LiquidRex
Member
He will comment, it's coming.No comments about the L3 cache.
What a.
-
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.
PS5 Die Shot has been revealed
- Thread starter LivingD3AD
- Start date
FatKingBallman
Banned
It does, fps wise. I'm not making stuff up.
So, higher quality shadows and higher image resolution with worse fps is league above PS5?
Last edited:
FatKingBallman
Banned
PS4 Pro is league above X1X just like XSX is league above PS5?
longdi
Banned
no... it's more than that. remember last gen where the cpu was stalled because of jaguar, developers have not need to go beyond a 2500k, 2600k for the most ambitious.
that's part of the excitement this gen, finally we can move up to a full fat 8c/16t zen2 as minimum.
let's just hope sony does another deep dive for their reasons in cutting parts of zen2, or the developers still create their engines to take full use of avx256 without being held back.
J_Gamer.exe
Member
Not necessarily.
Also as a general point not aimed at you, there is much said by Matt about the caches, once coded for helping further boost performance.
Unless he's on about just the clock advantage in gpu side meaning they run faster, but it seems like he means some other optimisation/ tweak. Maybe latency....?
Last edited:
thicc_girls_are_teh_best
Member
Sounds like he's talking about the GPU cache scrubbers in the 2nd tweet; leveraging those would save on cache flushes and therefore cycle time to do the cache flush. I just don't know if they're using patrol scrubbing or demand scrubbing, it has to be either of those two.
Also I might've underplayed a bit on number of game engines optimized for wider GPUs. That said considering PS4 Pro and One X were 36 CU & 40 CU design, if the vast majority of 3P games are using the consoles as base targets and simply scaling up visual fidelity and framerates for more powerful GPU cards, then the engines are probably simply leveraging wider designs for things that scale very easily, like texture resolutions and VFX.
So a lot of game engines are probably compatible with even wider designs, but the root game design governing the use of the engines are probably focused on narrower designs and just scaling up graphics settings and framerates with more powerful and wider GPU designs.
You think a single game dev trying to feed a family's gonna risk their livelihood and potentially freedom against a multi-billion dollar megaconglomerate?
I mean, they could still try, but they better some really strong VPN, use TOR, lots of burner email accounts and code the message. A big problem even with all that though is depending on the level of leaked info the company will have a vague idea of what person or people WOULD have access to that info to potentially let it leak, even if they use a contact as the leaker, so there's a good chance the dev (if they're the leaker) could still be in the crosshairs.
And revealing something like the true makeup of the PS5 APU, is not a small thing that'll wash over a weekend or get ignored by major mainstream sites and channels. It'd be literally everywhere, heck it could probably trend on Twitter
Last edited:
DaGwaphics
Member
Which is a smart move. Would be stupid to compromise the resolution and settings of the whole affair to avoid a few drops on the outskirts of one stage. LOL Plus, it stays well within VRR range and avoids requiring a patch when whatever comes next is released. But that's completely off topic in here.
Kenpachii
Member
Sounds like he's talking about the GPU cache scrubbers in the 2nd tweet; leveraging those would save on cache flushes and therefore cycle time to do the cache flush. I just don't know if they're using patrol scrubbing or demand scrubbing, it has to be either of those two.
Also I might've underplayed a bit on number of game engines optimized for wider GPUs. That said considering PS4 Pro and One X were 36 CU & 40 CU design, if the vast majority of 3P games are using the consoles as base targets and simply scaling up visual fidelity and framerates for more powerful GPU cards, then the engines are probably simply leveraging wider designs for things that scale very easily, like texture resolutions and VFX.
So a lot of game engines are probably compatible with even wider designs, but the root game design governing the use of the engines are probably focused on narrower designs and just scaling up graphics settings and framerates with more powerful and wider GPU designs.
You think a single game dev trying to feed a family's gonna risk their livelihood and potentially freedom against a multi-billion dollar megaconglomerate?
