Technical Comparison Sony PS4 and Microsoft Xbox

[dagamer]

Vita soc is going to be outpaced once devices with series 6 GPU cores come out.

If it hasn't been already.

Doubtful Sony is going to double down on a failed product, other than a cost reduction revision.

Are there any mobile games which on a technical level pass the best available on the PS2? They are too few and far between, perhaps Real Racing 3 would be one.

 

[bkilian]

While what he said is certainly true, what he didn't address is that the GPU can do other work during that time, now if it takes up a huge amount of frame time its going to be a problem. put if its doesn't then it really won't be as the GPU is still doing work whilst its waiting on memory. He also didn't mention the texture cache that exists per CU whilst small it probably has a decent hit ratio reducing this problem even further.

Each SIMD unit also contains a 64KB register file, which seems huge to me. How much use for textures I dunno. but dayum.

You are correct, the GPU will switch to another wavefront (64 thread work unit) on a memory stall. To make efficient use of the GPU you need enough wavefronts available so that memory stalls don't impact you too badly. Switching wavefronts has a cost though (something around 4 cycles I think). A memory stall can cost upwards of a thousand cycles, that you have to fill with other wavefronts. Like I said, a big GPU cache will not help for large streaming jobs, only for small, diverse jobs. It remains to be seen if it will have any practical difference once developers start coding to it.

 

[3dilettante]

I'm not sure which switch penalty would be four cycles. Four cycles is the round-robin delay for instruction issue amongst the SIMDs, which is part of steady-state execution.

Wavefronts don't really switch out until they are done executing, and I'd imagine the resource release and startup of a new wavefront should take longer.

 

[dumbo11]

Silly question, but in talk of ESRAM does it seem likely to anyone that the kinect depth buffer/HDMI input sit in there?

Either a 'reserved timeslice' (the GPU dumps the contents of ESRAM into DDR3, does it's stuff, then reloads the data from DDR3) or via an actual reservation?

That just seems rather essential to me...

 

[ERP]

Silly question, but in talk of ESRAM does it seem likely to anyone that the kinect depth buffer/HDMI input sit in there?

Either a 'reserved timeslice' (the GPU dumps the contents of ESRAM into DDR3, does it's stuff, then reloads the data from DDR3) or via an actual reservation?

That just seems rather essential to me...

I'd be surprised if either one were ever in ESRAM.

 

[zupallinere]

Sony: latency vs gigabytes

The implementation for Sony's memory bus and cache subsystem is also not known well enough to know what workarounds are available. Some optimizations such as assigning output and input buffers to certain addresses may allow for more optimal traffic patterns on a per memory channel basis, even if the global traffic appears to be pathological.

Sony may have opted for a known but consistent longer-latency base case on a multi-gigabyte allocation, whereas Microsoft has a shorter-latency case with a cliff at the end of the eSRAM allocation.

I agree it would seem very possible that Sony will want to put any extra memory to good use on that front. Setting aside the CPU, what would PC graphics card manufacturers do, if they had say 6 gigs to play with, and what would they do to deal with long but consistent latency issues

I am sure this has been discussed before so forgive me for being redundant but one might posit that AMD itself or others might be looking at how their drivers and PC GPU tech could be tuned for such memory abundance (or pushed for more optimizations if they have started that already). Since the PS4 shares the same CPU ( allegedly ) and is basically a PC card from the GPU back to memory there might be companies that target PS4/PC first. Just a thought.

If and I mean IF the PS4 holds say a 1.5 to 2 gig advantage in memory might that be a useful resource that is relatively handy even if the hardware target for a cross-platform game is the XB1 5 gigs.

 

[Pete]

let's take a gpu intensive game as the first PC version of crysis (I think the game that relies more heavily on GPU in our recent history)
[...]

Spoiler

That (heck, most) 10x7 benchmark is clearly CPU-limited, judging from the tiny avg fps spread b/w the slowest and fastest GPUs. Let's avoid SLI/XF issues and look at single GPUs at a more reasonable (and hopefully more relevant) res, 1080p: http://www.hardware.fr/articles/890-25/recapitulatif-performances.html

The comparison b/w the 7770, 7790, and 7850 are kind of interesting.

