ChuckeRearmed
Member
This actually has always been bugged me - why did/does on PS the first party games usually look better than on Xbox? Better art style? Design? What is the reason?
Last edited:
-
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.
Xbox Series X’s Advantage Could Lie in Its Machine Learning-Powered Shader Cores, Says Quantic Dream
- Thread starter Lunatic_Gamer
- Start date
- News
This actually has always been bugged me - why did/does on PS the first party games usually look better than on Xbox? Better art style? Design? What is the reason?
rnlval
Member
Sorry, RX 6800 or RTX 3080 is the minimum GPU purchase for me.You are back to double counting things and making it now in a laundry list format (yet more DCU’s feeding off the same L1 does not seem to be a minus for some reasons). You cannot quote more CPU cores and more CPU registers as if they were two separate things
... well unless you are just trying to have a e-peen numbers war and cannot accept that having “just” an 18% advantage is not bad even though it is not the monster advantage you may have wanted.
I'm not going to support RTX 2070 OC to 2.3Ghz argument against my RTX 2080 AIB OC near 2Ghz
assurdum
Banned
The hell ... are you really trying to say all the people need to prize more a 16% of difference in a gpu? Is it a serious claim?
Last edited:
longdi
Banned
If you are talking about PS4 first party games? Because PS4 is more powerful than One.
Power counts, in the right hand.
ChuckeRearmed
Member
No, even on PS3 games look neat despite being weaker than Xbox 360. Multiplatform was better, but that's it.If you are talking about PS4 first party games? Because PS4 is more powerful than One.
Power counts, in the right hand.
geordiemp
Member
Each DCU has a Local Data Share which scales with DCU count. XSX's 26 DCU LDS / PS5's 18 DCU LDS = ~44% advantage for XSX. LOL
More DCU has the following
1. more wave32 processing on the chip instead of outside the chip.
2. more Texture Filter Units
3. more texture load/store units
4. more L1 cache
5. more branch & message unit
6. more scalar units
7. more RT function blocks (RDNA 2)
Trips to the external memory bus have a higher cost.
Also the shared L1 cache between shader arrays is the possible big IPC gain for RDNA2 and frequency gating per WGP which we will soon learn in RDNA2 white paper I guess.
But answer me this, why is every pC card have shader arrays less than 10 CU if sharing a LDS is the way forward. Maybe larger shader arrays will be in RDNA3 and its XSX performance gain ?
/s
Last edited:
ChuckeRearmed
Member
It would be the best-case scenario of just by porting games to PC, you are able to execute them on Xbox. Might bring more games to Xbox from PC.
assurdum
Banned
Gpu was weaker. The whole hardware was more powerful to the 360 but limited with tons of bottlenecks, to be honest
Last edited:
Nhranaghacon
Banned
He is not citing super-resolution in what is specifically quoted - he is specifically citing DirectML -
Direct Machine Learning, or DirectML that is hardcoded into DX12U/Hardware as a type DLSS style feature.
Direct Machine Learning, or DirectML that is hardcoded into DX12U/Hardware as a type DLSS style feature.
rnlval
Member
Each DCU has LDS. 10 CU with RDNA is 5 DCU.
XSS APU's APU chip size is close to Ryzen 7 4800U APU chip size already.
assurdum
Banned
People trolling Sony when said the pro was basically a 8,4TF machine for FP16 shader. Now MS uses the same rethoric to call an unprecedented perfomance on the console. Good Lord PR stuff has no shame.
Last edited:
Andodalf
Banned
Eh, Forza Horizon and Halo 4 both looked really good on 360. I’d only put NDs games as definitely better on that front. Sony had a lot more output than MS did
geordiemp
Member
You did not answer me, RDNA2 chips revealed and leaked all have 10 or less CU (5 DCU) per shader array.
Every one and ps5, except XSX has 7 DCU.
Why ? XSX is the outlier here and you know it both in architecture, layout and clocks
Last edited:
McCheese
Member
Nah, this is "Power of the cloud" bullshit all over again. Machine learning could be used for all sorts of stuff, upscaling graphics, smart dynamic shaders, even increased framerate via smart interpolation.
