I believe if Apple is looking at AMD they're looking at the lowend range (Macbook, Mac mini) and the high end (Mac Pro)
Intel is probably going to own the iMac configs. Who knows though but my guess is that AMD is discussing 2011 products because it appears that this is their "make it or break it" year with Bulldozer.
I definitely see AMD CPUs being a viable option in the low-end due to the stronger IGPs, but I don't see the justification for AMD CPUs going into the Mac Pro. One of the major new features of Bulldozer and Sandy Bridge are AVX instructions and 256-bit execution units. Whereas each core in Sandy Bridge has 4 256-bit AVS execution units on 3 ports (MUL, ADD, SHUFFLE, and BOOL) each core in Bulldozer only has 1 admittedly more powerful 128-bit FMAC, but to execute
256-bit AVX instructions it seems both cores in a module need to combine into a into a single 256-bit FMAC unit. There are no doubt more intricacies involved, but it definitely seems Sandy Bridge will have better 256-bit AVX instruction throughput.
What's more, Sandy Bridge seems heavily skewed toward high-end CPUs, particularly Xeons that Apple uses in the Mac Pro. Whereas notebook and mainstream desktop Sandy Bridge seem to be targeting 1.5MB to 2MB of L3 cache per core, Xeon Sandy Bridge seem to be targetting up to 2.5MB of L3 cache per core. Xeon Sandy Bridge also look to have quad channel memory controllers and PCIe Gen 3. Bulldozer's 16 core version seems impressive, but presumably it's actually 8 modules so I wouldn't count an 8-core Sandy Bridge out.
Quote:
Originally Posted by bitemymac
AMD Thuban (Phenom II X6) is scheduled for late April launch(in about 10 days). Some of these chips are already in hands of many enthusiast in Asia. The performance expectations and early leaked benchmarks of Thuban is clock per clock comparable or even better than i7 9xx platform, exception of i7 980X with HT on. This can be attained by almost 1/2 to 1/3 of the price point of intel chips, not to mention intel mobo costing 2x as well.
AMD Thuban incorporates turbo mode which allows increasing the multiplier by 2 to 2.5 (400 to 500 MHz) boost when only 3 cores or less are running just like intel's offering but it does with all three cores at same speed boost. This is the current line up of AMD's offer. Why would you want to use intel i5 or i7 for much higher price for less performance? Not to mention these chips come with 95W/125W flavor.
It looks like AMD is back on track, like the old days of AMD64 X2 vs. Intel P4 days. You may also want to look up Bulldozer for the future line up.
The issue is that Thuban's closest competitors are not the Bloomfield Core i7 which Intel has let stagnate, but Lynnfield Core i7 which have a much more reasonable 95W TDP and much higher Turbo Boost modes than Bloomfield. Intel is also preparing to launch a faster 3.06GHz Core i7 880 Lynnfield and will probably do a price cascade if Thuban is that threatening. If Thuban causes Intel to wake up and speed up the replacement of Bloomfield Core i7 with cut-down Gulftowns I'm all for it.
Not bad. If Thuban is close to I7 then things would indeed be looking very good because AMD can make up performance delta in some areas via superior GPU.
From those benchmarks it looks like Thuban is still slightly clock for clock slower than Nehalem despite a 1.5x core count advantage. But regardless, it does show AMD really closed the performance lead even if it's not done in the most economically feasible way given the larger die size and cost of production for Thuban. It does give great expectations for Bulldozer. I think the top performing X6 1090T has a 140W TDP though so it's not appropriate even for a Mac Pro.
We still have several AMD servers - all are great machines and all of them still run 24/7. Starting with Athlon MP through several Opteron lines ending with the newest dual socket 4 core Opterons, all running Linux. I can imagine Special edition 24 core Opteron Mac PRO - one ugly MF.
The other wild-card may be that Sandy Bridge sees a return of the CPU microarchitectural stewardship to the Haifa team in Isreal who was responsible for the Pentium M and it's descendants the Core and Core 2 product lines and for regaining the performance and power lead for Intel. Nehalem and Westmere was designed by the Hillsboro team who was responsible for Netburst and Pentium 4 and we can see their influence on getting Nehalem to regain the server performance lead for Intel with lots of memory bandwidth, large slow L3 cache, and Hyperthreading. I'm expecting the Haifa team to implement major in tweaking the microarchitecture to maximize performance/watt using things like decreasing cache latency, hopefully substantially in the case of the L3 cache, increasing interconnect bandwidth, expansion of micro and macro-op fusion, expansion of the loop stream detector, increasing the number of instructions kept in flight and associated buffers to maximize Hyperthreading in anticipation of Bulldozer's SMT, optimizing clock gating, power plane distribution, and the PCU for higher Turbo Boost modes, and hopefully other new tricks.
I'm wondering about Sandy Bridge improvements almost as much as I am with Bulldozer. Someone on the XtremeSystems forums has been claiming for a while that the Sandy Bridge core for the 6/8 core variants will be tweaked compared to the Sandy Bridge core for the 2/4 core variants, presumably to counter Bulldozer.
Quote:
Originally Posted by ltcommander.data
Most of the rumours seem to indicate quad core mobile Sandy Bridge to maintain the 45W TDP due to the integration of the IGP, but you would think some 35W TDP quad cores would be possible considering the CPU moves from 45nm to 32nm over Clarksfield and the IGP also moves from 45nm to 32nm over Arrandale combined with the Sandy Bridge microarchitecture being tuned for power efficiency should open up some thermal room.
