Mihael Mello Keehl
Banned
Herewego
-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
EDGE: "Power struggle: the real differences between PS4 and Xbox One performance"
- Thread starter Musiol
- Start date
Here we go again.
Balance! Would love to have seen the meeting where they decided that 'balance' would be the response/buzz word from now on.
SenjutsuSage
Banned
I think it's literally impossible to say this with any sort of certainty. We don't even know where about 4 to 5MB of the supposed 47MB worth of on-chip storage Microsoft says is built into the Xbox One SoC. Where that 4 to 5MB, or potentially more if I'm mistaking the numbers, went could be fairly important. Just saying.
Mmmmm, the 204GB/s number rears its head again. DIdn't Penello "correct" it to 218GB/s in one of these threads?
Gemüsepizza
Member
Well Kaz Hirai himself said they are using off the shelf parts:
http://www.psxextreme.com/ps4-news/590.html
Edit: and yes - I've seen the second part of the sentence.
Sony designed the CELL with IBM and Toshiba, Kaz meant that PS4 APU is built by AMD, just like X1. but his statement doesn't mean that Sony didn't add some customizations at all.
all your questions are answered in Mark Cerny Gamasutra interview, but people tend to ignore what they don't like. tidbits from Cerny interview :
Source:
http://www.gamasutra.com/view/feature/191007/inside_the_playstation_4_with_mark_.php?page=2
Theres redundancy built into a number of parts, including the 2 deactivated CU's in the GPU that would contribute to this, don't think they have magic hiding, but don't stop believing
.brain_stew
Member
You can't leave off:
4) 16 ROPs vs. 32 ROPs
5) 72 TMUs vs. 48 TMUs
Bandwidth, CUs, ROPs and TMUs are the fundamental building blocks of any GPU and different games and different scenarios within those games will stretch each area differently. You can't be dramatically slower in every single important area of GPU performance and expect to compete. You can try and offset this difference in the future with a more forward looking design that is better setup to take advantage of new rendering techniques and algorithms but to make matters worse it's Sony that have the advantage here with a GPU more tailored to GPU compute.
The difference will be obvious on day one and the gap will only become wider as the generation progresses.
Mihael Mello Keehl
Banned
Dat balance! Whoever makes the thread put balance in there.
ConteZero76
Member
Yeah, he's 100% fair.
Obvious dGPU NDA in place~
SenjutsuSage
Banned
Theres redundancy built into a number of parts, including the 2 deactivated CU's in the GPU that would contribute to this, don't think they have magic hiding, but don't stop believing
They don't count redundancies as part of their chip design when mentioning on chip storage. Sure, t hey likely did that for the "5 billion transistors," but I certainly don't think they would for on-chip storage. That's real storage that's being used for whatever it's being used for inside the system. I'm obviously in no position to say what that is.
The 47MB of on-chip storage isn't redundancy. They wouldn't have mentioned it in such a way if it was. There's likely redundant CUs on the GPU that aren't used. No, not CUs that will magically be enabled one day, just backup to protect against chip failure, and I'm sure other things, but I wouldn't have any idea what those are, since this is an example that lines up well with what was done on the xbox 360. There were more unified shader pipelines on the 360 GPU than what was mentioned, but those were disabled for redundancy reasons, but Microsoft didn't magically include them as part of the overall chip design, which makes me doubt they would do such a thing with on-chip storage, if some of that were redundant and unusable by the system. Also, And, the LZ compression/decompression and JPEG decode isn't ONLY for Kinect, but nice try trying to make it appear that way.
They're boxing with shadows here.
I'm seeing them construct arguments against the design that they can tear down that people weren't actually fronting?
The whole 'you can't add bandwidth' criticism was not about concurrent access to the memory pools. I'm not sure who was claiming you couldn't concurrently access eSRAM and DDR3?
The argument that eSRAM is all fine and dandy from a development ease perspective because 360 was the easier console to dev for in the last cycle is also ignoring the fact that there are simpler designs around today and criticism levelled at it around development ease are in that context.
Going by more recent comments from Cerny they're also misinterpreting/twisting the VG Leaks information about Sony's design...
Their talk about Move Engines is also, again, focussing on how it helps in the context of XB1's design while making it sound like they're an advantage in general.