I mean, they could still try, but they better some really strong VPN, use TOR, lots of burner email accounts and code the message. A big problem even with all that though is depending on the level of leaked info the company will have a vague idea of what person or people WOULD have access to that info to potentially let it leak, even if they use a contact as the leaker, so there's a good chance the dev (if they're the leaker) could still be in the crosshairs.
And revealing something like the true makeup of the PS5 APU, is not a small thing that'll wash over a weekend or get ignored by major mainstream sites and channels. It'd be literally everywhere, heck it could probably trend on Twitter
Hey i wouldn't do it don't get me wrong. still its interesting that nobody actually pushes the information out that work with it. Even while it totally makes sense.
FrankWza
Member
These exist? Where?
I hope it’s not next to the sausage factory
Last edited:
iHaunter
Member
Practically no one in this thread knows what they're talking about, including myself. So I'm not going to pretend to.
DaGwaphics
Member
XSX is the system with the higher quality shadows, just sayin. LOL (framerate is identical in 99% of the title, but you knew that)It does, fps wise. I'm not making stuff up.
So, lower resolution shadows and higher image resolution with worse fps is league above PS5?
Last edited:
S0ULZB0URNE
Member
One stage out of six with dips as low as 40fps=1%?
DaGwaphics
Member
It's not one stage, it's a very small area near the outer edge of one stage. And even there the fps is almost always at least in the 50s. Nice effort though.
there is not a single one amd rdna2 6xxxx card with with a narrow design.
The rx 6800 game clock of 1815Ghz and 60 cu's is basically in line with the xsx with it's 1825ghz (constant) and 52cu's
the 6800xt have 72 CU's ....etc etc so no absolutely their design is more in line with what we see with the wider xsx design. Certainly both have the boost clock in the 6800 reaching 2000+ ghz But we've always had reduced clocks on consoles to maintain reasonable temperatures, The classification as not "fast" is more given by the sony counterpart that from the it's own real speed, because it is actually "fast and wide" as every other rdna2 gpu. Imagine the jump in speed considering that ps4 pro had a clock of 911Mhz the One X 1172Ghz while a Vega 56 1471Ghz and the little one GPU rdna1 5xxx starts with a game clock of 1670 Ghz. The truth here is that the ps5 has demonstrated innovation getting very close to the speed of the desktop versions, even though it's too early to assess the real benefits (I would love to see a benchmark between two hypothetical ps5 with and without active variable clock).
But to conclude, all gpu manufacturers prefer the advantage of having many CU's which increase the parallelism.
Last edited:
I hope not really wanting a update on the cpu change it's fascinating. It must be really slick to spend that much time on it. Figured something would of come up on a patent search by the internet detectives.
S0ULZB0URNE
Member
It's as low as 40fps and the % would be much higher than 1%.It's not one stage, it's a very small area near the outer edge of one stage. And even there the fps is almost always at least in the 50s. Nice effort though.
Sir Vilhelm
Member
Mark your words, yes!
TransTrender
Gold Member
Can't believe this thread is 20 pages long.
What on earth is going on in here?
What on earth is going on in here?
ethomaz
Banned
Mystery.
alabtrosMyster
Banned
You guys need to stop this, there is no way to reconcile your views, I tried so hard in the Hitman 3 thread...
Either way, whoever picked the settings for the xbox version decided to sacrifice frame rate for image quality, whoever made the equivalent choice on the PS5 version decided to not make that sacrifice.
So we are left with questions:
- How much should the xbox version sacrifice to lock its frame rate (1900p or 1700p?) and how would they configure those shadows?
- How high the settings of the PS5 version would have to go for it to drop frame like the series x version?
If you can give me a real answer I will concede to you that the series x is in a league of its own compared to the PS5.
Sure, and the scope stopping the game, etc. I mean, you can't tell me how it would run on the PS5 if it had the same settings and resolution as the series x version and don't act like you could.It's not one stage, it's a very small area near the outer edge of one stage. And even there the fps is almost always at least in the 50s. Nice effort though.