         7770    7790    7850   Xb1?*  PS4?*
tri/s    1       2      ~2      1     1
GFLOPS   1      ~1.4    ~1.4    1     1.5
Gtex/s   1       1.4     1.4    1     1.5
Gpix/s   1       1       1.7    1     2
GB/s     1      ~1.3    ~2      1     1.???

avg fps  1       1.3     1.55

*assuming same clocks

Now, given that the games tested are coded at a higher level and for a broader set of target hardware, we can at least see that perf scales roughly with ALUs/TMUs/bandwidth. So, assuming the same clocks, no CPU bottleneck, and the same graphics settings, a game not optimized for one console over the other should run faster on the PS4 on the basis of its faster GPU. But who knows how next-gen ports will be coded, and how evident that framerate difference will be once we account for effects tailored for each console's strengths.

 

[Strange]

the 6990 is not 50% but 100%+ faster than 6970
and how this is in real world?

so using a GPU 100%+ faster the gain is 4 FPS in 1024x768 (5%) and 24 FPS in 1280x1024 with AA on (20%)

so in the worst scenario, (a) are you really think you'll see a 30 FPS VS 45 FPS or 30 FPS VS 32-36 FPS?

A much more genuine comparison would be, well, from the same link

Which takes you away from the CPU bound world

From your criteria comparing a 6990 to a 6970 (~100% increase)

1920*1200
66.7 fps vs 37.4 fps (+78%)
and

2560*1600
43.1 fps vs 23.5 fps (+83%)

So in a sense yes, a 50% increase in GPU capability (Xbox One vs PS4) would probably result in a much better frame rate.

I'm personally fully expecting a ~40% increase in frame rate, without even considering bandwidth issues which could make matters even worse (or at best on par) for the Xbox One.

So put in real terms, if the Xbox One is running at 30 fps Vsync (assuming around 40 fps average to actually do that without hiccups), I wouldn't be surprised at all if the PS4 can easily hit ~50 fps on average.

 

[MBDF]

What is the jump in polygons/triangles per second like going from Gen 7 to Gen 8.. is it substantial?

 

[dagamer]

What is the jump in polygons/triangles per second like going from Gen 7 to Gen 8.. is it substantial?

It should be pretty significant though we aren't going to be able to look at polygon counts of character models for comparisons because tessellation is expected to be used this time around. You aren't going to see a fully tessellation model on screen all at once.

 

[rockaman]

Are there any mobile games which on a technical level pass the best available on the PS2? They are too few and far between, perhaps Real Racing 3 would be one.

There are a few I'm sure. Remember that tablets can run GTA VC with relative ease, at least some of them. And at higher resolutions than PS2.

Comparing the same games to Vita though.... I'd be very surprised if there are any games currently available on iOS or Android that can surpass Uncharted GA or Killzone Mercenary on a technical level aside from texture resolution. Some of the Killzone Vita screens look almost like Killzone 2 and 3...

 

[Betanumerical]

A much more genuine comparison would be, well, from the same link

Which takes you away from the CPU bound world

From your criteria comparing a 6990 to a 6970 (~100% increase)

1920*1200
66.7 fps vs 37.4 fps (+78%)
and

2560*1600
43.1 fps vs 23.5 fps (+83%)

So in a sense yes, a 50% increase in GPU capability (Xbox One vs PS4) would probably result in a much better frame rate.

I'm personally fully expecting a ~40% increase in frame rate, without even considering bandwidth issues which could make matters even worse (or at best on par) for the Xbox One.

So put in real terms, if the Xbox One is running at 30 fps Vsync (assuming around 40 fps average to actually do that without hiccups), I wouldn't be surprised at all if the PS4 can easily hit ~50 fps on average.

What I don't understand is why he didn't link to the very last graph which averages every single result in the entire article and shows you relative performance

Spoiler

It shows the 6970 as having 69% of the power of the 6990 which is roughly correct given that each of the 6990's two 'GPUs' are only 94% as power as a 6970.

You'll also notice that higher resolutions that scaling is nearly perfect.

Spoiler

 

[Strange]

What I don't understand is why he didn't link to the very last graph which averages every single result in the entire article and shows you relative performance

Spoiler

It shows the 6970 as having 69% of the power of the 6990 which is roughly correct given that each of the 6990's two 'GPUs' are only 94% as power as a 6970.

