How the PlayStation 4 is better than a PC

[3DVagabond]

Games will not use 8 cores, at best they will use 7 since 1 core is dedicated to the OS. Just like 1GB memory is reserved as well.

You are trying to tell me there is some fairy dust. I tell you there is reality.

PS4 demos capped at 30FPS already due to limitations.

I never said the games would use all 8 cores, just that they are designed to run on that specific 8 core APU. That's not fairy dust.

I guess the memory is weak when trying to remember the failed hype rumours.

http://www.pcgameshardware.de/Ninte...e-Vielleicht-doch-mit-Intels-Larrabee-702080/
http://gonintendo.com/?p=108637
http://forums.gametrailers.com/viewtopic.php?f=38&t=987238

And so on and on.

There will be no APU in PS4 and Xbox720.

Oh wait, you people talk about the Audio Processing Unit? Yes it will most likely be AMD based

You're just mad because there's an AMD APU in the latest consoles because you were so vocal that there wouldn't be. There's reality for you.

 

[raghu78]

The PS4 will have a 1.6Ghz core clock. And that memory will have no effect on it. Its just for the GPU part. Lastly there is not so much OS difference as you wish to believe. It might end up to 5% or less in real world.

Its getting very clear that the nextgen consoles are discount solutions for yesterday, rather than tomorrow.

please get the facts right. the PS4 SOC has an integrated 256 bit GDDR5 memory controller which serves both the GPU and CPU.

http://www.gamasutra.com/view/feature/191007/

"The CPU and GPU are on a "very large single custom chip" created by AMD for Sony. "The eight Jaguar cores, the GPU and a large number of other units are all on the same die," said Cerny. The memory is not on the chip, however. Via a 256-bit bus, it communicates with the shared pool of ram at 176 GB per second."

http://www.techpowerup.com/180765/amd-working-on-stripped-down-playstation-4-soc-for-pcs.html

"The PlayStation 4 SoC uses a 256-bit wide GDDR5 memory interface, with a stellar memory bandwidth of 176 GB/s"

http://www.extremetech.com/gaming/1...u-memory-should-appear-in-kaveri-xbox-720-ps4

even better is the architetctural integration of a unified address space and fully coherent memory between CPU and GPU.

 

[ShintaiDK]

please get the facts right. the PS4 SOC has an integrated 256 bit GDDR5 memory controller which serves both the GPU and CPU.

http://www.gamasutra.com/view/feature/191007/

"The CPU and GPU are on a "very large single custom chip" created by AMD for Sony. "The eight Jaguar cores, the GPU and a large number of other units are all on the same die," said Cerny. The memory is not on the chip, however. Via a 256-bit bus, it communicates with the shared pool of ram at 176 GB per second."

http://www.techpowerup.com/180765/amd-working-on-stripped-down-playstation-4-soc-for-pcs.html

"The PlayStation 4 SoC uses a 256-bit wide GDDR5 memory interface, with a stellar memory bandwidth of 176 GB/s"

http://www.extremetech.com/gaming/1...u-memory-should-appear-in-kaveri-xbox-720-ps4

even better is the architetctural integration of a unified address space and fully coherent memory between CPU and GPU.

I think you missed the point entirely. Try look at different memory speeds and amounts for current highend CPUs and look on the difference it makes. If a 5Ghz i7 3770K got 25GB/sec or 250GB/sec doesnt matter at all, simply because its not a limitation.

The PS4 CPU is just extremely weak compared to any regular CPU we use today in normal laptops/desktops.

 

[ShintaiDK]

I never said the games would use all 8 cores, just that they are designed to run on that specific 8 core APU. That's not fairy dust.

And that will make all the difference in the world, right? :whiste:

 

[3DVagabond]

I think you missed the point entirely. Try look at different memory speeds and amounts for current highend CPUs and look on the difference it makes. If a 5Ghz i7 3770K got 25GB/sec or 250GB/sec doesnt matter at all, simply because its not a limitation.

The PS4 CPU is just extremely weak compared to any regular CPU we use today in normal laptops/desktops.

Look at what happens with AMD's APU's, not i7's, as that's what's in the PS4.

And that will make all the difference in the world, right? :whiste:

Of course it will.

 

[ShintaiDK]

Look at what happens with AMD's APU's, not i7's, as that's what's in the PS4.

Of course it will.

How does AMDs APUs CPU performance scale with memory speed again?