I was a little surprised that quad-core didn't move down either in market segment (so it seems?) or in TDP. The IGP has something to do with it so if it was turned off (when a discrete GPU is on), the 45 W quad-core would probably act like ~35 W.
Quote:
Originally Posted by ltcommander.data
I definitely see AMD CPUs being a viable option in the low-end due to the stronger IGPs, but I don't see the justification for AMD CPUs going into the Mac Pro.
I can see AMD in the Mac Pro as long as the dual-die variants are used.
Quote:
Originally Posted by ltcommander.data
From those benchmarks it looks like Thuban is still slightly clock for clock slower than Nehalem despite a 1.5x core count advantage. But regardless, it does show AMD really closed the performance lead even if it's not done in the most economically feasible way given the larger die size and cost of production for Thuban. It does give great expectations for Bulldozer. I think the top performing X6 1090T has a 140W TDP though so it's not appropriate even for a Mac Pro.
That (and roadmaps from Intel and AMD) tells me that AMD is going for more cores for more performance while Intel is going for higher per-core performance (clocks, microarchitectural).
We still have several AMD servers - all are great machines and all of them still run 24/7. Starting with Athlon MP through several Opteron lines ending with the newest dual socket 4 core Opterons, all running Linux. I can imagine Special edition 24 core Opteron Mac PRO - one ugly MF.
So ugly it'll look identical to the Xeons as Apple will swap out the bus architecture for Hypertransport [the same transport they helped pioneer] and the CPUs in the exact same location.
From those benchmarks it looks like Thuban is still slightly clock for clock slower than Nehalem despite a 1.5x core count advantage. But regardless, it does show AMD really closed the performance lead even if it's not done in the most economically feasible way given the larger die size and cost of production for Thuban. It does give great expectations for Bulldozer. I think the top performing X6 1090T has a 140W TDP though so it's not appropriate even for a Mac Pro.
AMD absolutely has a competitive product for workstations and servers, but what I don't see is any competitive product for notebooks (and AIOs, likely Apple's second biggest Mac category).
YoY stats consistently show notebook purchases growing. Does AMD having anything that compete with Intel on the mobile front? From what I've read on AnandTech for a long time (and I'm sure you have too since I've seen you in the forums) AMD is still woefully behind in this area.
I was a little surprised that quad-core didn't move down either in market segment (so it seems…) or in TDP. The IGP has something to do with it so if it was turned off (when a discrete GPU is on), the 45 W quad-core would probably act like ~35 W.
Well there is be hope since there are reports that quad core desktop Sandy Bridge are targeting a TDP from 65-95W TDP, so on desktop TDP is dropping despite adding an IGP. Hopefully the same will be true for mobile quad core Sandy Bridge to have some lower 35W TDP models.
Quote:
Originally Posted by gabberattack
We still have several AMD servers - all are great machines and all of them still run 24/7. Starting with Athlon MP through several Opteron lines ending with the newest dual socket 4 core Opterons, all running Linux. I can imagine Special edition 24 core Opteron Mac PRO - one ugly MF.
Quote:
Originally Posted by iMacmatician
I can see AMD in the Mac Pro as long as the dual-die variants are used.
That (and roadmaps from Intel and AMD) tells me that AMD is going for more cores for more performance while Intel is going for higher per-core performance (clocks, microarchitectural).
Quote:
Originally Posted by solipsism
AMD absolutely has a competitive product for workstations and servers, but what I don't see is any competitive product for notebooks (and AIOs, likely Apple's second biggest Mac category).
The problem with extreme focus on core count is that even professional software is not well optimized for more than 8 cores, much less the 2x12 cores that are possible with Magny-Cours.
Magny-Cours is really targetted at servers, whereas it fails to beat 6-core Gulftown in workstation usage as in the Mac Pro. The Mac Pro only uses 95W TDP CPUs when configured dual processor, which limits Magny-Cours to the 2.2GHz 80W ACP/115W TDP Opteron 6174 while a slightly cheaper Gulftown alternative is a 2.66GHz 95W TDP Xeon X5650. The Xeon X5650 is overall faster, sometimes much faster. What's worse is that the Opteron 6174 is the fastest 12 core Magny-Cours whereas the X5650 is one of the slower 6-core Gulftown with a 2.93GHz X5670 available at the same 95W TDP. Apple will see no benefit in adopting AMD CPUs so soon with Magny-Cours.
AMD's core count focus will no doubt be better realized in Bulldozer which should adopt Turbo Core like the upcoming desktop Thuban to increase clock speed in low threaded applications, which are going to be most of them for a while. The interesting thing will be seeing how well multithreading solution in Grand Central scales to extreme core counts, although we should already be able to get an idea if there are any well implemented Grand Central applications that someone can test on DP 16 thread Nehalem Mac Pros and presumably future DP 24 thread Gulftown Mac Pros.
Quote:
Originally Posted by solipsism
YoY stats consistently show notebook purchases growing. Does AMD having anything that compete with Intel on the mobile front? From what I've read on AnandTech for a long time (and I'm sure you have too since I've seen you in the forums) AMD is still woefully behind in this area.
I believe most of AMD's performance in the mobile arena is due to pricing rather than performance. I believe AMD's current Caspian mobile CPUs are based on K10 which would make them competitive with the original Meroms and maybe approaching Penryn, but certainly not approaching Nehalem or Westmere.
