-
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.
Rumor: PS5 Pro Codenamed "Trinity" targeting Late 2024 (some alleged specs leaked)
- Thread starter MisterXDTV
- Start date
- Rumor Platform
THE:MILKMAN
Member
Seems to be mixing up GDDR6 bandwidth with DDR4. Looks like he is picking bits of info from different insiders...
TheRedRiders
Member
I wouldn't bother speculating on Tidux's leaks, he doens't know crap.
Also funny how every insider all of sudden has detailed specifications on things like memory bandwidth and teraflops only after Tom Henderson's leak, but before that they didn't know jack.
Also funny how every insider all of sudden has detailed specifications on things like memory bandwidth and teraflops only after Tom Henderson's leak, but before that they didn't know jack.
Mr.Phoenix
Member
First off...24TF is not only...
second, that tweet doesn't make sense to me. I can't think of a RAM config that gets you those numbers.
Bry0
Member
Yup. Is that 24 calculated with dual issue or not? Sounds like he pulled numbers off a forum. Lol
Don't expect more than 16MB L3 total so V-cache definitely won't happen. L3 cache for the GPU is a maybe/probably, it seems most likely they'd just use the 4x MCD configuration from Navi 32 which this GPU will lift basically everything from.
Last edited:
Tqaulity
Member
These are close to the core rumors leaks but seem a bit optimistic. 20 GB of RAM and 24 TFLOPs? Bring it on! This would be awesome and on the upper end of what Sony could do for a reasonable price by next year.
Ok yes it's silly to speculate over speculation I get it. But IF these specs turned out to be true, let's think about how you could get to 24TFLOPs with only a 54 CU part. The rumored 7700XT with 54 CUs clocked at 2.6Ghz boost is only ~18 TFLOPs. That's already a pretty heft clock increase over the base PS5. How do they get another 33% TFLOP increase with that config? Possibly an "astronomical" clock speed (i.e >3Ghz)? Maybe the bits from RDNA 4 provides some IPC gains not seen with RDNA 3? Or maybe they use a full 60 CU part (even then, would still need a ~3Ghz clock to hit 24TFLOPs). Makes you wonder doesn't it
At the end of the day, I would love 24TFLOPs however they get there. Make it happen Sony!
I thought that guy was known for just making up stuff?
James Sawyer Ford
Member
24 TF at 60 CUs???
It's going to be a ~35-40 TFLOPS part if you present it as AMD does other quad-issuing RDNA3 GPUs. 24TF is not happening either way as it would either have low 1.75GHz clock or astronomically high 3.5GHz (which is well in excess of what any current overclocked 5nm GPU can do on air cooling).
That's a 60 CU clocked at 3.2 or slightly under GHz, which translates into 48 or so RDNA3 TFLOPS.
Also,why not use 20GBs of GDDR6 instead of a split like the Series X?
The specs have to be BS, 18GBs GDDR6 can't get 448 gbs on any bus width.
Last edited:
Loxus
Member
He's still using RDNA2 to calculate TFlops.
24,000,000 ÷ 60 ÷ 32 ÷ 2 ÷ 2 = 3,125GHz
3,125GHz is too high for a console.
Mr.Phoenix
Member
Don't read into that guy... that shit reeks of BS to me. Seem like he is just trying to jump on the leaks bandwagon for attention.These are close to the core rumors leaks but seem a bit optimistic. 20 GB of RAM and 24 TFLOPs? Bring it on! This would be awesome and on the upper end of what Sony could do for a reasonable price by next year.
Ok yes it's silly to speculate over speculation I get it. But IF these specs turned out to be true, let's think about how you could get to 24TFLOPs with only a 54 CU part. The rumored 7700XT with 54 CUs clocked at 2.6Ghz boost is only ~18 TFLOPs. That's already a pretty heft clock increase over the base PS5. How do they get another 33% TFLOP increase with that config? Possibly an "astronomical" clock speed (i.e >3Ghz)? Maybe the bits from RDNA 4 provides some IPC gains not seen with RDNA 3? Or maybe they use a full 60 CU part (even then, would still need a ~3Ghz clock to hit 24TFLOPs). Makes you wonder doesn't it
At the end of the day, I would love 24TFLOPs however they get there. Make it happen Sony!
60CU means 54CU active. Realistically speaking. Getting 54CU to 24TF? That's a 3.5Gh clock. Nope... BS. And even if it was somehow the full 60CU being used (which makes no sense but hey, we're speculating right?) that's still a 3.1Ghz clock. Nope... BS again.