 

[raghu78]

I think you missed the point entirely. Try look at different memory speeds and amounts for current highend CPUs and look on the difference it makes. If a 5Ghz i7 3770K got 25GB/sec or 250GB/sec doesnt matter at all, simply because its not a limitation. The PS4 CPU is just extremely weak compared to any regular CPU we use today in normal laptops/desktops.

you quoted Temash performance which has a very low bandwidth single channel DDR3 1600 memory controller and substantially lower clocks. in that context i mentioned improved bandwidth and much higher clocks will significantly improve performance. also the fact that quad core Temash matches core i3 in multi threaded performance at same clocks for lower power means given a fixed power and die size budget the other options would have been even worse.

How does AMDs APUs CPU performance scale with memory speed again?

AMD small cores like Jaguar do not have L3 cache like core i5 or core i3. So definitely performance will improve when moving from low bandwidth single channel DDR3 memory to high bandwidth GDDR5 memory especially on demanding workloads like games.

 

[galego]

This should only be for single-precision operations. Ivy Bridge can do 8 single-precision MULs and 8 ADDs per cycle per core.

Really, the only reason Jaguar is on the PS4 is it's the best that AMD can do.

Disagree and my question went in the opposite side. Why did Epic selected an Intel i7? I offered a tentative explanation using the info available to me then. The PS4 CPU gives a raw GFLOP comparable to i5/i7, but the i5 lacks hyper-treading and PS4 games will be using above four cores (demos are already using more).

Meanwhile other posters have provided more relevant info. Seeing a 2.0 GHz 4x1 jaguar beating a 2.5 GHz 2x2 SB i3 explains why a 1.6 GHz 4x2 jaguar chip on a console will be competing with a 3.5 GHz 4x2 SB/IB i7. Now makes more sense why Epic selected an i7.

 

[PrincessFrosty]

If it tests that way, how is it not? IIRC, Geforces were faster with compression than without by the NV20 (GF 3), though definitely were by the NV40 (GF 6xxx, benchmarks below).

Because dedicating specific parts of the pipeline to certain (even common) tasks, doesn't give you that processing for free, it's sacrificing more general purpose computing resources for fixed function. You're usually using some part of the GPU silicon to achieve this.

Obviously fixed function hardware is faster at the task but you're making a fixed sacrifice out of your available silicon to get it, which otherwise could be spent on more general compute or something else.

Benchmarks don't really show this, well in fact they do. You need to ask the reverse question, such as "when I disable this specific feature in my software/game which uses fixed function hardware why is the overall frame rate not increasing?"

The lesson to learn here is really that it depends on what kind of work you're going to be doing, the massive problem is that console have fixed hardware for 7+ years and developers make games spec'd out for the lowest common denominator...well most developers did for the lifespan of the last generation, only a few made any extra effort for the PC in the latter years of that cycle.

So things like heavy trade offs for processing vs memory space make sense for consoles with only 256mb of memory, but do not make sense for modern PC's which have more memory than we really know what to do with.

 

[Enigmoid]

Disagree and my question went in the opposite side. Why did Epic selected an Intel i7?

Probably mainly for appearance reasons. To look the part of a top end pc (they want to show that UE4 is killer and cool and running it on a gtx 460 + i3 doesn't really show that).

Tech demos are never cpu intensive (well very rarely) compared to actual games. (They can cut out a lot of cpu heavy stuff such as AI, all character computations because there is no character onscreen). The graphical tests of 3dmark for example put little stress on the cpu. (I don't know if this is true in this case but it wouldn't be surprising if it was).

Look at BF3 singleplayer on ultra. Require very little cpu. Its too early to say that they ran it on an i7 because an i7 was required.

 

[beginner99]

So definitely performance will not improve when moving from low bandwidth, low latency single channel DDR3 memory to high bandwidth, high latency GDDR5 memory especially on demanding workloads like games.

fixed

 

[showb1z]

Disagree and my question went in the opposite side. Why did Epic selected an Intel i7? I offered a tentative explanation using the info available to me then. The PS4 CPU gives a raw GFLOP comparable to i5/i7, but the i5 lacks hyper-treading and PS4 games will be using above four cores (demos are already using more).

Meanwhile other posters have provided more relevant info. Seeing a 2.0 GHz 4x1 jaguar beating a 2.5 GHz 2x2 SB i3 explains why a 1.6 GHz 4x2 jaguar chip on a console will be competing with a 3.5 GHz 4x2 SB/IB i7. Now makes more sense why Epic selected an i7.