AMD absolutely has a competitive product for workstations and servers, but what I don't see is any competitive product for notebooks (and AIOs, likely Apple's second biggest Mac category).
YoY stats consistently show notebook purchases growing. Does AMD having anything that compete with Intel on the mobile front? From what I've read on AnandTech for a long time (and I'm sure you have too since I've seen you in the forums) AMD is still woefully behind in this area.
Guess it depends on how well the Llano APU performs in the Sabine platform. The darkhorse here is OpenCL. AMD is on top of things with this regard. AMD can make up some ground but we need some realworld benefits to OpenCL before we can grok how the GPU can boost general computing software.
Guess it depends on how well the Llano APU performs in the Sabine platform. The darkhorse here is OpenCL. AMD is on top of things with this regard. AMD can make up some ground but we need some realworld benefits to OpenCL before we can grok how the GPU can boost general computing software.
I'm actually curious how good ATI's OpenCL performance is currently compared to nVidia? I believe at least initially, ATI's HD4000 series was woefully behind nVidia's G80 performance, there were even reports that the HD4870 was slower in OpenCL than the 9400M. It's clear that the HD4000 series wasn't fully designed for OpenCL in mind and is penalized by incomplete support for local shared memory in hardware having to drop back to VRAM, but has performance improved by better drivers? Similarly, the HD5000 series has these limitations solved so is performance significantly improved to be competitive to the nVidia GT200 and Fermi?
Early benchmarks show the HD5870 is still at a distinct disadvantage against the GTX285. I was annoyed that Apple chose to include the GT330M in the new MacBook Pros compared to the HD5650 which would be faster in games and cooler, but the GT330M would presumably be a superior OpenCL choice. As such, it concerns me whether even going with AMD CPUs and Llano will yield an OpenCL advantage over the 320M considering Llano is still based on the same HD5000 architecture. If AMD IGPs don't give an OpenCL advantage over the current nVidia 320M IGP, then maybe Apple should consider just dropping the IGP/chipset licensing moral stance, and redesign the 13" MacBook Pro so it can accommodate a low-end discrete GPU in addition to Intel CPUs which would yield better performance than any IGP, AMD or nVidia, anyways.
Arrandale's IGP does support GPGPU operation with Intel reported to be working on a DirectX Compute Shader driver for later this year. If Apple and Intel could write an OpenCL driver for Arrandale's IGP, it'd be very interesting to see how well it does. At 12 shaders I'm not expecting a miracle, but I wonder if they'll having compute efficiency more similar to what nVidia can do with a small number of stream processors compared to ATI's many SP VLIW strategy which seems to scale better for graphics than compute.
The Mac Pro only uses 95W TDP CPUs when configured dual processor, which limits Magny-Cours to the 2.2GHz 80W ACP/115W TDP Opteron 6174 while a slightly cheaper Gulftown alternative is a 2.66GHz 95W TDP Xeon X5650. The Xeon X5650 is overall faster, sometimes much faster. What's worse is that the Opteron 6174 is the fastest 12 core Magny-Cours whereas the X5650 is one of the slower 6-core Gulftown with a 2.93GHz X5670 available at the same 95W TDP. Apple will see no benefit in adopting AMD CPUs so soon with Magny-Cours.
I see a lot price comparisons to AMD and Intel with AMD always being much lower, but rarely do I see it mentioned how much this cost savings is after 2 or 3 years of power usage for AMD over Intel on servers.
Quote:
Originally Posted by hmurchison
Guess it depends on how well the Llano APU performs in the Sabine platform. The darkhorse here is OpenCL. AMD is on top of things with this regard. AMD can make up some ground but we need some realworld benefits to OpenCL before we can grok how the GPU can boost general computing software.
Nvidia might not be on par with AMD/ATI with OpenCL but it looks like Nvidia has plenty of time to get it figured out.
It's funny that Khronos Group ratified OpenCL with in a record 5 months, yet 15 months later it seems dead in the water yet with so much potential. Hopefully that will change this year with a hopeful revision of many apps to Cocoa/64-bit, including iTunes X around September.
From those benchmarks it looks like Thuban is still slightly clock for clock slower than Nehalem despite a 1.5x core count advantage. But regardless, it does show AMD really closed the performance lead even if it's not done in the most economically feasible way given the larger die size and cost of production for Thuban. It does give great expectations for Bulldozer. I think the top performing X6 1090T has a 140W TDP though so it's not appropriate even for a Mac Pro.
The real benchmark actually scales much better due to more aggressive L2/L3 cache timing vs. prior phenom II X4 and does perform better than the estimated bench scores on the prior link.
Here's a link of one of the enthusiasts playing around the with Thuban 1055T @ 2.8GHz with turbo boost to 3.3GHz on three cores(current base x6 model). He has done some overclocking to 3.2GHz to match i7 980 on cinebench 10 & 11.5 which does beat out i7 at the same clock speed. There are few more posters with their own experience on different threads with other benchmark comparison to i7 scores.
BS. No matter what, AMD is considered to be a second tier supplier.
Only in your limited view of the world! Besides why do you think there are so many super computers built with AMD chips?
[quote]
Sure, if you want a $400 piece of junk, get an AMD machine.[/QUOTE
]You can buy $400 junk computers with Intel processors in them so what is your point?