You'll also notice that higher resolutions that scaling is nearly perfect.

Spoiler

I'm not going to comment on why he didn't link to the more representative graphs, but I wouldn't use these totals either, since they still do include
those CPU bound cases that are most likely not representative of next gen console games.

I mean, if you know you're CPU bound and you're easily hitting 100 fps+, there's very little reason to not add more eye candy or up the resolution.

 

[temesgen]

I'm not going to comment on why he didn't link to the more representative graphs, but I wouldn't use these totals either, since they still do include
those CPU bound cases that are most likely not representative of next gen console games.

I mean, if you know you're CPU bound and you're easily hitting 100 fps+, there's very little reason to not add more eye candy or up the resolution.

Because those graphs don't show the power of cloud computing.... duh....

 

[Cyan]

What is the jump in polygons/triangles per second like going from Gen 7 to Gen 8.. is it substantial?

Could it be that modern high quality can draw two triangles per primitive compared to one triangle per primitive from the olden days?:smile2layStation 4 GPU and Xbox One GPU, they don't have a name for now btw, are in that league-

 

[hesido]

I'm personally fully expecting a ~40% increase in frame rate, without even considering bandwidth issues which could make matters even worse (or at best on par) for the Xbox One.

So put in real terms, if the Xbox One is running at 30 fps Vsync (assuming around 40 fps average to actually do that without hiccups), I wouldn't be surprised at all if the PS4 can easily hit ~50 fps on average.

What I've been saying. Tho as others pointed, the Xbone having ESRAM complicates things and makes it harder to draw comparisons from the PC Cards. It may indeed provide massive gains in some areas (e.g. according to sebbbi Xbox's EDRAM saves up to 76% bandwidth in some -although very specific- cases), but whether it will be enough to bridge that gap remains to be seen.

 

[goonergaz]

What I've been saying. Tho as others pointed, the Xbone having ESRAM complicates things and makes it harder to draw comparisons from the PC Cards. It may indeed provide massive gains in some areas (e.g. according to sebbbi Xbox's EDRAM saves up to 76% bandwidth in some -although very specific- cases), but whether it will be enough to bridge that gap remains to be seen.

But are we essentially reversing this gen? So most cross plats will be better on PS4 however exclusives will be able to look as good?

Whilst I'm no tech expert I find it hard to understand how any cross plat games won't be made for PC/PS4 with little effort and 'shoe-horned' into the Xbone (ie less FPS/effects) - after all that seems to be what happened this gen in reverse so why not? I can't see devs wasting too much time and money on the bespoke Xbone ESRAM as they didn't with the PS3 memory/cell setup.

In the end I'm just thinking that Sony went to the devs and asked them what they wanted, Sony (kind of) over delivered with the RAM but gave them exactly what they desired and surely they are the ones who know what's best?

Or am I missing something?

 

[dumbo11]

I'd be surprised if either one were ever in ESRAM.

I've got no idea about HDMI input, but I'd be surprised if the kinect data wasn't present in ESRAM before it gets processed by the GPU. (the depth buffer may only be around 640kb [16bit 640x480?]).

 

[Betanumerical]

What I've been saying. Tho as others pointed, the Xbone having ESRAM complicates things and makes it harder to draw comparisons from the PC Cards. It may indeed provide massive gains in some areas (e.g. according to sebbbi Xbox's EDRAM saves up to 76% bandwidth in some -although very specific- cases), but whether it will be enough to bridge that gap remains to be seen.

Im not sure the eSRAM will provide any bandwidth benefits (outside of its actual bandwidth) to the XBONE in comparison the PS4. Latency sure, but I don't see how having a low latency cache can make more bandwidth then it already has (unless I'm misunderstanding the question).

 

[Gubbi]

I've got no idea about HDMI input, but I'd be surprised if the kinect data wasn't present in ESRAM before it gets processed by the GPU. (the depth buffer may only be around 640kb [16bit 640x480?]).

Why would the GPU work on the Kinect depth buffer? AFAICT they have offloaded all the Kinect chores to dedicated silicon.

Cheers