But you shouldn't wave machine learning abstract concepts about as a real hardware advantage until something is on the market that is actually using it.
Similar to the cloud bullshit that sounded great on paper, we could go the whole generation before they've figured out an actual use case for it.
But you shouldn't wave machine learning abstract concepts about as a real hardware advantage until something is on the market that is actually using it.
Similar to the cloud bullshit that sounded great on paper, we could go the whole generation before they've figured out an actual use case for it.
Last edited:
Lysandros
Member
Both PS5 and XSX have 2 shader engines and 4 shader arrays. There is one L1 cache block (128 KB in RDNA1) per array shared between 13 CU's for XSX and 9 CU's for PS5. L1 cache amount isn't directly tied to CU count, how do you know about its size for XSX, since when this info is public?Each DCU has a Local Data Share which scales with DCU count. XSX's 26 DCU LDS / PS5's 18 DCU LDS = ~44% advantage for XSX. LOL
More DCU has the following
1. more wave32 processing on the chip instead of outside the chip.
2. more Texture Filter Units
3. more texture load/store units
4. more L1 cache
5. more branch & message unit
6. more scalar units
7. more RT function blocks (RDNA 2)
Trips to the external memory bus have a higher cost.
Last edited:
Panajev2001a
GAF's Pleasant Genius
I think he meant L1 which seems to be the same across both consoles get it is feeding more DCU’s (L1 is partitioned per Shader Array and the XSX Shader Arrays house more DCU’s, so you have a higher DCU per L1 ratio).
longdi
Banned
360 has better GPU and overall better design
PS3 has the more flexible CPU but held back by the last minute dated GPU that nvidia dumped onto them. Needless to say, Sony became friends with Amd onwards.
So multiplatform games looked better on 360. While first party games, PGR and Gears look just as good. Halo ran better than KZ. MS then winded down their first party titles at the end of 360 in favour of kinnect. So we didnt get to see much XGS games, especially TPS games that are more suitable to showcase the visuals
except MS did not count this hadware ML into the TF wars?
assurdum
Banned
SSD faster (and I/O of course) means potentially the gpu on ps5 needs less RAM or bandwidth for the same tasks , it's not just faster loading time. People continue to downplay this aspect but there is a reason if amd did infinite cache on new Navi.
Last edited:
01011001
Banned
Halo 4 arguably looks better than any PS3 exclusive and it runs better too with a mostly stable 30fps while games like TLoU regularly drop to 20fps in any demanding scene. also absolutely terrible AA
Last edited:
assurdum
Banned
What? 24 TF machine for ML they said. It's the same shit again360 has better GPU and overall better design
PS3 has the more flexible CPU but held back by the last minute dated GPU that nvidia dumped onto them. Needless to say, Sony became friends with Amd onwards.
So multiplatform games looked better on 360. While first party games, PGR and Gears look just as good. Halo ran better than KZ. MS then winded down their first party titles at the end of 360 in favour of kinnect. So we didnt get to see much XGS games, especially TPS games that are more suitable to showcase the visuals
except MS did not count this hadware ML into the TF wars?
Last edited:
oldergamer
Member
That number was when using the dedicated hardware for raytracing setup.
longdi
Banned
That was talking about hardware raytracing, which supposedly can run parallel with traditional rasterization.
El Pistolero
Banned
Those words are from XSXS lead engineer Andrew Goosen. What is wrong with them?
Nothing wrong with them, they are just designations for rendering techniques, for the most part. The problem resides with fans that start throwing them willy nilly as if they were easily activated options, usable at will with complete disregard to the projects end goals, the overall power envelope of the machine and the obvious tradeoffs that plague development in general. Plus, Sony has their own version of such features, so it remains to be seen which approach ends up on top, especially that APIs will play a big role in extracting the performance out of the two boxes. In other words, wait and see...