My final comment would be that it's a little unfortunate Richard Leadbetter was the channel for this. It does little to dispel the notion that he's been a bit too close to MS sources of late. That there are echoes of themes that he communicated in previous articles seems rather coincidental too.
I'm seeing them construct arguments against the design that they can tear down that people weren't actually fronting?
The whole 'you can't add bandwidth' criticism was not about concurrent access to the memory pools. I'm not sure who was claiming you couldn't concurrently access eSRAM and DDR3?
The argument that eSRAM is all fine and dandy from a development ease perspective because 360 was the easier console to dev for in the last cycle is also ignoring the fact that there are simpler designs around today and criticism levelled at it around development ease are in that context.
Going by more recent comments from Cerny they're also misinterpreting/twisting the VG Leaks information about Sony's design...
Their talk about Move Engines is also, again, focussing on how it helps in the context of XB1's design while making it sound like they're an advantage in general.
My final comment would be that it's a little unfortunate Richard Leadbetter was the channel for this. It does little to dispel the notion that he's been a bit too close to MS sources of late. That there are echoes of themes that he communicated in previous articles seems rather coincidental too.
Did MS engineer just admit that GDDR5 is the best memory setup?
also when asked about the CU number compared to PS4, he brought 12 CUs are the most balance shit, then claim that Sony agree with him in VGleaks article? What about Cerny who declined such rumor?
How in hell do you guys want me to believe anything MS says, seriously guys!!
shinra-bansho
Member
So much balance. The word is in the article like 15 times. I also like how they've included a video of their representative Xbox One Radeon 7850 and PS4 7870 XT...
Is the implication supposed to be that the PS4 is imbalanced?
Edit: They seemed to have missed people's contention with the bandwidth adding. On the matter of the ESRAM, people weren't just taking issue with whether one could add bandwidth, they were saying it was disingenuous to add the bandwidth when it only applies to a very small part of the RAM, compared to the unified pool on the PS4.
On the idea that there are bottlenecks with regard to CU scaling, they seem to be ignoring that other components of the PS4's GPU hold even more of an advantage than the 40% the CUs do?
I don't know why they're talking about the old VGLeaks document when it's well known now that there is no division between the compute units and developers can used them for whatever they see fit.
Is the implication supposed to be that the PS4 is imbalanced?
Edit: They seemed to have missed people's contention with the bandwidth adding. On the matter of the ESRAM, people weren't just taking issue with whether one could add bandwidth, they were saying it was disingenuous to add the bandwidth when it only applies to a very small part of the RAM, compared to the unified pool on the PS4.
On the idea that there are bottlenecks with regard to CU scaling, they seem to be ignoring that other components of the PS4's GPU hold even more of an advantage than the 40% the CUs do?
I don't know why they're talking about the old VGLeaks document when it's well known now that there is no division between the compute units and developers can used them for whatever they see fit.
They don't count redundancies as part of their chip design when mentioning on chip storage. Sure, t hey likely did that for the "5 billion transistors," but I certainly don't think they would for on-chip storage. That's real storage that's being used for whatever it's being used for inside the system. I'm obviously in no position to say what that is.
The 47MB of on-chip storage isn't redundancy. They wouldn't have mentioned it in such a way if it was. There's likely redundant CUs on the GPU that aren't used. No, not CUs that will magically be enabled one day, just backup to protect against chip failure, and I'm sure other things, but I wouldn't have any idea what those are, since this is an example that lines up well with what was done on the xbox 360. There were more unified shader pipelines on the 360 GPU than what was mentioned, but those were disabled for redundancy reasons, but Microsoft didn't magically include them as part of the overall chip design, which makes me doubt they would do such a thing with on-chip storage, if some of that were redundant and unusable by the system. Also, And, the LZ compression/decompression and JPEG decode isn't ONLY for Kinect, but nice try trying to make it appear that way.
Really don't count redundancies, how do you know?.
Also, JPEG, not used in games, seriously google game texture formats and learn why your idea is a horrible one, you'd blow out all your caches, its primary purpose per the interview is for Kinect it might have another use somewhere in the realm of decoding JPEGs from external devices but other then that..
The interview also practically confirms that its HD7790 also.
Funny thing that they are quoting VGleaks article that was debunked by Cerny himself ! that's so funny it's sad actually.
