Relative performance of AMD 2.9GHz APU vs current consoles?

RudeCurve

Banned
Now that AMD has a quad core procoessor + dual core graphics on a single chip, I was wondering how it compares to current generation consoles specifically Xbox360 and the upcoming WiiU. So is it more powerful, equal or less powerful?
 
Yes, I guess it has a single GPU core. I think I read somewhere the specs and it was wrong. Anyway how does it compare?

My second question is since it's fabbed on 32nm process, would a single chip Xbox CPU+GPU made on 32nm process be bigger or smaller than the AMD APU?
 
I believe a Llano has more than twice the transistors as a single chip Xbox 360 CPU/GPU. Presumably the 360 chip would be significantly smaller.
 
The 360 Soc is only 35mm^2 @ 45nm, so there's already no contest. You could fit 6 360 SoCs inside Llano.
 
Last edited by a moderator:
The 360 Soc is only 35mm^2 @ 45nm, so there's already no contest. You could fit 6 360 SoCs inside Llano.

Haha thanks for that, so I guess using 32nm an updated 360 SoC with the same die size as AMD's APU would be more powerful...eg at minimum 12PPU cores + 1000 shader cores.:smile:
 
Actually they had to go out of their way to not make this one more powerful:

...[T]he purpose of the "FSB replacement block" may not be obvious. This particular block essentially implements a kind of on-die "frontside bus" with the exact same latency and bandwidth characteristics as the older bus that connected the CPU and GPU when they were discrete parts.

It would have been easier and more natural to just connect the CPU and GPU with a high-bandwidth, low-latency internal connection, but that would have made the new SoC faster in some respects than the older systems, and that's not allowed. So they had to introduce this separate module onto the chip that could actually add latency between the CPU and GPU blocks, and generally behave like an off-die FSB.

http://arstechnica.com/gaming/news/2010/08/microsoft-beats-intel-amd-to-market-with-cpugpu-combo-chip.ars
 

That has no relation to adding more cores though. That's just something they added to keep signal timing the same when going from two external chips into one SoC.

What I'm talking about is more CPU cores and more shader cores packed into the same space as an APU = more powerful. For running X360 games you just turn off all the cpu cores except 3. For running X720 games you turn on all 12 cpu cores.
 
Good question, common sense would dictate it would be much larger than that :oops:. I was looking at Page 3 of these slides: http://www.hotchips.org/uploads/archive22/HC22.23.215-1-Jensen-XBox360.pdf What do they mean by 35 x 35 mm package?

The dies are in a package. The package is connected to the motherboard. 35mm x 35mm would be 1225mm2 :smile:. That is the size of package.

Often times the chips and the package are covered by heatspreaders and some confuse the the spreader as the chip.
 
Last edited by a moderator:
So even though it has inside A8-3850 gpu with 400 strean processors, if Xenos gpu and Xenon cpu (x360S) integrated at 32nm with about half size and relation watts vs size vs power processing would have a better ratio performance for game porposes(despite only DX9 graphics level)?
 
Last edited by a moderator:
So even though it has inside A8-3850 gpu with 400 strean processors, if Xenos gpu and Xenon cpu (x360S) integrated at 32nm with about half size and relation watts vs size vs power processing would have a better ratio performance for game porposes(despite only DX9 graphics level)?

No. CPU on APU has one more core and is superscalar vs. Xbox360 in order counterpart. For general code AMD solutions should be way faster than what Xenon can offer.
On GPU front there is simply no comparison. XBOX GPU only savior is it's EDRAM for virtually free AA, but this comes with downsides as well.

Where APU is hold back is software ecosystem (lack of direct optimization for specific hardware) and low memory bandwidth (somewhat mitigated by DDR3-1866).
 
No. CPU on APU has one more core and is superscalar vs. Xbox360 in order counterpart. For general code AMD solutions should be way faster than what Xenon can offer.
On GPU front there is simply no comparison. XBOX GPU only savior is it's EDRAM for virtually free AA, but this comes with downsides as well.

Where APU is hold back is software ecosystem (lack of direct optimization for specific hardware) and low memory bandwidth (somewhat mitigated by DDR3-1866).


