given that a Million a year is a massive bump. To that end Apple is going to need some very interesting configurations to reach a million a year. By interesting I'm looking at price and performance. I don't see a million a year happening unless they can introduce a viable model for under $2000. Yes I do expect a lower cost to Apple. The question is by how much. The success of this new Mac Pro could be a huge win for AMD giving their hardware a very visible platform upon which to shine.
As a side note a Mac Pro with one decent GPU card would come real close to what I imagine an XMac should be. Sell it for $1500 or less and you have a very acceptable machine that sits between the Mini and the Pro in the line up. There are still a lot of unknowns here with respect to the new Mac Pros configurations. Until the unit ships we really don't know what pricing or configuration is.
Quote:
Originally Posted by wizard69
I doubt that you will see anything from NVidia. Such cards would require NVidia being willing to work with Apple on hardware design and to aggressively support OpenCL.
This arguably already happens. Either NVidia or PNY went back and added OpenCL support for OSX on the Quadro 4000 where it didn't exist initially. Their cards also support OpenCL under Windows. Some of the workstation cards aren't always tuned for it. In fact I've seen a lot of solutions that strapped additional cooling to gaming cards, simply because some calculations are very fast on them. Typically with workstation gpus you have certain driver features enabled and tuned for high precision OpenGL. Tuning them for computation as well has been more of a recent thing. So far NVidia in general has been much more stable when it comes to computation. I would look at more than benchmarks too. Marvin used this example earlier. I do have some familiarity with that software. The gpu leveraged functions are for draft modes, as many of their target customers would be restricted by ram and number of material nodes if everything was run on the gpu.
www.spot3d.com/vray/help/rt100/render_gpu.htm
Quote:
Supported hardware and drivers
NVidia also has a nice page on OpenCL.
https://developer.nvidia.com/opencl
That links to further gpu leveraged libraries. This is one of those things where I think if Apple pushes OpenCL libraries, both vendors will contribute low level work. As I think I mentioned somewhere, 7970 drivers for the Mac Pro have been showing up in random developer builds of Mountain Lion. It may have happened once in Lion too. These could have been under development prior to Kepler. After all these cards have been available retail for almost a year at this point. It's fully possible that they will include rebranded 7970s as an option with the Firepro moniker used to designate some of the driver and firmware options. I'm not sure what they will show up as in Bootcamp, but I haven't been using it lately.
I'm still wondering about a custom spin to meld well with advanced TB displays of greater than 4K resolution. I could see them tweaking the output ports for this and giving the chip a custom part number. A workstation for the 4K market would ideally be able to drive larger than 4K displays. We have heard little about new displays from Apple even though they are obviously needed. Maybe it is wishful thinking but I see a big splash coming with a complete system from Apple. In other words the Mac Pro is released with a new display option and a disk array option. The timing seems right for this to happen.
Quote:
Originally Posted by wizard69
I still have to wonder if NVidia would even be remotely interested in doing custom cards for Apple.
I've seen mention of those drivers.
PNY put out an after market Quadro 4000 for the Mac Pro. Early drivers were absolutely terrible and caused kernel panics. Supposedly the later ones were good. It outperformed other applications in some software, but it varied. OpenCL 1.1 support was added later in spite of this being an after market card. The hardware would have been identical to the Windows version with whatever driver development. A lot of people have mentioned generic NVidia cards having received some level of support as of Mountain Lion, even if they won't display boot screens. I haven't tried it. The EVGA 680 for Mac is basically the same hardware again. I don't see any reason why NVidia would be opposed to doing a custom order for Apple. If this is a pricing thing, they would have less reason to undercut their other offerings on Quadro branded cards, given how much they depend on margins there. NVidia does have Titan at $1000 with 6GB of ram. As I mentioned the framebuffer size is a big deal with these if they're leveraged for computation, as they can't rely on virtual memory. The problem either runs or it doesn't.
Quote:
This whole thing with so called Pro GPUs can be confusing as often the only thing pro about the cards is some firmware tweaks and different drivers. However admittedly pro cards do offer some advantage these days such as ECC RAM. Obviously both AMD and Apple will have to agree on how these cards are branded.
That is very true. Even on Windows sometimes gaming cards do work. You just have to be careful as some are terrible in certain applications. The 5870 had a ton of Windows bugs in a lot of Autodesk apps, and it was never officially supported there or even recommended on their compatibility list. It'll say certified, recommended, and mention any known caveats found in their testing. Some of the other gaming cards were fine. The 5870 did okay in OSX for its time, but I wouldn't want to buy one today, especially not at the same price. It's normal to expect a faster card with an update, as software tends to become more aggressive in its features when the people using it are running newer hardware. I definitely think it's silly to interpret everything based on current market specs of hardware that has been available for at least a year at its current pricing. When I looked up the top 12 core mac pro option, the cpu + gpu costs were still significantly below what it would cost using the parts everyone seems to be projecting as long as the stratospheric pricing on those gpus remains. If low volume is the problem that everyone suggests with the current machine, I don't think they're going to push it even higher unless they're sure they have a market for it somewhere. If anything it needs to be significantly more competitive at the sub $3k configuration if it's going to maintain any kind of volume. This is especially true with the possible need to invest in additional das solutions when moving to the new one.