SenjutsuSage
Banned
I see this all the time, and I think it's totally bogus. Richard Leadbetter has written and is involved in so many, in my view, awesome articles that have been put up on that site, a lot of them very awesome articles (some of the best actually) on the PS4, but nobody gives him any credit when he does those or accuses him of somehow being too close to Sony or Sony devs when he writes those? I say whatever war people are waging with the guy is beyond ridiculous. The guy isn't allowed to report on the Xbox One or relay information on it without people attacking his credibility, but he can report plenty of times on the PS4, often with very positive and very insightful information that we weren't aware of before, without a negative word uttered anywhere?
Again, this ridiculous proxy fight that some on GAF keep manufacturing about Leadbetter's motivations and allegiances is tired, played out, and, above all, extremely pitiful and transparent. Some of the articles that people most cite on here for some of the great insight we've gotten on the PS4 have been written by Richard Leadbetter, or haven't people actually realized this fact? Do people's eyes only turn red with anger when he's reporting anything that could be construed as mildly positive Xbox One information?
You need to read more on what JPEG decoding is doing on vgleaks. It's converting the jpegs into different pieces of information that apparently can be used in games.
New secret sauce: MS will unlock CUs in future firmware update.
No they won't because according to MS 12 is more BALANCED than 14
B_Boss
Member
The most interesting aspect concerning this home console 'power struggle' between MS & Sony seems so obvious to me: Mark Cerny lol....this guy is either one of the most arrogant human beings on earth or he is on to something...and with a personality and background such as his I'd probably assume well that when he speaks about Computer & Game Design he knows exactly what he is talking about.
So lets skip alot of the rumors, hearsay, etc. What has Cerny said? Lets take a look (emphasis mine)..
I think its very clear what he is implying and saying in more ways than one here:
Perhaps many details can be discussed, argued, etc, but I think the 'bigger picture' has been clearly seen and understood by Cerny himself. The man has called the PS4 the most powerful console ever made, why are we still arguing such a moot point now lol? (rhetorical). Cerny 'dropped the mic' a long time ago.
(source: http://www.theverge.com/2013/8/22/4646696/ps4-mark-cerny-interview-gamescom-2013)
So lets skip alot of the rumors, hearsay, etc. What has Cerny said? Lets take a look (emphasis mine)..
I think its very clear what he is implying and saying in more ways than one here:
Perhaps many details can be discussed, argued, etc, but I think the 'bigger picture' has been clearly seen and understood by Cerny himself. The man has called the PS4 the most powerful console ever made, why are we still arguing such a moot point now lol? (rhetorical). Cerny 'dropped the mic' a long time ago.
(source: http://www.theverge.com/2013/8/22/4646696/ps4-mark-cerny-interview-gamescom-2013)
Don't forget about:
1. 768 Shader Cores vs. 1152 Shader Cores
2. 16 ROPs (Render Output Pipelines) vs. 32 ROPs
3. 12.8 Gigapixels/sec vs. 25.6 Gigapixels/sec
Brad Grenz
Member
Yeah, the FUD strategy seems to be repeating the word "balance" as many times as possible to imply the PS4 is unbalanced.
Yeah, cute that they omitted the cost part of that particular cost/benefit analysis.
Anyway, they should probably retitled the article. This isn't DF vs Xbox One Architecture, it's Xbox One Architecture PR via DF.
Yet your still not listening, its the most retarded format on the planet to use for textures or anything game related.
1. The GPU doesn't support it internally in its caches and TMU's so you get a huge blow out on the cache and on the eSRAM as it has to be decoded.
2. Its meant for photos not game textures or game data, which is why its not used.
3. To even use it in RGB space you have to convert it via the shaders.
4. I can't think of a single reason to use them, can you give me one?.
i HeaR MusiC
Member
nvm
Deep dives on individual games are a little different to competitive hardware comparisons.
And I'm just talking about the latest run of articles. There's just some fairly unquestioning acceptance of what he's presented with. Maybe Richard isn't in a position to challenge what they're saying, maybe he's better at benchmarks than understanding hardware - he's excellent at benchmarks. He may not have the technical confidence to challenge some of the more slippery/questionable suggestions made by MS here. And it's not about giving companies/developers a platform to share views, but his own commentary and who/what it aligns with perhaps a little too easily. It may not be wilful bias, but his critics will probably only find fuel in stuff like this.