Thanx a lot to enlight me,there are many others ascpects APU can shine....cause i was already thinking where to put multiple chips X360s would be the best option.:oops:
 
The XCGPU is ~500M transistors, while the 4-core Llano has 1450M transistors.
Theoretically, one could put 3x the XCGPU's functional units "inside" a Llano. That means a 9-core PPU, a 144 vec4+scalar (720 "shaders" comparable to how many VLIW5 shaders? 480?) and 30MB eDRAM.

However, even the 45nm XCGPU is already consuming >80W at load, so a "3xXCGPU" chip could have a prohibitve power consumption for a console, if they retained the same clocks.



Going to the question at hand, it's a good guess that a 4-core, 400sp Llano is at least 2x more powerful than the XCGPU in the graphics department, but may not retain the same advantage in some FP-intensive CPU stuff like physics.

It doesn't go without saying that Llano is essentially a laptop chip, so it's "squeezed" to operate in the single-digit watt power consumption, as long as possible.
 
The XCGPU is ~500M transistors, while the 4-core Llano has 1450M transistors.
Theoretically, one could put 3x the XCGPU's functional units "inside" a Llano. That means a 9-core PPU, a 144 vec4+scalar (720 "shaders" comparable to how many VLIW5 shaders? 480?) and 30MB eDRAM.

However, even the 45nm XCGPU is already consuming >80W at load, so a "3xXCGPU" chip could have a prohibitve power consumption for a console, if they retained the same clocks.



Going to the question at hand, it's a good guess that a 4-core, 400sp Llano is at least 2x more powerful than the XCGPU in the graphics department, but may not retain the same advantage in some FP-intensive CPU stuff like physics.

It doesn't go without saying that Llano is essentially a laptop chip, so it's "squeezed" to operate in the single-digit watt power consumption, as long as possible.

This is interesting debacle, but we need to remember things designed for purpose are much more efficient at given tasks. For instance PowerPC chip inside XCGPU has only 1MB L2 cache shared across all cores where Llano has 4x1MB L2. Same goes for L1 caches which for AMD are bigger. These are few hundred millions of transistors spent on caching alone on AMD side just to eek few percent better IPC from varied code base this chip needs to run. Then we have interconnect networks, various FIFO buffers, etc to 'glue' together all the components when adding cores and shaders efficiently. One can fantasize how powerful Xbox360 chip could be if given 1.45 billion transistor budget and extra power headroom, but in reality things don't scale that easily. Definitely not for free, it cost extra transistors :)

Anyway I bet AMD / Intel / IBM can easily come up with some fantastic chip given nearly 2 billion transistors and rumored 28nm process. Not adhering to x86 is a bonus in efficiency terms and producing closed system helps achieving very high throughput computing given proper compiler / software optimization.
I wonder if next gen CPU for consoles will still be in order or they will move to OoO cores.

GPU wise only today I've read AMD won all 3 console contracts, which means Sony, Microsoft and Nintendo will run using AMD's (ATI) IP. Interesting if true! How much different PS4 and XBOX 720 GPU's will be given that same company is designing them and performance target should be similar?! Will AMD bother to design them in a very different way or both chips will be very similar at low level hardware?
 
Last edited by a moderator:
xenos=48x5=240

I know that Xenos has vec4+scalar ALUs, being capable of 5D operations, hence the 720 number (240*3).

However, the Xenos' shaders are not VLIW5 like the desktop/laptop AMD GPUs, and the R600->Barts shaders are more efficient.

Xenos' transistor count is lower than RV635 (HD36xx, 120 shader processors, 8 TMUs, 4 ROPs), and the latter has no eDRAM, half the TMUs and half the ROPs.
You can tell there are a lot more transistors dedicated to the shader processors in the VLIW5 shaders than the X360's Vec4+scalar.


This is interesting debacle, but we need to remember things designed for purpose are much more efficient at given tasks. For instance PowerPC chip inside XCGPU has only 1MB L2 cache shared across all cores where Llano has 4x1MB L2. Same goes for L1 caches which for AMD are bigger. These are few hundred millions of transistors spent on caching alone on AMD side just to eek few percent better IPC from varied code base this chip needs to run. Then we have interconnect networks, various FIFO buffers, etc to 'glue' together all the components when adding cores and shaders efficiently. One can fantasize how powerful Xbox360 chip could be if given 1.45 billion transistor budget and extra power headroom, but in reality things don't scale that easily. Definitely not for free, it cost extra transistors

That's true, however they wouldn't need to triple everything. I/O blocks, the memory controller and ROPs wouldn't make much sense to triple in transistor amount, so they could save a lot in there.