Either way, whoever picked the settings for the xbox version decided to sacrifice frame rate for image quality, whoever made the equivalent choice on the PS5 version decided to not make that sacrifice.
So we are left with questions:
- How much should the xbox version sacrifice to lock its frame rate (1900p or 1700p?) and how would they configure those shadows?
- How high the settings of the PS5 version would have to go for it to drop frame like the series x version?
If you can give me a real answer I will concede to you that the series x is in a league of its own compared to the PS5.
It depends how you take your sample, I'm not sure if I would have sacrificed resolution to fix that if... then again, when I'm fiddling with graphics settings on PC drops below 60 become annoying pretty fast. Either way, the drops are pretty important in terms of value, and when they happen it lasts for a while, so they aren't insignificant (unless you are in the 0.001% of people who have a VRR TV, in which case they probably won't annoy you).
Chris_Rivera
Member
DaGwaphics
Member
If that's what helps you sleep at night buddy.
@
alabtrosMyster
pointless whataboutism is just that. It is how it is, and I would take the 4k and improved settings if my tradeoff was some stutter out in the bushes on the periphery, to each their own. Also, find me a quote where I said XSX is in another league in regards to this game. LOL
Last edited:
onesvenus
Member
Yes they maybe reduced the latency between both CCXs. On PS4 the CCXs could read each others L2 cache but it was with much higher latency (190 cycles).
Can any of you provide me with a link that explains how distance between components has latency implications?
From what I remember from my chip design classes at the university, latency on on-chip networks is strictly a function on bandwith, router latency and hops.
anothertech
Member
Xbox dudes still in denial their box is underperforming Sony and cost $100 more.
like the 44 % more pixels denial?
Codeblew
Member
Are you talking about that 1 game out of 50?
I'm talking about what to expect from its consoles compared to what's built into it. You know the engine of a car? here we are in this thread because finally we are starting to discovering what is there, a part from PR and marketing, under the hood of the PS5. And already we have found out that the unified cache of the CPU does not exist and does not exist either the mythical IC on the gpu and from a first sight there's nothing so weird in the layout to resemble something that is different from a rdna1 or rdna2 gpu, so also that rdna3 rumor 99.9% was just false.The software can be fixed, improved, patched but the hardware will always remain the limit,based on what's inside the console, for the achievable performances. And there is practically the whole generation ahead of us to learn how to use those machines
Last edited:
ZehDon
Member
We've read different threads and shop on different planets, it seems.
nowhat
Member
I doubt you know him so well you'd be on first-name basis, so that would be Cerny your words.
Garani
Member
A game which, by the way, doesn't even do RT as much as the PS5, or they would have had to lower resolution and framerate.
Panajev2001a
GAF's Pleasant Genius
With far more distant components you could actually have either more hops in between or since you have more distance to travel memory access split into more pipeline stages (to hit desired clock speed) or both.
In the second tweet yes, but his comment on last generation vs optimised current generation engines (still in dev) is in his first tweet and your remark was about the “narrow and fast” vs “wide and slow” with the former being better for last generation engines and him saying that he believes the opposite to be the case (in many not all cases
):
FranXico
Member
He's talking about Hitman 3, not Control.
Hitman 3 is The One True Hardware Benchmark!
And why? Because it gave the result the experts wanted!
Last edited:
FatKingBallman
Banned
Sorry, higher resolution shadows. Wrote in wrongly and will edit it. But regarding fps, XSX version is worse and you knew that
Garani
Member
Can I go now and play at SpiderMan MM with RT at 60 fps? Thanks!He's talking about Hitman 3, not Control.
Hitman 3 is The One True Hardware Benchmark!
And why? Because if gave the result the experts wanted!
prinz_valium
Member
Jo
onesvenus
Member
Yup, but wouldn't those be visible in the Die shot?