SenjutsuSage
Banned
This part I personally think is just blatant common sense, and people have been fighting this basic notion forever on here, suggesting what is and isn't possible, and here it is straight from the horses mouth restating what other much less technically proficient posters have been saying on here forever. ESRAM is but a basic extension of EDRAM on the Xbox 360, and just like they designed EDRAM to work well with system memory on the Xbox 360, they did the same yet again with the Xbox One, only there are less limitations compared to eDRAM.
But, you know, these guys don't know what they're talking about and are just spreading PR lies.
Nobody complains when he writes fantastic articles on the PS4, also largely accepting what he is told with fairly unquestioning acceptance. This fantastic article about the crew port to PS4 was also written by Richard Leadbetter, and many more articles that I'm feeling too lazy to bother go looking up.
http://www.eurogamer.net/articles/digitalfoundry-how-the-crew-was-ported-to-playstation-4
Where's the controversy about his motives when he does things like this? And these literally aren't the only fantastic articles this guy has on or relating to the PS4. Anyway, that's enough about this. You guys can carry on. Forgot how hardcore this thread was.
But, you know, these guys don't know what they're talking about and are just spreading PR lies.
Nobody complains when he writes fantastic articles on the PS4, also largely accepting what he is told with fairly unquestioning acceptance. This fantastic article about the crew port to PS4 was also written by Richard Leadbetter, and many more articles that I'm feeling too lazy to bother go looking up.
http://www.eurogamer.net/articles/digitalfoundry-how-the-crew-was-ported-to-playstation-4
Where's the controversy about his motives when he does things like this? And these literally aren't the only fantastic articles this guy has on or relating to the PS4. Anyway, that's enough about this. You guys can carry on. Forgot how hardcore this thread was.
I'm inclined to believe the 14CUs with 16ROPs would still be a more balanced GPU, just look at the 7790. I dont recall seeing it ROP limited in any of the benches as far as I can remember, please correct me if I'm wrong. Yeah yeah I know I'm not an engineer whose hands are literally dripping with the silicon logic of the Xbone APU so what do I know!
brain_stew
Member
Sound like they confirmed that they chose DDR3 + eSRAM for cost, "power draw" and memory capacity reasons as we all expected. If even Microsoft can't pretend it was chosen as superior to GDDR5 then that about closes that silly argument. It's a band aid to hit 8GB at low cost, nothing more, nothing less.
Edit: Nice of them to admit they are fillrate limited as well (explanation of why they get more performance from increasing clocks rather than enabling CUs). So much for people saying the ROP advantage didn't matter if it's already proving to be a major bottleneck in launch software.
Now that I've actually finished reading it, there's a lot of interesting tidbits in there. It still leaves me with a bunch of questions though.
This confirms that the ESRAM is absolutely not a cache but a developer-controlled scratchpad, which renders the Intel 32MiB EDRAM cache size arguments irrelevant.
This still doesn't really explain it to me. Unless there's some special purpose hardware there that can do a read-modify-write on the ESRAM side the of the 1024-bit bus, there's just no way to send more than 1024-bits per cycle, which tops out at 109GB/s at 853MHz.
This is really telling. This is an admission that the CUs can (and do) starve due to deficiencies elsewhere in the GPU pipeline. It's too bad that the bottlenecks aren't actually elaborated.
The GPR comment is pretty weird. GPRs allocation for shaders is done during compilation of the shader, and the only real rule is to minimize them without having to resort to spilling to some sort of scratch buffer. What he could mean, instead of allocation, is just filling them in the first place (loading uniforms and the like). That's a sort of difficult problem, but one I would hope that they have a handle on.
The GCN DMA engines are part of what they're calling a Data Move Engine? Is the Data Move Engine really just a bunch of other bits of fixed function that they've grouped together even though they're actually separate?
Here comes latency. Are talking about the GDDR5 vs. DDR3 latency, or ESRAM vs. main memory, or something else entirely?
This confirms that the ESRAM is absolutely not a cache but a developer-controlled scratchpad, which renders the Intel 32MiB EDRAM cache size arguments irrelevant.
This still doesn't really explain it to me. Unless there's some special purpose hardware there that can do a read-modify-write on the ESRAM side the of the 1024-bit bus, there's just no way to send more than 1024-bits per cycle, which tops out at 109GB/s at 853MHz.
