PS4's APU Exposed

Not exactly everything but still, there are some new details here and there since Sony's too scared to release the full specs of the console.

Despite continuing claims from some that the SoC must contain a higher-end Kaveri/Steamroller-class CPU, the tiny x86 cores implemented here are clearly based on Jaguar/Kabini. Each core is roughly 3.1 millimeters square — exactly the size AMD gave for that chip. The large (rather plaid) boxes in each quad-core arrangement are the L2 cache. Memory I/O wraps all three sides of the die, which makes sense — the SoC itself uses a GPU-style memory layout. It’s not clear from this diagram whether HSA is implemented on the chip or not, though it might be possible to identify the IOMMU unit that HSA requires with a close analysis.

Die size on the chip is 328 mm sq, and the GPU actually contains 20 compute units — not the 18 that are specified. This is likely a yield-boosting measure, but it also means AMD implemented a full HD 7870 in silicon. According to Chipworks, the GPU is 88 mm sq, and takes up about a third of the total die. Looking at AMD’s published figures for the HD 7870, however, the Pitcairn GPU core is a 228 mm sq part.

The Wii U’s design was a bit prophetic

One of the interesting things about the PS4′s design is how closely it echoes the Wii U’s design. While the PS4 is far, far more powerful than the Wii U, Sony made very similar decisions about where to spend its transistor budget. As some of you may recall, the Wii U’s GPU is far larger than its CPU — the latter is a triple-core version of the Broadway core that powered the Wii (which was, itself, a higher-clocked version of the GameCube CPU). The GPU, while still underpowered by current standards, integrated far more computational horsepower than the old Hollywood GPU inside the Wii.

More Here

Nice. No wonder PS4 outperforms modern gaming pc's.

I hope that was sarcasm...

? uh... no. Why would it be?

Get out.

don't take the bait

Why? Because it was a pretty dumb thing to say. My PC is 4+ years old with a video card that is about a year old and it blows my PS4 out of the water. No comparison.

Because is not even remotely true.

all evidence proves otherwise.

Depends highly on what you call a "modern gaming PC."

In pure CPU capability it's grossly outclassed. Even 4 year old quad core Phenom IIs or current dual core i3's have as much CPU GFLOPS and a similar amount of serial CPU processing as the 8 Jaguar cores in the PS4's APU. The graphics array is certainly equal to more capable that a general mainstream gaming PC (not really sure what I would call a 'median' graphics card, maybe a 7770 or Geforce GTX 650). However, a 20 CU AMD graphics processor like the 7870 not only has full access to all 20 of it's CUs but also will be clocked close to 1 GHz, not 800 MHz like on the PS4. That's 1.84 TFLOPS vs 2.56 TFLOPS.

To play Devil's Advocate, the PS4 is certainly well equipped to maintain a relative feature parity with PC gaming, and it's GPGPU capabilities are still somewhat a mystery until we see full documentation or real world examples of games that make full use of it. On PC, obviously this would be more difficult to implement, but the PS4 must use the CUs at the cost of rendering time. On PC we may see GPGPU used on an AMD APU, on an actual GPU or both. Obviously these components can be upgraded too.

It's interesting that the two powerhouse consoles this cycle show no real performance advantage over high-medium end PC hardware, especially considering how cheap PC hardware is now. Anyone with an i5 or FX-8 + a Radeon 7950 or R9 270 or better is pretty much set for this generation unless games get truly GPGPU happy.

As for myself, I'm stlll sporting a 4 year old Radeon 5850 1 GB. While it is long in the tooth with only 1 GB of memory and much weaker tessellation hardware than today's modern GPUs, in many ways it still holds it's own at 1080p and we know how bad the Xbox One is equipped for 1080p ;D

PS4 don't need no powerful cpu cuz it don't need to run antivirus and spyware programs like PC does when hermits play farmville and world of warcraft.

It just needs to run its' bloated OS.

And that only costs 2 cores & 37% of the RAM. Chump change, really.

It's really the GFLOPS in a CPU that a console needs more than anything to run animations, physics, sound, etc Jaguar cores offered good TDP/performance/mm² with a very comprehensive feature and easily adaptable to various APU solutions. However their thermals and power needs go up quick like any CPU core would as you increase the clock speed. Modern Intel Core architecture offers twice as many GFLOPS per clock (256 bits wide) than a Jaguar core (128 bits wide) and on desktop PCs, they'll easily run upwards of 3.2 GHz, so even dual core Intel i3's have as much GFLOPS as the 8 Jaguar cores in the PS4's APU. Scalar performance matters too and the advantages in scalar performance and GFLOPS that an i5, i7 or octo core AMD FX CPU offer can get PC gamers the 60 FPS or smoother framerate they may be craving while the two consoles may be limited to 30 FPS in CPU heavy scenes and/or experience even less if the scene is incredibly taxing.

The Xenon CPU in the 360 was a fine example of a a reasonably sized processor with heavy focus on high GFLOPS output and good for it's time general CPU performance which was necessary for advancing the amount of physics and quality of animation in games while still having the grunt to handle good AI and do background tasks.