In any event there is a common theme in this thread that AMD offers up only low end solutions. That isn't the case at all. They have weak spots right now but so does Intel. The important thing with AMD is that in many ways they have been given a second chance at the low end with Intels Arrandale which as we all know has just about the worst shipping GPU going.
In any event the idea that Apple can't implement a viable AMD system is silly. AMD has a number of products that Apple could successfully implement. Especially in the Mini where Apple needs a product that Intel no longer offers.
As an aside I have to wonder if anybody here has even a remote clue as from where the current architecture came from? It wasn't Intel boys.
As more and more of the CPU's heavy lifting is shared with the GPU, a lessor capable CPU wouldn't necessarily be felt if the lessor capable CPU was equipped with a better GPU then a competing more capable CPU equipped with a lessor capable CPU. So in theory overall performance could be the same or better with two main CPU's that aren't equal in processing power depending on the quipped CPUs.
Further, Apple has been designing processors for a long time through it POWER PC partnership with IBM and Motorola. It's expertise is seen in it's new iPads that are quite snappy for relatively low powered devices. If Apple were to use AMD, it would make sure performance wasn't an issue especially if Apple were giving AMD advice.
I might not trust AMD, but I would trust Apple.
Quote:
Originally Posted by ghostface147
I understand that Apple wants to get great deals on chips, but no way they should be satisfied with possibly getting 80% performance at 60% cost. They need to get 100% performance at all times when possible. If AMD can deliver, more power to them. I personally wouldn't buy an AMD-based Mac. I've had minor stability issues with them on the Windows side.
As an aside I have to wonder if anybody here has even a remote clue as from where the current architecture came from? It wasn't Intel boys.
This isn't another all roads lead to DEC Alpha question is it?
Intel's Haifa team, who's working on Sandy Bridge, does have a history with many of the features we're seeing adopted. In the late '90s, they were working on the Timna which among other things included an on die IGP and memory controller. Timna was cancelled, but the experience in SoC design and micro-architecture efficiencies would eventually become the Pentium M and form the basis of the current CPU microarchitecture. Fittingly, the Haifa team will finally get to produce what I believe is the first x86 CPU with native on die IGP in Sandy Bridge. The Intel Pineview Atom does have a on die IGP, but appears to be structured as a traditional northbridge connected via FSB just on the same die rather than something completely integrated.
Quote:
Originally Posted by TBell
I might not trust AMD, but I would trust Apple.
Well, after making fun of PCs for using Intel IGPs, Apple switched from the dedicated Radeon 9200 in the PowerPC Mac Mini to the Intel GMA 950. The GMA 950 was decidedly slower than the Radeon 9200 in anything graphically intensive, but did have the advantage of having more flexible/programmable shaders even if they were slower, but sufficient to support light Core Image acceleration which was what Apple promoted. Trusting Apple doesn't necessarily mean you'll get better value or more performance, just that there will be a feature that will be well marketed that may or may not otherwise justify other sacrifices that might be made.
Consumers don't care and they see AMD branded hardware in every Best Buy across the US. Opteron has enjoyed a good reputation amongst IT professional.
And I find the concern over the AMD brand even more puzzling as Apple doesn't promote Intel. Yes, they mention them in the text, but I find it easy the only logo's you see in the iMac pages are for ATI an NVIDIA, no Intel logo. Apple also doesn't advertise Intel on the outside of the box like other PC manufacturers.
I don't think the average Apple user would notice or care if it was AMD inside instead of Intel inside.
BS. No matter what, AMD is considered to be a second tier supplier.
Sure, if you want a $400 piece of junk, get an AMD machine.
I've got two AMD PC's and they work just as fine as Intel machines.
A cheap AMD will be just as crappy as a cheap Intel, and a good AMD machine will be just as good as a good Intel machine.
I think AMD gets branding issues because they tend to be in the purely bargain basement machines so they get stigmatized by the sucky components their CPUs get surrounded with.
Want to understand why Apple isn't in the under $700 market? Look at the comments quoted above. There are definite advantages to not playing in the ultra-low margin, high volume "value" segment.
Apple ditched ATI and has in the past leaked 'high level talks' to goad its partners into accommodating its needs. I'd expect bing maps before AMD chips...
Nooooo! Apple should stick with Intel (or maybe just start making their own processors like the A4) and intel should let apple use the nvidia chips with their processors. AMD and ATI suck. And intels graphics chips suck so bad. lol
And I find the concern over the AMD brand even more puzzling as Apple doesn't promote Intel. Yes, they mention them in the text, but I find it easy the only logo's you see in the iMac pages are for ATI an NVIDIA, no Intel logo. Apple also doesn't advertise Intel on the outside of the box like other PC manufacturers.
I don't think the average Apple user would notice or care if it was AMD inside instead of Intel inside.
I can't help but wonder if the allowable exclusion of the stickers, which I assume is part of the licensing, is in no small part because Apple agreed to only use Intel chips in their Macs.
The fact is that software is where the biggest bottleneck is in performance right now.
And I would wager Apple is farther along than any other company in this regard right now. You don't think if they weren't contemplating AMD that they haven't been updating the LLVM compiler with the necessary optimizations?
Me, I think it's a no brainer. They ran OSX on Intel in the labs for years. Who's to say they haven't had AMD specific optimizations running the whole time as well? All I know is I wouldn't bet against Apple being prepared...