Panajev2001a
GAF's Pleasant Genius
The extra hop maybe (I cannot spot them, but I am not as eagle eyed as other might be), the additional cycles as the access could be broken into to satisfy the increased signal propagation time would be even harder to spot.
J_Gamer.exe
Member
Another interesting set of tweets is this one, open to see all.Sounds like he's talking about the GPU cache scrubbers in the 2nd tweet; leveraging those would save on cache flushes and therefore cycle time to do the cache flush. I just don't know if they're using patrol scrubbing or demand scrubbing, it has to be either of those two.
Also I might've underplayed a bit on number of game engines optimized for wider GPUs. That said considering PS4 Pro and One X were 36 CU & 40 CU design, if the vast majority of 3P games are using the consoles as base targets and simply scaling up visual fidelity and framerates for more powerful GPU cards, then the engines are probably simply leveraging wider designs for things that scale very easily, like texture resolutions and VFX.
So a lot of game engines are probably compatible with even wider designs, but the root game design governing the use of the engines are probably focused on narrower designs and just scaling up graphics settings and framerates with more powerful and wider GPU designs.
You think a single game dev trying to feed a family's gonna risk their livelihood and potentially freedom against a multi-billion dollar megaconglomerate?
I mean, they could still try, but they better some really strong VPN, use TOR, lots of burner email accounts and code the message. A big problem even with all that though is depending on the level of leaked info the company will have a vague idea of what person or people WOULD have access to that info to potentially let it leak, even if they use a contact as the leaker, so there's a good chance the dev (if they're the leaker) could still be in the crosshairs.
And revealing something like the true makeup of the PS5 APU, is not a small thing that'll wash over a weekend or get ignored by major mainstream sites and channels. It'd be literally everywhere, heck it could probably trend on Twitter
So as I see it Sony have faster caches giving a decent boost guaranteed.
20% faster lets say.
The cache scrubbers provide a further boost, how much will differ but he does suggest he thinks it'll matter more than cu count or clocks. It saves bandwidth with no re fetching.
He also mentions the latency of caches and seeing what sony has done with the io breakthroughs, what we can see of the re design of the zen2 cpu in places, its not really a surprise to claim they may have reconfigured some more bits to improve latency, otherwise there's no point modifying and you may as well go stock parts if there's no improvement.
We have the io that won't be a bottleneck and can feed data at unseen of levels and reduce the amount needed to be stored in ram leaving more for whats on screen, not what might be used.
The whole system design is clear, where possible latency is reduced and speed and efficiency is essential.
So all these things add up, the marginal gains I have mentioned before. Some large benefits like the clocks that benefit so much and give ps5 some large advantages. But then add in the ssd and io, add in the smalls gains here and there and all together at times this is allowing ps5 to match series x and even outperform it sometimes.
Last edited:
Garani
Member
Which is what Cerny has been saying all along: you gotta give work to those CUs and it's easier to do it without latencies and with faster clocks. "A rising tide lifts all boats"
This doesn't mean the XSX is a bad console, it's just that Microsoft has decided to be closer to the PC implementation (because of DX12U) instead of being innovative in certain aspects. It's not a bad choise, it's just a different choice.
When I see people crying because the PS5 performs similarly, if not better at times, than a XSX i don't get it: Sony worked A LOT on the I/O assembly, on the GE, the Latencies and clock. All was because they had the vision of pushing data FAST around the system, without wasting cycles. It's a different paradigm, and Sony can do it because they have one monolitic architecture, while MS has that "write once, run everywhere" approach, which by the way is the base of bloated Java software, but I digress.
So, all in all, it's good to see different solutions, it helps the process of pushing ideas around.
Last edited:
ToTTenTranz
Banned
Scaling down is what some devs have been attempting in these latest 9th-gen ports. The result has been the SeriesS version lacking features like raytracing altogether.
When actual 9th-gen engines start appearing that use raytracing for e.g. shadows without a fallback to traditional shadowmaps, how are they going to solve that on the SeriesS? The only way is to develop for the SeriesS first, SeriesX second.