But as you said, even if 1.45B transistors could be enough to "triple" a XCGPU, I'm pretty sure that wouldn't be of the best interest to Microsoft, AMD or IBM :)



GPU wise only today I've read AMD won all 3 console contracts, which means Sony, Microsoft and Nintendo will run using AMD's (ATI) IP. Interesting if true! How much different PS4 and XBOX 720 GPU's will be given that same company is designing them and performance target should be similar?! Will AMD bother to design them in a very different way or both chips will be very similar at low level hardware?

That rumour came from HardOCP alone and I have a hard time believing in it. Nintendo sharing the GPU maker with any other is okay because their consoles are displaced from the others in generational leaps.
However, I don't see any console maker buying a GPU design without a "GPU exclusivity" contract, and I don't see AMD having the man-power to design 2 console GPUs + discrete GPUs + iGPUs at the same time.
 
Last edited by a moderator:
I remember in year 2005 when Xbox 360 launched, some AAA launch title developer said the Xbox CPU is slightly less powerful than dual core Athlon 64 for running average code. At that point Athlon 64 X2 was the most powerful PC CPU, and it ran at 2.0 GHz. Assuming this old info is even somewhat true, we can draw some conclusions... which of course can be debated a lot, since programmers didn't know that well how to efficiently utilize a 6 HW thread CPU at that time.

According to official AMD launch information (collected from reviews around the net), we get the following info about their architecture IPC increase over the years:
Athlon64(k8) -> Phenom(k10) +25%
Phenom -> Phenom2 +12%
Phenom2 -> Llano(k10.5) +6%

So, the bottom line is: Athlon64(k8) -> Llano(k10.5) +48.4%

So compared to Athlon 64, Llano has 48.4% better IPC at same clocks, has 45% higher clocks (2.0->2.9), and has four cores instead of two. So in theory Llano 2.9Ghz performance relative to Athlon 64 2.0GHz should be 430% (over four times as fast). So no matter how inaccurate the comparison between Xbox 360 CPU and dual core Athlon 64 was at that time, Llano CPU should be considerably faster.

The theoretical GPU estimation can be easily calculated from the available info. Of course the real efficiency of various architectures is different (but it's safe to say that newer architectures get more job done per cycle). So yes, Llano GPU half should be considerably faster also. Overall it's pretty safe to say that Llano should be over 2x faster than either of the current consoles (both are pretty near when we are talking about real games). But that off course assumes the game would be highly optimized for Llano. I doubt many PC game developers are spending months and months to optimize just for Llano. Maybe we are lucky and get a next gen Bulldozer based APU in a future console, so we can actually see how these fusion chips fare when developers start to really optimize for them.

Update:
Found a forum with lots of Brazos E-350 APU benchmark links. Brazos is a netbook chip designed to compete with ATOM (much lower end than Llano).

Brazos fares pretty well in current console games:
Battlefield 2 (2005) - 1024x768, all max, 25-35 fps
Crysis 2 (2011) - 1280х720, 20 fps
Dirt 3 (2011) - 1024x600, 19-25 fps (ultra low noAA 25-28 fps)
Mass Effect 2 (2011) - 1024х600, 20 fps
Call of Duty: Modern Warfare 2 (2009) - 1280х720, 25 fps

Current consoles run those same games at 30 fps at similar resolutions. These games are definitely not optimized for Brazos architecture. Maybe we get some netbooks for the Christmas that are more capable than the current consoles :)
 
Last edited by a moderator:
Llano has about the same IPC as Athlon II, not Phenom II. Phenom II is to varying degrees faster than both.

I'm really skeptical about Xenon being particularly competitive with an Athlon 64 X2 without a lot of attention to how the program works. As far as we know Xenon might be more comparable to Brazos or Atom.
 
Back
Top