With OpenCL in Snow Leopard Apple could switch to AMD and not lose any processing performance, right? Wouldn't apps be able to take advantage of better, ATI, graphics to make up for slightly weaker processor speed?
Who's to say the processor will be weaker or slower? Apple is at the fore front of LLVM compiler technology which makes abstractions and optimization for low-level hardware much simpler. This is over and above what OpenCL brings to the table, BTW...
Comments
I believe if Apple is looking at AMD they're looking at the lowend range (Macbook, Mac mini) and the high end (Mac Pro)
Intel is probably going to own the iMac configs. Who knows though but my guess is that AMD is discussing 2011 products because it appears that this is their "make it or break it" year with Bulldozer.
I definitely see AMD CPUs being a viable option in the low-end due to the stronger IGPs, but I don't see the justification for AMD CPUs going into the Mac Pro. One of the major new features of Bulldozer and Sandy Bridge are AVX instructions and 256-bit execution units. Whereas each core in Sandy Bridge has 4 256-bit AVS execution units on 3 ports (MUL, ADD, SHUFFLE, and BOOL) each core in Bulldozer only has 1 admittedly more powerful 128-bit FMAC, but to execute
256-bit AVX instructions it seems both cores in a module need to combine into a into a single 256-bit FMAC unit. There are no doubt more intricacies involved, but it definitely seems Sandy Bridge will have better 256-bit AVX instruction throughput.
What's more, Sandy Bridge seems heavily skewed toward high-end CPUs, particularly Xeons that Apple uses in the Mac Pro. Whereas notebook and mainstream desktop Sandy Bridge seem to be targeting 1.5MB to 2MB of L3 cache per core, Xeon Sandy Bridge seem to be targetting up to 2.5MB of L3 cache per core. Xeon Sandy Bridge also look to have quad channel memory controllers and PCIe Gen 3. Bulldozer's 16 core version seems impressive, but presumably it's actually 8 modules so I wouldn't count an 8-core Sandy Bridge out.
AMD Thuban (Phenom II X6) is scheduled for late April launch(in about 10 days). Some of these chips are already in hands of many enthusiast in Asia. The performance expectations and early leaked benchmarks of Thuban is clock per clock comparable or even better than i7 9xx platform, exception of i7 980X with HT on. This can be attained by almost 1/2 to 1/3 of the price point of intel chips, not to mention intel mobo costing 2x as well.
AMD Thuban incorporates turbo mode which allows increasing the multiplier by 2 to 2.5 (400 to 500 MHz) boost when only 3 cores or less are running just like intel's offering but it does with all three cores at same speed boost. This is the current line up of AMD's offer. Why would you want to use intel i5 or i7 for much higher price for less performance? Not to mention these chips come with 95W/125W flavor.
It looks like AMD is back on track, like the old days of AMD64 X2 vs. Intel P4 days. You may also want to look up Bulldozer for the future line up.
The issue is that Thuban's closest competitors are not the Bloomfield Core i7 which Intel has let stagnate, but Lynnfield Core i7 which have a much more reasonable 95W TDP and much higher Turbo Boost modes than Bloomfield. Intel is also preparing to launch a faster 3.06GHz Core i7 880 Lynnfield and will probably do a price cascade if Thuban is that threatening. If Thuban causes Intel to wake up and speed up the replacement of Bloomfield Core i7 with cut-down Gulftowns I'm all for it.
http://forums.extremeoverclocking.co...34&postcount=2
Not bad. If Thuban is close to I7 then things would indeed be looking very good because AMD can make up performance delta in some areas via superior GPU.
From those benchmarks it looks like Thuban is still slightly clock for clock slower than Nehalem despite a 1.5x core count advantage. But regardless, it does show AMD really closed the performance lead even if it's not done in the most economically feasible way given the larger die size and cost of production for Thuban. It does give great expectations for Bulldozer. I think the top performing X6 1090T has a 140W TDP though so it's not appropriate even for a Mac Pro.
The other wild-card may be that Sandy Bridge sees a return of the CPU microarchitectural stewardship to the Haifa team in Isreal who was responsible for the Pentium M and it's descendants the Core and Core 2 product lines and for regaining the performance and power lead for Intel. Nehalem and Westmere was designed by the Hillsboro team who was responsible for Netburst and Pentium 4 and we can see their influence on getting Nehalem to regain the server performance lead for Intel with lots of memory bandwidth, large slow L3 cache, and Hyperthreading. I'm expecting the Haifa team to implement major in tweaking the microarchitecture to maximize performance/watt using things like decreasing cache latency, hopefully substantially in the case of the L3 cache, increasing interconnect bandwidth, expansion of micro and macro-op fusion, expansion of the loop stream detector, increasing the number of instructions kept in flight and associated buffers to maximize Hyperthreading in anticipation of Bulldozer's SMT, optimizing clock gating, power plane distribution, and the PCU for higher Turbo Boost modes, and hopefully other new tricks.
I'm wondering about Sandy Bridge improvements almost as much as I am with Bulldozer. Someone on the XtremeSystems forums has been claiming for a while that the Sandy Bridge core for the 6/8 core variants will be tweaked compared to the Sandy Bridge core for the 2/4 core variants, presumably to counter Bulldozer.