So not only is the ps4's gpu equal to 2 Titans, the tablet cpu in it rivals a 3770...
Don't let the lack of facts hold you back.

 

[Cerb]

Because dedicating specific parts of the pipeline to certain (even common) tasks, doesn't give you that processing for free, it's sacrificing more general purpose computing resources for fixed function. You're usually using some part of the GPU silicon to achieve this.

Obviously fixed function hardware is faster at the task but you're making a fixed sacrifice out of your available silicon to get it, which otherwise could be spent on more general compute or something else.

Benchmarks don't really show this, well in fact they do. You need to ask the reverse question, such as "when I disable this specific feature in my software/game which uses fixed function hardware why is the overall frame rate not increasing?"

The lesson to learn here is really that it depends on what kind of work you're going to be doing, the massive problem is that console have fixed hardware for 7+ years and developers make games spec'd out for the lowest common denominator...well most developers did for the lifespan of the last generation, only a few made any extra effort for the PC in the latter years of that cycle.

So things like heavy trade offs for processing vs memory space make sense for consoles with only 256mb of memory, but do not make sense for modern PC's which have more memory than we really know what to do with.

Unless you have a Titan, you generally don't have more memory than you know what to do with (and, with SLI/Xfire, 3-6GB is not wasted).

Most users have 1GB, though 2GB is becoming the norm. Without compression, that 1GB of 4GHz GDDR5 would be like having 500-700MB at 2-3GHz. I have been able to use more than I have for awhile now, I'm just waiting until a full system upgrade to bother (still using a C2D). 1GB VRAM is only enough for most games today because it's the 2nd tier lowest common denominator (current consoles being the true LCD). Remove compression, and that 1GB wouldn't be enough, for textures of the same size. It would also put more pressure on RAM bandwidth, which is something we usually do have in mild excess (using compression).

So, really, it's not even free, it's more like the other way around: lack of compressed textures is expensive, because it's going to mean needing to pay for denser or more RAM chips to keep the GPU busy, at any given quality level.

 

[Olikan]

After looking at the Temash review of Jaguar based cores. I would say my faith in any kind of performance for the PS4 is severely lowered. And I didnt have much hopes in it to begin with.

And thats a 1Ghz quadcore Jaguar.

No wonder we already see a 30FPS capped demo for the PS4.

well, PS4 cpu is almost 3 times faster than that... :whiste:
even with 1 core to the OS

 

[Erenhardt]

Not true. Texture decompression is free on any modern GPU as it's built right into the hardware, not software. In fact, using uncompressed textures might actually be slower since they use more memory bandwidth.

That being said, using compressed textures on a modern game engine is pretty much mandatory if you want to stay competitive since the storage gains approach 10:1 and memory space is limited.

False. Texture decompression uses GPU compute. It is executed on shaders. The main reason textures are compressed is hdd space memory capacity and bandwidth.
Nice read about civ 5 :
http://www.csee.umbc.edu/~olano/papers/texcompress.pdf

GPUs have become more and more effective at hiding memory latency with threads, making the bandwidth savings of DXT less important for many applications. For example, for the game shaders in Sid Meier’s Civilization
R V, rendering with DXT vs. uncompressed textures did not make any noticeable performance difference, counter to the folk wisdom that DXT is necessary for game rendering bandwidth.
In addition, DXT artifacts are becoming less acceptable as game quality expectations grow. Memory and disk savings are still a key reason to continue using texture compression.
Most games use more texture than fits into memory. To swap working sets, they pause with “loading” screens, or initially load low resolution textures, so the player sees the textures “res in” as they are playing. In a game like Civilization V, the player can switch at any time between the game screen and any of the world leaders with no warning. In this context, “loading” screens are unacceptable.

There is no dedicated part to texture decompression:

Some hardware extensions have been proposed that would allow more flexible on the fly compression. Inada and McCool [IM06] proposed B-tree indexing hardware to support random access within a variable bit rate compressed texture, and proposed a general configurable filtering unit. Our method decompresses variable bit rate textures using standard GPU computing.

Decompression is done in many small blocks using advatage of parallel computing architecture offered by GPUs

To achieve good GPU decoding speed, we divide the image into blocks, with GPU threads decoding the difference values for every block in every MIP level simultaneously. This introduces a tradeoff between compression rate and GPU occupancy. Smaller blocks increases the number of threads, increasing occupancy, but each block introduces overhead that reduces the compression rate.