The Cell in the PS3 was an example of focusing way too much on GFLOPS while having barely enough general scalar performance. Luckily for the PS3, the excess GFLOPS capability on the Cell could be used to augment the PS3's weaker graphics processing unit.

So cute when you see console gamers trolling like that.

Are you saying my R9-290 + Mantle can't do real time GpGPU?

Indeed,but you know it can always get dumber :P

TLHBO

@ronvalencia:

What I'm saying we don't know how GPGPU + Mantle will do on PC until

we actually see it

especially when we consider how efficiently it's implemented. Assuming we actually get to experience it, I'm going to guess that the best way to experience "real" (and not just effects physics) GPGPU will be with a Kaveri APU handling GPGPU with the benefit of HSA and hUMA so there is no extra saturation of the PCI-E bus with the likely associated communication latency between the CPU and GPU when doing GPGPU. It would leave the dedicated graphics card left to do graphics only. Kaveri with 800+ GFLOPS on the CUs is quite a bit of processing power yeah? I think it would be more efficient to keep it all handled on the same piece of silicon and let the graphics card do it's thing without being distracted. It could mean actual gameplay GPGPU physics that we hardly saw with PhysX.

You are referring to GpGPU not CGPU workloads. PCI-E 3.0 16 lanes can easily handle real time compute returns from the GPU i.e. the big bottleneck was Direct3D and AMD Mantle replaced Direct3D.

NVIDIA already has a low overhead dispatch driver via Tesla Cluster Compute (TCC) driver and NVIDIA is murdering AMD on HPC.

NVIDIA could have physics compute that can return within a frame render. NVIDIA just needs to stop milking the hermits release TCC's benefits for the general GeForce Windows drivers.

For games, the GpGPU results needs to be returned to the CPU within a frame render. From http://www.slideshare.net/zlatan4177/gpgpu-algorithms-in-games

AMD talks about CGPU/GpGPU workloads for both dGPU and IGP.

With PS4, the bandwidth for CPU to memory (i.e. main memory being use as GPU memory) is ~20 GB/s (peak total).

With PC and PCI Express 3.0 16 lanes, the bandwidth for CPU to GPU memory is ~32 GB/s (peak total).

----

With PS4, CPU to GPU interconnect bandwidth is ~20 GB/s or~10 GB/s per direction.

With PC and PCI Express 3.0 16 Lanes, CPU to GPU interconnect bandwidth is ~32 GB/s or ~16 GB/s per direction.

---

PS4 can easily beat Intel Sandybridge/AMD Trinity/AMD Piledriver era PCI Express 2.0 16 lane.

AMD Kaveri APU would be AMD's first PCI-E version 3.0 capable I/O solution that would break AMD's ~20 GB/s CPU I/O limits.

To speed up CPU-to-GPU interconnects, I would need to upgrade my Intel Sandybridge to Intel Ivybridge or Intel Haswell i.e. I'll update when it's required.

Cows are mad, what do lemmings, herms and everyone else have to do with this?

Who fvcin cares. just get some games out

AMD's GCN CU includes a scalar processor and it's not expose by the current Direct3D nor OpenCL (wait for HSA or Mantle drivers) i.e. programming GCN's at ASM level, GCN ISA acts like a cut down IA-64 with very wide SIMD.

You can't compare GFLOPS across different processor designs.

------------

http://beyond3d.com/showpost.php?p=1788668&postcount=693

bkilian Senior Member

Everyone talks about this 100GFLOPS, but all I've ever said is that the entire 360 CPU would not be able to run Shape at 100%. Remember that the 360 CPU gets, on average, an IPC of 0.2. So while the chip is _technically_ 100GFlops, you will never see that in real running code. Jaguar cores have an IPC of close to 1. A single jaguar core could almost outperform the entire 360 CPU on production code.

How would PS4 setup compare to a 7790 GPU ?

Would it be much faster, or just faster ?

Why do you care? Pc has no rpg's according to you.......

Why don't you post more? You're a far more worthwhile read than Tormentos VS Ronvalencia.

Also, it's worth considering how saturated the PCI-E bus actually is. Concensus seems to be that PCI-E2 is barely a bottleneck for even Titan, as far as I'm aware.

A quick look at the specs of the card show that the parts of the card range from 25% to 50% faster for ps4. If we're just estimating, the ps4 is probably ~33% faster. Of course, I have no idea how the 8gddr5 will play out in comparison to PC benchmarks.

Also, if you're asking specifically because of that digital foundry comparison, the 7790 is more powerful than what is in the xbone. The digital foundry comparison was based on a downclocked 7790 when a 7770 was a much more accurate comparison. I'm honestly not sure why they did that.

Edit: To correct myself, apparently they used a downclocked 7850 for their xbone comparison.

Oh look 20 CU like i always say 7870 2 CU disable for redundancy..lol

Oh is gimped 7790 vs gimped 7870..

How exciting.

like seriously dude, you once said that you had a BA on computer science but here you're making these questions... PS4 is faster even if they have similar Tflop output.