Most of the rumours seem to indicate quad core mobile Sandy Bridge to maintain the 45W TDP due to the integration of the IGP, but you would think some 35W TDP quad cores would be possible considering the CPU moves from 45nm to 32nm over Clarksfield and the IGP also moves from 45nm to 32nm over Arrandale combined with the Sandy Bridge microarchitecture being tuned for power efficiency should open up some thermal room.
I was a little surprised that quad-core didn't move down either in market segment (so it seems?) or in TDP. The IGP has something to do with it so if it was turned off (when a discrete GPU is on), the 45 W quad-core would probably act like ~35 W.
I definitely see AMD CPUs being a viable option in the low-end due to the stronger IGPs, but I don't see the justification for AMD CPUs going into the Mac Pro.
I can see AMD in the Mac Pro as long as the dual-die variants are used.
From those benchmarks it looks like Thuban is still slightly clock for clock slower than Nehalem despite a 1.5x core count advantage. But regardless, it does show AMD really closed the performance lead even if it's not done in the most economically feasible way given the larger die size and cost of production for Thuban. It does give great expectations for Bulldozer. I think the top performing X6 1090T has a 140W TDP though so it's not appropriate even for a Mac Pro.
That (and roadmaps from Intel and AMD) tells me that AMD is going for more cores for more performance while Intel is going for higher per-core performance (clocks, microarchitectural).
We still have several AMD servers - all are great machines and all of them still run 24/7. Starting with Athlon MP through several Opteron lines ending with the newest dual socket 4 core Opterons, all running Linux. I can imagine Special edition 24 core Opteron Mac PRO - one ugly MF.
So ugly it'll look identical to the Xeons as Apple will swap out the bus architecture for Hypertransport [the same transport they helped pioneer] and the CPUs in the exact same location.
Stunning!! It's sooo damn fugly!
From those benchmarks it looks like Thuban is still slightly clock for clock slower than Nehalem despite a 1.5x core count advantage. But regardless, it does show AMD really closed the performance lead even if it's not done in the most economically feasible way given the larger die size and cost of production for Thuban. It does give great expectations for Bulldozer. I think the top performing X6 1090T has a 140W TDP though so it's not appropriate even for a Mac Pro.
AMD absolutely has a competitive product for workstations and servers, but what I don't see is any competitive product for notebooks (and AIOs, likely Apple's second biggest Mac category).
YoY stats consistently show notebook purchases growing. Does AMD having anything that compete with Intel on the mobile front? From what I've read on AnandTech for a long time (and I'm sure you have too since I've seen you in the forums) AMD is still woefully behind in this area.
I was a little surprised that quad-core didn't move down either in market segment (so it seems…) or in TDP. The IGP has something to do with it so if it was turned off (when a discrete GPU is on), the 45 W quad-core would probably act like ~35 W.
Well there is be hope since there are reports that quad core desktop Sandy Bridge are targeting a TDP from 65-95W TDP, so on desktop TDP is dropping despite adding an IGP. Hopefully the same will be true for mobile quad core Sandy Bridge to have some lower 35W TDP models.
We still have several AMD servers - all are great machines and all of them still run 24/7. Starting with Athlon MP through several Opteron lines ending with the newest dual socket 4 core Opterons, all running Linux. I can imagine Special edition 24 core Opteron Mac PRO - one ugly MF.
I can see AMD in the Mac Pro as long as the dual-die variants are used.
That (and roadmaps from Intel and AMD) tells me that AMD is going for more cores for more performance while Intel is going for higher per-core performance (clocks, microarchitectural).
AMD absolutely has a competitive product for workstations and servers, but what I don't see is any competitive product for notebooks (and AIOs, likely Apple's second biggest Mac category).
The problem with extreme focus on core count is that even professional software is not well optimized for more than 8 cores, much less the 2x12 cores that are possible with Magny-Cours.
http://www.bit-tech.net/hardware/cpu...x5650-review/1
Magny-Cours is really targetted at servers, whereas it fails to beat 6-core Gulftown in workstation usage as in the Mac Pro. The Mac Pro only uses 95W TDP CPUs when configured dual processor, which limits Magny-Cours to the 2.2GHz 80W ACP/115W TDP Opteron 6174 while a slightly cheaper Gulftown alternative is a 2.66GHz 95W TDP Xeon X5650. The Xeon X5650 is overall faster, sometimes much faster. What's worse is that the Opteron 6174 is the fastest 12 core Magny-Cours whereas the X5650 is one of the slower 6-core Gulftown with a 2.93GHz X5670 available at the same 95W TDP. Apple will see no benefit in adopting AMD CPUs so soon with Magny-Cours.
AMD's core count focus will no doubt be better realized in Bulldozer which should adopt Turbo Core like the upcoming desktop Thuban to increase clock speed in low threaded applications, which are going to be most of them for a while. The interesting thing will be seeing how well multithreading solution in Grand Central scales to extreme core counts, although we should already be able to get an idea if there are any well implemented Grand Central applications that someone can test on DP 16 thread Nehalem Mac Pros and presumably future DP 24 thread Gulftown Mac Pros.
YoY stats consistently show notebook purchases growing. Does AMD having anything that compete with Intel on the mobile front? From what I've read on AnandTech for a long time (and I'm sure you have too since I've seen you in the forums) AMD is still woefully behind in this area.
I believe most of AMD's performance in the mobile arena is due to pricing rather than performance. I believe AMD's current Caspian mobile CPUs are based on K10 which would make them competitive with the original Meroms and maybe approaching Penryn, but certainly not approaching Nehalem or Westmere.