Last edited:
rnlval
Member
PS4 Pro's 8.4 TFLOPS FP16 spec is fine when coupled with higher memory bandwidth, but Sony gimped PS4 Pro APU's potential.
geordiemp
Member
I put a big blue square around what all other parts are doing, XSX is an enlarged / stretched shader array and is different.
The L2 is just scaled by GDDR6 PHY.
Is a larger shader array a benefit or not to efficiency, we shall see. I believe the 10 CU limit on many parts is tied to the shorter data path required for higher RDNA2 frequences and propagation.
But we will see soon enough.
jigglet
Banned
I think ML will only work if they're delivered as API's devs can just use out of the box, like DLSS. Not many - if any - game devs are going to attempt to train ML models themselves. It's a very specific skill set.
Remember they tried to do something similar with the cloud on XB1 and it was never adopted. At the time they just said "use the cloud for expanded computation...but how you ask? Na, you write that shit yourself from scratch, have fun k thx bye lol!!"
Remember they tried to do something similar with the cloud on XB1 and it was never adopted. At the time they just said "use the cloud for expanded computation...but how you ask? Na, you write that shit yourself from scratch, have fun k thx bye lol!!"
Last edited:
assurdum
Banned
And Sony said was for FP16 shader. What's the difference exactly?That was talking about hardware raytracing, which supposedly can run parallel with traditional rasterization.
Andodalf
Banned
Because implying you’re using fp 16 for all your rasterization is a meme. It has uses, but they’re limited. Using lower precision for ML is an actual thing.
El Pistolero
Banned
That was talking about hardware raytracing, which supposedly can run parallel with traditional rasterization.
24 TF is a bogus number. If you run Raytracing, you will automatically take up resources that could otherwise have been dedicated to traditional rasterization, forfeiting other graphical features in the process. Believing otherwise is positing that MS was somehow able to produce an alien technology powered by the clouds...
Last edited:
longdi
Banned
Thats what i said. The 2nd gen games from XSXS will be surprising. The advancements in the Series X continue to impress.
btw, when MS is being open and excited to share the designs they took in XSXS = pr buzzword?
Does that mean Nvidia, Amd, Apple etc are guilty of the same thing? Only those who prefer to keep secretive are those we should praise and respkt?
Last edited:
Nhranaghacon
Banned
Not sure how DirectML, a feature of DX12Ultimate to which all PC hardware is now geared for - and something with dedicated Hardware on the Series X is rhetoric - utilizing lower FP operation's for machine learning isn't just talk according to Nvidia DLSS. The software's ability to improve performance at up to 86% is very real when sampling from all other Machine Learning Examples - how Microsoft uses this with SeriesX is anyone's guess but I expect a future announcement from Microsoft boasting over a 65% increase in performance with DirectML programming properly implemented.
Last edited:
assurdum
Banned
They never implied anything.
assurdum
Banned
I would be very careful with such claims. Nvidia has dedicated core hardware for such stuff. Serie X hasn't. You can't count with the same gain boost on serie X just because they have software stuff in the horizon. The propaganda around this stuff by MS blinds the rationality of many people to my opinion.
Last edited:
Krisprolls
Banned
There are multiple reasons :
- More talented people including artists so that means better art style yes
- More money / more time to improve and polish the games
- Better custom API / software and hardware optimizations (Sony has a dedicated cross team helping studios named ICE Team)
- It only has to work on a single machine they know perfectly, not hundreds of unknown PC configurations, so optimization is a lot easier.
Last edited:
El Pistolero
Banned
Thats what i said. The 2nd gen games from XSXS will be surprising. The advancements in the Series X continue to impress.
The 2nd gen games of any console that has ever hit the market tend to be surprising, and more impressive than the first wave. What advancements are you talking about when the damn machine is not out yet? Time to get off that DeLorean, my friend...
Gen.Grievous
Member
It took Nvidia a lot of time to get DLSS implemented properly and with a lot of dedicated hardware (tensor cores). RDNA2 doesn't have that much dedicated hardware and I don't have as much faith in Microsoft's ability in this area compared to Nvidia.
Pretty obvious XBsX has a hardware advantage though, why some people try to downplay that amazes me.