PS4 has twice the Rops. PS4 has ACE. PS4 has more bandwidth. PS4's was customized to allow 64 queue. PS4 has a complete advantage on programming to the metal, no layers of Windows or drivers getting in the way either.

godly made thank you sony

Hooray, now where are the good games?

I agree with this sentiment ^.

Is there a transistor count for this APU? I can't find one in the article.

My Ti has more.

tormentos vs ronvalencia incoming. HIDE!!

I am out of the GPU loop for a while, i am not one of those that pretend to know everything, especially something i have not followed or read about in years

What does that have to do with my degree ? Knowing how to design a CPU in FPGA for example does not make me an expert on the latest cutting edge GPU tech or how each GPU compares to each other

In any case, here is the situation. I am developing a RPG game rather heavy on graphics and i want to target PS4 with an option to be able to port for xbox one later. I already had an old PC with dual core CPU and 4GB ram and i upgraded the ATI 4500 to AMD 7790 in order to be somehwere between xbox one and PS4

PS4 was thought to be like 7790 before, but with double the ROPs. I am aware PS4 would be faster due to not having my ancient CPU and ram setup, where i am mostly limited in right now in Unity anyway.

The question is, if i can get 20-30fps for my game now on 7790 and my PC setup, how much gain could i expect from PS4 and how many sacrifices would i have to make for xbox one

That is what i want to get a feeling on

What is ACE btw ? :)

@clyde46: and mine

if you agree with the sentiment then why did you even click on the thread and took the time to write the comments? You want to talk about games then go and create your own thread about it. It's that easy.

Looks like a Battlefield 4 mini map

lol - I agree with Cainetao's sentiment about desiring more games are released shortly. The 'who fucking cares' part, I don't agree with that specifically. Besides if I did, wouldn't I just quote that alone and write as little as possible?

Think about it Gue1. Is it not possible to want games to come out, and to want to know the transistor count for the PS4's APU?

Is there a transistor count for this APU or not?

Why spam the boards with more meaningless threads?

it has everything to do with your degree because on the BA for CS you need to take 2 low level classes. One is Assembly Language where you need to understand CPU x86 architecture to be able to program for it. You learn stuff like the registers of the CPU, the FPU, etc. Then there's another course about Computer Graphics where you learn to use openGL in conjunction with C where you go a bit deeper on GPU architecture and how it works before you even start to render your first openGL 2D shape and vertices.

Like seriously... This is only basic stuff. By looking at a barebones spec sheet you should be able to tell all of this in an instant.

BTW ACE is AMD's fancy way to say: queue list to feed the CUs with workloads. ACEs decides the resource allocation, context switching, and task priority. Each compute engine can take up 2 sources at a time but for the PS4 it was customized for each one of them to be able to handle up to 8 sources (8 compute engines * 8 sources = 64 sources).

@Gue1 said:

@psymon100 said:

@cainetao11 said:

Who fvcin cares. just get some games out

I agree with this sentiment ^.

Is there a transistor count for this APU? I can't find one in the article.

if you agree with the sentiment then why did you even click on the thread and took the time to write the comments? You want to talk about games then go and create your own thread about it. It's that easy.

I understand the PC gamer caring about stuff like this because they have the ability to upgrade and change it. In consoles its ridiculous FOR me. I cant change it so who cares.

@MlauTheDaft:

LOL I don't care for Gamespot's new look but more so it's just now starting to run as efficiently now that bugs have been worked out so I've only just recently started posting regularly again.

And Ronvalencia is pretty damn smart, he's just not always coherent with the way he writes his messages.

Tormentos is just Tormentos.....lol

I'm just trying to approach this generation as practically as possible. At the beginning of last gen, I had just began my tenure as a PC gamer (mid 2005), but still did not completely understand the differences between the consoles and PC architectures, nor did I have a good gaming PC until early 2007. My current PC is old, but ironically enough the CPU is on par with either the PS4 or Xbone's CPU and in some ways my GPU (Radeon 5850) is just as capable, but practically it lacks the kind of tessellation that either console has thanks to more modern GCN graphics architecture implementation as well as decent GPGPU.

What still kind of blows my mind is how AMD hasn't made any plans to release server versions of the PS4 APU. For a general desktop solution, it's CPU cores are too slow both in clock and too slow per clock, but for a server focused GPGPU, it would be good all in one solution not requiring separate control CPUs and GPUs, much like Cell but with a practical basic CPU capability.

Kaveri certainly isn't looking bad but it should've been around a year ago and the highest end desktop version should've been quad channel or AMD should've made a "super Kaveri" variant with 768 SPs with quad channel support. It would be good for cheap multimedia desktops, servers, and most importantly a great entry level Steam Box APU which could sell millions.

Is the PS4 more powerful than three R290X?

Heck no, PS4 GPU 1.84 TFLOP, R290X 5.4 TFLOP along with more memory and bandwidth you do the math

So the R290X is slightly more powerful?

didnt know 3x+ more powerful is slight.....

I agree normally. But with console optimization they can squeeze out more power.

One shouldn't forget the PS4's 8gb GDDR5 is very powerful.

I think most multiplats will run the same with the PC GPU having a bit of an edge in resolution.