Yes, but that always happens. Cores maintain coherency (i.e. checking each others' calculation results) by snooping through the others' last level cache (LLC). The LLC in Jaguar was the L2 that served 4-core blocks, and in Zen1/2/3 the LLC is the L3.
If the cores couldn't access the instruction + data in each others' LLC, then they'd need to maintain coherency through the RAM, which is very slow in comparison.
The big difference between "unified cache" (Zen3) and two blocks of cache that can be accessed by cores that are external to that complex (Zen2) is the number of cycles needed to access the cache on that "far end" CPU core.
I don't know the numbers so this is just an example, but if in Zen3 it takes 8 cycles to access any part of the unified L3, in Zen2 it takes 8 cycles to access the L3 in its own CCX, but 40 cycles to access the L3 in the other CCX.
This makes a sizeable difference in gaming performance when the GPU isn't a bottleneck (usually high framerates), and it's a good part of the reason why Zen3 is so good for games.
There are two blocks of L3 seen in the PS5's die shot (one for each CCX), but it's also true that both CCXs are close together, and there's a relatively large unidentified space between the CPU and GPU blocks (which could be just glue logic).
It could be that the PS5's custom Zen2 has specific optimizations for reducing the cycles for inter-CCX L3 accesses, to the point where it behaves closer to an unified L3 than a separate one.
Something is responsible for the PS5 getting consistently better performance at framerate levels where the CPU is usually a bottleneck (120FPS modes), despite the SeriesX getting a non-variable higher clockrate and a full fledged Zen2 FPU.
And I don't buy the "current devkits are bad but future ones will bring the magic sauce" theories for the SeriesX. I expect both consoles to mature at parallel levels.
Last edited:
John Wick
Member
Is that why Actual Developers have praised the PS5 design eh?
Andrea Pessino
@AndreaPessino
Dollar bet: within a year from its launch gamers will fully appreciate that the PlayStation 5 is one of the most revolutionary, inspired home consoles ever designed, and will feel silly for having spent energy arguing about "teraflops" and other similarly misunderstood specs.
6:36 PM · Mar 19, 2020
You need to stop your downplaying of the PS5 and move on...........
Clear
CliffyB's Cock Holster
Unless you are exclusive to PS5, you need to have a non-RT fallback.
On PC you need to support older cards, and on Xbox there's series S to think about where although its supported its a very heavy thing in terms of overall budget.
Always been the problem with RT effects, unless you're good with decimating your addressable market, you need to implement legacy solutions as well.
On PC you need to support older cards, and on Xbox there's series S to think about where although its supported its a very heavy thing in terms of overall budget.
Always been the problem with RT effects, unless you're good with decimating your addressable market, you need to implement legacy solutions as well.
prinz_valium
Member
Always a good laugh when this gets posted. Hope we'll still see it in 3 to 5 years. Hopefully even when next gen starts and no one cares about the old consoles anymoreIs that why Actual Developers have praised the PS5 design eh?
Andrea Pessino
@AndreaPessino
Dollar bet: within a year from its launch gamers will fully appreciate that the PlayStation 5 is one of the most revolutionary, inspired home consoles ever designed, and will feel silly for having spent energy arguing about "teraflops" and other similarly misunderstood specs.
6:36 PM · Mar 19, 2020
You need to stop your downplaying of the PS5 and move on...........
PaintTinJr
Member
On the Cache question, last-line of cache(LLC) and L3 caches have been used to mean the same thing AFAIK, and with caches being SRAM, and the IO complex using SRAM (IIRC) and it interfacing directly to the CPU, in this scenario is the IO complex a level 4 cache LLC - or even a L3 extended/shared cache, with each Zen Core pair still having the 4MB setup?
I glanced an extremetech article, today about L3 caches, that said L3 caches began life off-chip on the motherboards, so if the IO complex SRAM is big enough, and on-chip or off-chip then it might be an LLC - in addition to being a monster decompression IO unit, with priority levels.