AMD absolutely has a competitive product for workstations and servers, but what I don't see is any competitive product for notebooks (and AIOs, likely Apple's second biggest Mac category).
YoY stats consistently show notebook purchases growing. Does AMD having anything that compete with Intel on the mobile front? From what I've read on AnandTech for a long time (and I'm sure you have too since I've seen you in the forums) AMD is still woefully behind in this area.
Guess it depends on how well the Llano APU performs in the Sabine platform. The darkhorse here is OpenCL. AMD is on top of things with this regard. AMD can make up some ground but we need some realworld benefits to OpenCL before we can grok how the GPU can boost general computing software.
Guess it depends on how well the Llano APU performs in the Sabine platform. The darkhorse here is OpenCL. AMD is on top of things with this regard. AMD can make up some ground but we need some realworld benefits to OpenCL before we can grok how the GPU can boost general computing software.
I'm actually curious how good ATI's OpenCL performance is currently compared to nVidia? I believe at least initially, ATI's HD4000 series was woefully behind nVidia's G80 performance, there were even reports that the HD4870 was slower in OpenCL than the 9400M. It's clear that the HD4000 series wasn't fully designed for OpenCL in mind and is penalized by incomplete support for local shared memory in hardware having to drop back to VRAM, but has performance improved by better drivers? Similarly, the HD5000 series has these limitations solved so is performance significantly improved to be competitive to the nVidia GT200 and Fermi?
http://www.anandtech.com/show/2977/n...th-the-wait-/6
Early benchmarks show the HD5870 is still at a distinct disadvantage against the GTX285. I was annoyed that Apple chose to include the GT330M in the new MacBook Pros compared to the HD5650 which would be faster in games and cooler, but the GT330M would presumably be a superior OpenCL choice. As such, it concerns me whether even going with AMD CPUs and Llano will yield an OpenCL advantage over the 320M considering Llano is still based on the same HD5000 architecture. If AMD IGPs don't give an OpenCL advantage over the current nVidia 320M IGP, then maybe Apple should consider just dropping the IGP/chipset licensing moral stance, and redesign the 13" MacBook Pro so it can accommodate a low-end discrete GPU in addition to Intel CPUs which would yield better performance than any IGP, AMD or nVidia, anyways.
Arrandale's IGP does support GPGPU operation with Intel reported to be working on a DirectX Compute Shader driver for later this year. If Apple and Intel could write an OpenCL driver for Arrandale's IGP, it'd be very interesting to see how well it does. At 12 shaders I'm not expecting a miracle, but I wonder if they'll having compute efficiency more similar to what nVidia can do with a small number of stream processors compared to ATI's many SP VLIW strategy which seems to scale better for graphics than compute.
The Mac Pro only uses 95W TDP CPUs when configured dual processor, which limits Magny-Cours to the 2.2GHz 80W ACP/115W TDP Opteron 6174 while a slightly cheaper Gulftown alternative is a 2.66GHz 95W TDP Xeon X5650. The Xeon X5650 is overall faster, sometimes much faster. What's worse is that the Opteron 6174 is the fastest 12 core Magny-Cours whereas the X5650 is one of the slower 6-core Gulftown with a 2.93GHz X5670 available at the same 95W TDP. Apple will see no benefit in adopting AMD CPUs so soon with Magny-Cours.
I see a lot price comparisons to AMD and Intel with AMD always being much lower, but rarely do I see it mentioned how much this cost savings is after 2 or 3 years of power usage for AMD over Intel on servers.
Guess it depends on how well the Llano APU performs in the Sabine platform. The darkhorse here is OpenCL. AMD is on top of things with this regard. AMD can make up some ground but we need some realworld benefits to OpenCL before we can grok how the GPU can boost general computing software.
Nvidia might not be on par with AMD/ATI with OpenCL but it looks like Nvidia has plenty of time to get it figured out.
It's funny that Khronos Group ratified OpenCL with in a record 5 months, yet 15 months later it seems dead in the water yet with so much potential. Hopefully that will change this year with a hopeful revision of many apps to Cocoa/64-bit, including iTunes X around September.
From those benchmarks it looks like Thuban is still slightly clock for clock slower than Nehalem despite a 1.5x core count advantage. But regardless, it does show AMD really closed the performance lead even if it's not done in the most economically feasible way given the larger die size and cost of production for Thuban. It does give great expectations for Bulldozer. I think the top performing X6 1090T has a 140W TDP though so it's not appropriate even for a Mac Pro.
The real benchmark actually scales much better due to more aggressive L2/L3 cache timing vs. prior phenom II X4 and does perform better than the estimated bench scores on the prior link.
Here's a link of one of the enthusiasts playing around the with Thuban 1055T @ 2.8GHz with turbo boost to 3.3GHz on three cores(current base x6 model). He has done some overclocking to 3.2GHz to match i7 980 on cinebench 10 & 11.5 which does beat out i7 at the same clock speed. There are few more posters with their own experience on different threads with other benchmark comparison to i7 scores.
http://www.xtremesystems.org/forums/...d.php?t=249606
The official bench cores will probably flood out as soon as NDA expires on the 26th of April.
cinbench 10 scores of Thuban running at 2.8GHz
estimated bench:
and cinbench 11.5 scores
BS. No matter what, AMD is considered to be a second tier supplier.