NVidia owns a lot of the workstation market, and it's always rumored that it makes up a small portion of AMD's overall revenue. If you look at what they market as their low end Firepro cards, they're based on chips of the 5870 era. I couldn't find any breakdown in their quarterly report though. I do think it could be partly a branding thing here but still focused on computation and professional graphics. The choice of branding may just be to signify the emphasis even if in terms of hardware we're looking at a Radeon with a larger framebuffer.
By the way, I did read somewhere that memory speeds were a big factor in bringing about heterogeneous computing. I suspect they want something that can maintain lockstep with the gpu portion of the APU.
Quote:
I'm still wondering about a custom spin to meld well with advanced TB displays of greater than 4K resolution. I could see them tweaking the output ports for this and giving the chip a custom part number. A workstation for the 4K market would ideally be able to drive larger than 4K displays. We have heard little about new displays from Apple even though they are obviously needed. Maybe it is wishful thinking but I see a big splash coming with a complete system from Apple. In other words the Mac Pro is released with a new display option and a disk array option. The timing seems right for this to happen.
You would need larger than 4K if Apple wanted to maintain the current size of their displays and apply the same doubling at the 27" level. That would bring you to 5120 x 1880. As I've mentioned thunderbolt was really designed with portable devices in mind more than anything else. I don't particularly care for it for a number of reasons, mainly it doesn't easily shoehorn into everything, it's often more expensive at equivalent performance levels, and it's tied to intel rather than something like PCI which has its own group to set specific standards. I also hate serialized chains of equipment. The availability of 6 ports is nice as it at least alleviates the issue of things that don't play well with others on the chain or placement concerns. Displays have to go at the end of the chain. With the often proposed solution of placing storage in another room, it couldn't go on the same port as a display. Displaysport displays supposedly cannot go on the end of a thunderbolt chain either, although displayport 1.2 support would allow for daisy chaining displays as long as it supports the full standard. The wikipedia article on displayport is actually very well sourced. Displayport 1.2 was finalized December 2009. It increased bandwidth to 17.28 Gb/s. Raw bandwidth is actually higher, but there is some overhead. A lot of its functionality overlaps with thunderbolt in terms of chaining displays and implementing usb hubs.
I don't really care for closed standards in most cases. With NVidia as I mentioned the difference is that they do a lot of extra development that some of these developers are simply implementing in their own software. That is apparently how Adobe obtained their CUDA based raytracer. It's basically the same thing Apple does where OSX and certain software packages are tied to Macs. They add value to Macs and none to any machine that doesn't run Apple hardware. TB 2.0 introduces channel bonding just to get to 4K. I'm not entirely confident you'll see anything beyond that anytime soon, especially given the cost of custom development for what isn't one of the higher volume products (large displays). I was surprised just to see the amount of custom work on the imac. It was less surprising on portables given that their numbers.
The desktop display market in general has been pretty conservative for years. The manufacturers that made or at least designed extremely high quality panels, such as Hitachi, all pulled out of it. They may still be making specialized equipment. Speaking of Hitachi, they were the ones that pioneered IPS in the late 90s. In spite of its flaws it seems like it outships Samsung's PVA. PVA was nice in some ways. It just had gamma shift issues at some viewing angles. I will say that some of the non - LG branded panels had far fewer flaws, but basically everything uses LG now. That is why I kind of facedesk whenever someone says Apple should go elsewhere. Unless the elusive Sharp displays start to show up in large quantities, what are they going to use? Even 2003-2009 when others were available, Apple wasn't one of the adopters, probably due to cost and possibly required quantity.
Quote:
The desktop display market in general has been pretty conservative for years.
No kidding...
Lemon Bon Bon.
The 4k 'revolution' or ultra HD may FINALLY get things moving again re: monitor tech'.
I'm wondering if you even need a monitor if you can have a 4k tv? Would that be able to serve as a computer monitor?
4k. I'd love for the next iMac to be that. I'm not sure we'll see it this year?
Though 4k panels for an iMac and a Pro monitor...would get the price down..?
Maybe 2014? For pricing to drop?
Lemon Bon Bon.
Quote:
Originally Posted by Lemon Bon Bon.
Though 4k panels for an iMac and a Pro monitor...would get the price down..?
Desktop displays are often produced at 16:9 to allow for some reuse of parts initially designed for televisions. As of today those are mostly 1080 components. You would have a much greater component overlap if both went to 4K compared to today where a high quality 27" display is 2560x1440. Desktop displays in general tend to be a very conservative point in terms of panel technology relative to notebooks, phones, tablets, etc. If you start to see enough 4K televisions, they will definitely impact the price of 4K displays. I think overall sizes are good where they are. I like 24", but I can do 27". No matter how many pixels you cram on, a notebook display loses too much space to UI elements for me personally. It depends on the user, but 2-3 display setups are used by people outside of creative professions as a means of organizing multiple windows that are each designed to fill most of the screen. Someone brought up the topic of curved displays in another thread. I wouldn't care for the visual distortion, although I do angle secondary displays.
Mac Pro ~~
4k 'revolution' or ultra HD may FINALLY get things moving again
I need to think about the mac pro~~~