Decompression can be redirected to CPU in case if GPU is too slow and can't spare any resources on decompression.

For low-end hardware, textures are decompressed on the CPU at a reduced resolution and paged into the GPU after the leader scene is already playing. For DX11-class hardware, all leader textures are kept resident on the GPU in compressed form and decompressed when needed.

Compressed textures give huge memory space and bandwidth savings for "little"(depends on compression ratio\algorithm) cost of GPU power.

The total set of textures is almost 2 GB. Even in DXT form, the full set of textures is almost half a gigabyte. In VBR compressed form, these 18 leaders only take 142 MB, so we can keep all of them resident on the GPU. Decompressing the texture set for any leader takes from about 50 to 150 ms. If we include the additional expansion leaders (22 in all), the total is 2.25 GB of uncompressed texture, but only 180 MB with VBR compression.

While small compression seems to be always beneficial, higher compression ratios, often froced by the lack of memory capacity and/or bandwidth, are bad, resulting in lower texture quality and more GPU ultilization.
Here is a nice graph that explains alot:

You can see how increasing compression ratio by 1.0 (from 2.5 to around 3.0~3.5) increases compute time by 10 times (from 10^1 ms to 10^2 ms)

PS4 having lots of memory with nice bandwidth will enable high quality textures without need of high compression ratio, resulting in less work for GPU.

In linked article, on the last page, there are graphs showing compression ratio vs quality. Compression ratio bellow ~3.0 shows no (or close to none) image quality loss/artifacts.
You want your compression to be where it is fast enough to load compressed texture to memory, doesn't take too much space (both in HDD and memory), is fast to decompress and don't use too much shader power.

And here is nice table with texture loading times.

 

[R0H1T]

well, PS4 cpu is almost 3 times faster than that... :whiste:
even with 1 core to the OS

I would put it as 2~3 times more efficient but then again since almost everything is different, except x86 based parts, between a PC & a competing console there is no practical way to verify this claim !

Also why does this thread have to go on like this, the OP has been taken out of context & has been misinterpreted by anyone & everyone who wants to post a contrarian viewpoint

 

[Erenhardt]

Originally Posted by raghu78
So definitely performance will not improve when moving from low bandwidth, low latency single channel DDR3 memory to high bandwidth, high latency GDDR5 memory especially on demanding workloads like games.

fixed

Edited:
Ouch! You are talking about CPU performance, not GPU

About this "High latency":
http://en.wikipedia.org/wiki/GDDR5

GDDR5 operates with two different clock types. A differential command clock (CK) as a reference for address and command inputs, and a forwarded differential write clock (WCK) as a reference for data reads and writes. Being more precise, the GDDR5 SGRAM uses two write clocks, each of them assigned to two bytes. The WCK runs at twice the CK frequency. Taking a GDDR5 with 5 Gbit/s data rate per pin as an example, the CK clock runs with 1.25 GHz and WCK with 2.5 GHz. The CK and WCK clocks will be aligned during the initialization and training sequence. This alignment allows read and write access with minimum latency.

 

[frozentundra123456]

I would put it as 2~3 times more efficient but then again since almost everything is different, except x86 based parts, between a PC & a competing console there is no practical way to verify this claim !

Also why does this thread have to go on like this, the OP has been taken out of context & has been misinterpreted by anyone & everyone who wants to post a contrarian viewpoint

I agree the thread should have ended long ago. I do disagree that the OP has been misquoted and taken out of context. In contrast, the thread has continued endlessly because the OP takes without question statements from developers with a vested interest and no hard data to back them up, and instead of being willing to wait for hard data, endlessly repeats the same undocumented claims over and over again as if they were proven facts.

 

[Cerb]

False. Texture decompression uses GPU compute. It is executed on shaders. The main reason textures are compressed is hdd space memory capacity and bandwidth.
Nice read about civ 5 :
http://www.csee.umbc.edu/~olano/papers/texcompress.pdf

Where in that paper does it show how DXT1 to DXT5, FI, are being done?

They're talking about specifically adding a new texture compression method, using DirectCompute, much like Id's Megatexture will do with CUDA, to improve the quality v. size.