Only in your limited view of the world! Besides why do you think there are so many super computers built with AMD chips?
[quote]
Sure, if you want a $400 piece of junk, get an AMD machine.[/QUOTE
]You can buy $400 junk computers with Intel processors in them so what is your point?
In any event there is a common theme in this thread that AMD offers up only low end solutions. That isn't the case at all. They have weak spots right now but so does Intel. The important thing with AMD is that in many ways they have been given a second chance at the low end with Intels Arrandale which as we all know has just about the worst shipping GPU going.
In any event the idea that Apple can't implement a viable AMD system is silly. AMD has a number of products that Apple could successfully implement. Especially in the Mini where Apple needs a product that Intel no longer offers.
As an aside I have to wonder if anybody here has even a remote clue as from where the current architecture came from? It wasn't Intel boys.
Dave
Further, Apple has been designing processors for a long time through it POWER PC partnership with IBM and Motorola. It's expertise is seen in it's new iPads that are quite snappy for relatively low powered devices. If Apple were to use AMD, it would make sure performance wasn't an issue especially if Apple were giving AMD advice.
I might not trust AMD, but I would trust Apple.
I understand that Apple wants to get great deals on chips, but no way they should be satisfied with possibly getting 80% performance at 60% cost. They need to get 100% performance at all times when possible. If AMD can deliver, more power to them. I personally wouldn't buy an AMD-based Mac. I've had minor stability issues with them on the Windows side.
As an aside I have to wonder if anybody here has even a remote clue as from where the current architecture came from? It wasn't Intel boys.
This isn't another all roads lead to DEC Alpha question is it?
Intel's Haifa team, who's working on Sandy Bridge, does have a history with many of the features we're seeing adopted. In the late '90s, they were working on the Timna which among other things included an on die IGP and memory controller. Timna was cancelled, but the experience in SoC design and micro-architecture efficiencies would eventually become the Pentium M and form the basis of the current CPU microarchitecture. Fittingly, the Haifa team will finally get to produce what I believe is the first x86 CPU with native on die IGP in Sandy Bridge. The Intel Pineview Atom does have a on die IGP, but appears to be structured as a traditional northbridge connected via FSB just on the same die rather than something completely integrated.
I might not trust AMD, but I would trust Apple.
Well, after making fun of PCs for using Intel IGPs, Apple switched from the dedicated Radeon 9200 in the PowerPC Mac Mini to the Intel GMA 950. The GMA 950 was decidedly slower than the Radeon 9200 in anything graphically intensive, but did have the advantage of having more flexible/programmable shaders even if they were slower, but sufficient to support light Core Image acceleration which was what Apple promoted. Trusting Apple doesn't necessarily mean you'll get better value or more performance, just that there will be a feature that will be well marketed that may or may not otherwise justify other sacrifices that might be made.
For Geeks.
Consumers don't care and they see AMD branded hardware in every Best Buy across the US. Opteron has enjoyed a good reputation amongst IT professional.
And I find the concern over the AMD brand even more puzzling as Apple doesn't promote Intel. Yes, they mention them in the text, but I find it easy the only logo's you see in the iMac pages are for ATI an NVIDIA, no Intel logo. Apple also doesn't advertise Intel on the outside of the box like other PC manufacturers.
I don't think the average Apple user would notice or care if it was AMD inside instead of Intel inside.
BS. No matter what, AMD is considered to be a second tier supplier.
Sure, if you want a $400 piece of junk, get an AMD machine.
I've got two AMD PC's and they work just as fine as Intel machines.
A cheap AMD will be just as crappy as a cheap Intel, and a good AMD machine will be just as good as a good Intel machine.
I think AMD gets branding issues because they tend to be in the purely bargain basement machines so they get stigmatized by the sucky components their CPUs get surrounded with.
Want to understand why Apple isn't in the under $700 market? Look at the comments quoted above. There are definite advantages to not playing in the ultra-low margin, high volume "value" segment.
Apple ditched ATI and has in the past leaked 'high level talks' to goad its partners into accommodating its needs. I'd expect bing maps before AMD chips...
And I find the concern over the AMD brand even more puzzling as Apple doesn't promote Intel. Yes, they mention them in the text, but I find it easy the only logo's you see in the iMac pages are for ATI an NVIDIA, no Intel logo. Apple also doesn't advertise Intel on the outside of the box like other PC manufacturers.
I don't think the average Apple user would notice or care if it was AMD inside instead of Intel inside.
I can't help but wonder if the allowable exclusion of the stickers, which I assume is part of the licensing, is in no small part because Apple agreed to only use Intel chips in their Macs.
The fact is that software is where the biggest bottleneck is in performance right now.
And I would wager Apple is farther along than any other company in this regard right now. You don't think if they weren't contemplating AMD that they haven't been updating the LLVM compiler with the necessary optimizations?
Me, I think it's a no brainer. They ran OSX on Intel in the labs for years. Who's to say they haven't had AMD specific optimizations running the whole time as well? All I know is I wouldn't bet against Apple being prepared...
With OpenCL in Snow Leopard Apple could switch to AMD and not lose any processing performance, right? Wouldn't apps be able to take advantage of better, ATI, graphics to make up for slightly weaker processor speed?
Who's to say the processor will be weaker or slower? Apple is at the fore front of LLVM compiler technology which makes abstractions and optimization for low-level hardware much simpler. This is over and above what OpenCL brings to the table, BTW...