I glanced an extremetech article, today about L3 caches, that said L3 caches began life off-chip on the motherboards, so if the IO complex SRAM is big enough, and on-chip or off-chip then it might be an LLC - in addition to being a monster decompression IO unit, with priority levels.
Epic Sax CEO
Banned
From the discussions I'm seeing on more technical forums I think that I get what happened.
Both Microsoft and Sony had strong concerns about heat and prevent throttle, each one addressed this issue in different ways. Microsoft's way is more simpler, just put a bigger cooler. Both solutions work but are fundamentally different, Microsoft's way deals with the heat after it's generated, Sony tries to prevent the heat from being generated.
The FPU is the single greatest source of heat, so Sony went with a smaller FPU to have a CPU that generates less heat, uses less power, don't throttle and is easier to cool.
From analyzing the die shots people are concluding that Sony made AMD remove the FADD from the FPU leaving only the FMAC. The FPU can still do FADD math at native 256bits, only on more cycles than the dedicated FADD unit. So the FPU works the same, with a bit less throughput, and do this generating much less heat and using much less power to prevent the CPU from downclocking when stressed. This is really interesting!
Edit: contrary to what some say, the CPU may be better than GPU in doing physics if the right engines are used, because today's CPU have strong vector capabilities. With a 256bits FPU the CPU can calculate the physics of 8 objects with 32bits precision every 5 cycles (the number of cycles the Zen 2's FPU takes to do FADD math at the FMAC, some say).
Edit2: and the SeX would be able to do 8 additional calculations simultaneously every 3 cycles while turning into a blast furnace.
Both Microsoft and Sony had strong concerns about heat and prevent throttle, each one addressed this issue in different ways. Microsoft's way is more simpler, just put a bigger cooler. Both solutions work but are fundamentally different, Microsoft's way deals with the heat after it's generated, Sony tries to prevent the heat from being generated.
The FPU is the single greatest source of heat, so Sony went with a smaller FPU to have a CPU that generates less heat, uses less power, don't throttle and is easier to cool.
From analyzing the die shots people are concluding that Sony made AMD remove the FADD from the FPU leaving only the FMAC. The FPU can still do FADD math at native 256bits, only on more cycles than the dedicated FADD unit. So the FPU works the same, with a bit less throughput, and do this generating much less heat and using much less power to prevent the CPU from downclocking when stressed. This is really interesting!
Edit: contrary to what some say, the CPU may be better than GPU in doing physics if the right engines are used, because today's CPU have strong vector capabilities. With a 256bits FPU the CPU can calculate the physics of 8 objects with 32bits precision every 5 cycles (the number of cycles the Zen 2's FPU takes to do FADD math at the FMAC, some say).
Edit2: and the SeX would be able to do 8 additional calculations simultaneously every 3 cycles while turning into a blast furnace.
Last edited:
azertydu91
Hard to Kill
What I like about this gen is how different both are approaching hardware even though it is based on the same architecture.
This gen isn't about power it is about philosophy, one could end up being better than the other but it is more probable than one philosophy will be better suited for certain games and the other for different games.
This gen comparisons threads will be a bloodbath because I believe the ps5 will sometimes do better and the xsx will sometimes do better and people will argue about power.
alabtrosMyster
Banned
*Dropping frames
40/60 = 0.6666
More or less the same in both directions.
The claim we see all the time is that the series X has a power advantage because of the results in hitman 3... Maybe you didn't claim that the series X was 'in a league of its own', but making any claim in performance based on only the fact that one console has higher resolution and improved shadows without knowing how matching settings for both platforms is not very wise.If that's what helps you sleep at night buddy.
@
It's fair to say that you prefer one over the other, and they are close enough so that the argument can be made either way.
Calling 'whataboutism' when someone asks for hard fact on a technical subject doesn't do anyone a favor.