Those proposed hardware extensions they're talking about are for a flexible decompressor pipeline, which is not what DXTs, to date, involve. They are fixed size BTC style encodings.
http://msdn.microsoft.com/en-us/library/bb694531(v=vs.85).aspx

 

[R0H1T]

I agree the thread should have ended long ago. I do disagree that the OP has been misquoted and taken out of context. In contrast, the thread has continued endlessly because the OP takes without question statements from developers with a vested interest and no hard data to back them up, and instead of being willing to wait for hard data, endlessly repeats the same undocumented claims over and over again as if they were proven facts.

By OP I meant opening/original post btw are we talking about galego or the actual thread starter Cloudfire777

 

[Arzachel]

Lots o' stuff

If decompresing textures on the fly takes less power than moving the full sized ones around, then it is the way to go except in extreme corner cases, where the bus gets saturated or you absolutely can't do without that tiny bit of additional shading grunt. The actual advantage from more RAM is that not having to constantly shuffle data around means better performance and most likely less code complexity.

 

[BallaTheFeared]

AMD has the worst IMC in the industry, GDDR5 isn't going to help their cpu at all in fact if the latency is too high it will probably hurt it.

 

[PrincessFrosty]

Unless you have a Titan, you generally don't have more memory than you know what to do with (and, with SLI/Xfire, 3-6GB is not wasted).

Most users have 1GB, though 2GB is becoming the norm.

Going by something like the steam stats the majorety of gamers are using 1Gb or more GPUs and have 4Gb+ of system RAM.

However the average screen resolution is 1080p or below.

The overwhelming majority of AAA games from the last 7 years have been more or less straight console ports, even with a 2560x1600 screen I've managed to keep my GTX580 for what will soon be 2 generations with very little reason to upgrade because the number of games that can truly stress even hardware this old can be counted on 1, maybe 2 hands.

The simple fact is that the fixed hardware for the consoles has influenced the PC space with almost 100% penetration, and PC hardware is now in another league above these systems resources are plenty. Can you throw some arbitrarily high amount of SSAA or tesellation into a scene to stress these cards? Sure, but outside of a few minor exceptions this is simply not the reality of the situation.

When the new generation of consoles lands and the lowest common denominator gets increased in power I'm sure we'll see multi-platform games that look much better and stress modern hardware, again that isn't a reality right now.

 

[Olikan]

I would put it as 2~3 times more efficient but then again since almost everything is different, except x86 based parts, between a PC & a competing console there is no practical way to verify this claim

errr...PS4 cpu is 3 times faster than the temash reviewed....
just by looking @ clocks and cores
sure efficiency is a big win for the console's cpus too, not having to deal with a stack of software layers is very good

 

[Cerb]

Going by something like the steam stats the majorety of gamers are using 1Gb or more GPUs and have 4Gb+ of system RAM.

However the average screen resolution is 1080p or below.

The overwhelming majority of AAA games from the last 7 years have been more or less straight console ports, even with a 2560x1600 screen I've managed to keep my GTX580 for what will soon be 2 generations with very little reason to upgrade because the number of games that can truly stress even hardware this old can be counted on 1, maybe 2 hands.

The simple fact is that the fixed hardware for the consoles has influenced the PC space with almost 100% penetration, and PC hardware is now in another league above these systems resources are plenty. Can you throw some arbitrarily high amount of SSAA or tesellation into a scene to stress these cards? Sure, but outside of a few minor exceptions this is simply not the reality of the situation.

When the new generation of consoles lands and the lowest common denominator gets increased in power I'm sure we'll see multi-platform games that look much better and stress modern hardware, again that isn't a reality right now.

Except that such a statement is only true for un-modifiable AAA titles. Off the top of my head, I Crysis, Crysis 2, Crysis 3, BF3, Oblivion, Fallout 3, Fallout: New Vegas, Skyrim, and GTA 4 remove limitations due to suffering from consolitis (though Crysis 2 was quite the release botch, IIRC). With all compressed textures (I haven't seen uncompressed for anything but HUDs in a long time), 1GB VRAM often requires tweaking to stay under, and I'm not even up to 1080p, as of yet (I'm getting weak pixels, though, so I feel a forklift upgrade coming on ). For Oblivion, FO3, and F:NV, you don't even need mods to do it, just INI tweaks, so it shows more, and reduces LOD less aggressively.

Now, if all the textures took 2x-6x the VRAM, 2GB would be about as bad as 1GB (the higher-ratio formats suck for normals, so aren't as common, today).

I can even make use of my 1GB in Morrowind, though I admit about 800MB is the highest I've gotten it to use, even with MSGO.