This is really telling. This is an admission that the CUs can (and do) starve due to deficiencies elsewhere in the GPU pipeline. It's too bad that the bottlenecks aren't actually elaborated.
The GPR comment is pretty weird. GPRs allocation for shaders is done during compilation of the shader, and the only real rule is to minimize them without having to resort to spilling to some sort of scratch buffer. What he could mean, instead of allocation, is just filling them in the first place (loading uniforms and the like). That's a sort of difficult problem, but one I would hope that they have a handle on.
The GCN DMA engines are part of what they're calling a Data Move Engine? Is the Data Move Engine really just a bunch of other bits of fixed function that they've grouped together even though they're actually separate?
Here comes latency. Are talking about the GDDR5 vs. DDR3 latency, or ESRAM vs. main memory, or something else entirely?
No one is lying but they're arguing with a point I'm not sure (m)any were making, at least within more informed critiques of the architecture?
The argument about adding bandwidths together and the appropriateness of doing that is not about concurrent memory access as they're framing it. Adding bandwidths together wasn't appropriate on 360 either.
The reason it is not appropriate is that x amount of bandwidth to one small pool of memory + y amount of bandwidth to one larger pool of memory is not equivalent to x+y amount of bandwidth to one large pool. You don't have the same flexibility about how to use that bandwidth as in the latter case. In some use cases it may make little or no difference but in others, and thus in the general case, it does make a difference. Adding together bandwidths and presenting them as one number in order to compare to one batch of bandwidth to one pool of memory runs the risk of suggesting they're equivalent when they're not. That's all.
Gemüsepizza
Member
Jesus Christ, but it's not a simple evolution of the Xbox 360 design. Just look at this comparison:
Current gen
PS3: ~25GB/s
Xbox 360: ~23GB/s + eDRAM
Next gen:
PS4: 176GB/s
Xbox One: 68GB/s + eSRAM
Do you notice something?
The eSRAM is made from 4x 256bit memory controllers which each access 8 1MB modules.
You can read and write to separate controllers and therefore separate modules at the same time, in fact you have to get there peak bandwidth numbers but separately reading or writing will always max out at 109GB/s if your doing it over all 4 controllers.
SenjutsuSage
Banned
Gemüsepizza;83107225 said:
Yea, I notice something alright. You're playing the traditional console warrior card of not even mentioning the very real bandwidth that is on the ESRAM. Nice try leaving out the bandwidth figure on the 360's eDRAM also in an attempt to try and cover what your real intentions were for leaving out the embedded memory bandwidth on the Xbox One.
Yea, I notice something alright. You're playing the traditional console warrior card of not even mentioning the very real bandwidth that is on the ESRAM. Nice try leaving out the bandwidth figure on the 360's eDRAM also in an attempt to try and cover what your real intentions were for leaving out the embedded memory bandwidth on the Xbox One.
Ill fix it for him then.
brain_stew
Member
Perhaps he didn't have the opportunity to question MS. Perhaps if he took a harder approach he wouldn't get as much information as he does. Either way, I'm pleased that he gets this information out there, we can question it and/or rip it to shreds from there.
Sure. But 4x 256-bits is 1024-bits, and 1024-bits * 853MHz is 109GB/s. Even if you can choose for each individual lane to read or write, each cycle you only have enough lanes for 1024 bits.
Gemüsepizza
Member
Yea, I notice something alright. You're playing the traditional console warrior card of not even mentioning the very real bandwidth that is on the ESRAM. Nice try leaving out the bandwidth figure on the 360's eDRAM also in an attempt to try and cover what your real intentions were for leaving out the embedded memory bandwidth on the Xbox One.
My point is, on Xbox 360 the eDRAM was a bonus for developers. On Xbox One, the eSRAM is needed to counter the massive bandwidth difference of the main memory.
Sword Of Doom
Member
[Sums up my perspective as well, and on a much more broad level than just performance. Seemed MS believed that either Sony would be totally asleep at the wheel or that they themselves could throw out whatever device they wanted and the masses would lap it up thanks to the Xbox brand. I wonder if higher-ups at MS thought they had a chance at becoming the Apple of the living room and failed to notice they totally missed their mark./quote]
This!
Edit: Messed up the quote
Similar threads
- Locked