Pretty obvious XBsX has a hardware advantage though, why some people try to downplay that amazes me.
Last edited:
Lysandros
Member
Yes, if the L1 cache size remains the same as in RDNA1 this would mean 40% more readily available L1 data (per CU) to be send to L0/CU's at 20% higher bandwidth for PS5. That would mean 'something' for CU efficiency i guess..
rnlval
Member
More DCU enables more wavefronts to run on-chip. External memory has a higher latency cost.
RX 5700 XT's L1 cache has 16-way IO ports and it's unknown for XSX's L1 cache design.
Last edited:
Panajev2001a
GAF's Pleasant Genius
I put a big blue square around what all other parts are doing, XSX is an enlarged / stretched shader array and is different.
The L2 is just scaled by GDDR6 PHY.
Is a larger shader array a benefit or not to efficiency, we shall see. I believe the 10 CU limit on many parts is tied to the shorter data path required for higher RDNA2 frequences and propagation.
But we will see soon enough.
I was agreeing with you, but you were quoting the LDS which is inside the DCU and the L1 cache which is shared by all DCU’s in a Shader Array.
Panajev2001a
GAF's Pleasant Genius
More DCU enables more wavefronts to run on-chip. External memory has a higher latency cost.
RX 6800 has 128 KB L1 per shader array, hence four shader array yields 512 KB L1 cache which the same as RX 5700 XT's 512 KB L1 cache.
AMD Radeon RX 6800 XT Specs
AMD Navi 21, 2250 MHz, 4608 Cores, 288 TMUs, 128 ROPs, 16384 MB GDDR6, 2000 MHz, 256 bitwww.techpowerup.com
RX 6800 (60 CU) crushed RX 5700 XT(40 CU).
Where would you run more wavefront offchip?
again double counting things... you have the same L1 cache feeding 7 DCU’s on one side and 5 DCU’s on the other... period.You want more TFLOPS (more wavefronts/threads/operations) without additional Shader Engines (which has a bigger fixed cost HW wise... sure, but you have a tradeoff that the L1 cache is shared with more DCU’s.
Last edited:
Nhranaghacon
Banned
I fully expect a 86% jump in performance, as it plainly states the Series X can opt to utilize a portion of it's core hardware to ML if it so desires - considering all other ML examples cite extraordinary gains, I expect no less from Microsoft's own proprietary solution built into DX12Ultimate/SeriesX in the future.
If we are being pretentious, the DRS solution on One X was not a true Machine Learning solution and offered increases upwards of 46-65% performance improvement's through non-ML specific software optimizations.
Last edited:
geordiemp
Member
The 60 CU part is an 80 CU part with SE disabled. Its all 80 CU die design.
However, the part your missing is the L1 cache on RDNA2 will likely not be same as RDNA1. The recent AMD IPC patent is about shared L1 between shader arrays, but was not detailed in the Big navi reveal.
Hence why there is 10 CU per shader array on big Navi and ps5.
assurdum
Banned
You are lucky if you have around a 20-30% of gain in performance. The percentage expected to you are absurd and you are fallen in the trap of the MS propaganda.
Last edited:
Panajev2001a
GAF's Pleasant Genius
XSX has no special ML HW, “just” can process INT8/INT4 code at a faster rate than FP16 code.
assurdum
Banned
Wut. Sony did a vague statement about TF, MS did the same. To be honest lately MS seems overhyping tech stuff on serie X more than the reality. Never heard something comparable on ps5 in the Sony front. But I could be wrong or missed it.
Last edited:
Nhranaghacon
Banned
Nope, AI enhanced software is something I for the last 20 year's have stated will garner performance up to 85% on this gen of hardware in fact - so not a MS propaganda thing more a Computer Scientist rationale.
assurdum
Banned
You need dedicate hardware to support such enhancement. You take a bunch of tech stuff heard here and there, and you come with overoptimistic conclusion mixing all in the bag without rationality. It doesn't work like that.
Last edited:
too much eslim
Member
Both have geometry engine