And that RAM. That shit makes zero sense. And no one EVER talks about how much RAM is being reserved for the OS. And why would the OS RAm suddenly increase? Is he saying that the 4GB is a separate pool of RAM? O, then how does it have 448GB/s of bandwidth? All these things are just not possible.
HeisenbergFX4
Gold Member
lh032
I cry about Xbox and hate PlayStation.
lol we are still talking about TF?
Crispy Gamer
Member
Gpu would have to be about 3ghz clocked that sounds too dreamy
StereoVsn
Gold Member
Oh damn, hello fellow 40s crew! And yeah, if this thing is not going to go ballistic with the disk spinning up every 20 min I am pre-ordering.
Although to be fair my PS4 Pro sounded like jet taking off. Couldn't sell it fast enough to get a PS5, lol.
jm89
Member
Yes, if this dude told me the earth was spherical i'd believe it was flat.
Last edited:
StereoVsn
Gold Member
Yeah, considering you can get a good PCIE Gen 4 4TB for line $250-270, no point of going over 1TB built in. Save on coats toward other components.
CanYouFlyBobby?
Banned
4nm is quite surprising
TheRedRiders
Member
Refreshing as well, there's around a 20% increase in power efficiency over N5, I think the Pro will be around 350-400w total system power consumption.
Bry0
Member
That would be either a big or loud console. I would expect active power use in gaming to be like 250w maybe 300w absolute max.
TheRedRiders
Member
Agree, I think average power consumption will be around that ball park, PS5 has a total power draw of 350w but on operation it's around 200w and fluctuates.
Mr.Phoenix
Member
Yup.. its no mystery actually. We already have detailed specs of the GPU this is going to be based on.
That GPU is already rated to be a 200W GPU. On 5nm. And if its on 4nm? Then it wud be drawing even less power than that. And has a game and boost clock of 2400 and 2600 respectively. PS5pro can sit in the middle, so... 2500Mhz? And will obviously not have the massive 64MB infinity cache that GPU has. You throw in the CPU, I can see the PS5pro APU pulling no more than 230w under load. The system would still need a 300w+ PSU though.
Dream-Knife
Banned
Can you really compare an APU to a dGPU?
PS5 has the same "hardware" as the 6700, but has the performance of a 6600xt due to lower clocks.
DuduIsDisappointed
Banned
Insane Metal
Member
24TF? Lol. No. I bet it'll be<20.
Mr.Phoenix
Member
Yes, we can compare it. Its not going to be identical... but that is the best GPU to use to compare or extrapolate on what the PS5pro could be. Especially if you know what the architectural carryovers are. And if we end up being off on just clocks... thats still pretty damn close.
| 7700XT | PS5pro | |
| Fab Node | 5nm | 4nm/5nm |
| Compute Units | 54 | 54 |
| Clock (game/boost) | 2400Mhz/2600Mhz | 2500Mhz |
| FP32 (RDNA3 32) | 17.9TF (35.9TF) | 17.2TF (34.5TF) |
| Cache (l2/l3) | 2MB/64MB | 6MB/ - |
| TDP | 200W | 200W+ (CPU + GPU) |
| Mem Bus | 192bit | 256bit |
ChiefDada
Member
When has the 6600xt ever matched a PS5?
Dream-Knife
Banned
Returnal runs on high 1440p 60 on a 6600xt. Ps5 it runs at medium 1080p 60.
ChiefDada
Member
Returnal stands alone as the weirdest PS5 to PC port I've ever seen. At 1440p the bandwidth starved 6600xt with 8gb buffer should not be running faster than PS5. Simply put the game is not well optimized for PS5. Housemarque should have been patched it. It especially stings because it's my favorite game on PS5.
SegaSnatcher
Member
I just laugh that anyone would question Mark Cerny at this point. PS4, PS4 Pro and PS5 all have punched above their weight.
Look at the gaming experiences the PS4/PS4 Pro offered even with the crap tier CPU. Mark Cerny clearly knows how to design hardware and does a great job engineering the architecture to limit certain bottlenecks.
Go build a PC with a similar setup of the PS4/PS4 Pro and see how it handles the same games.
PS5 Pro will take a lot of those games with unlocked 40fps modes above the 48fps VRR threshold making for a much better experience.
Look at the gaming experiences the PS4/PS4 Pro offered even with the crap tier CPU. Mark Cerny clearly knows how to design hardware and does a great job engineering the architecture to limit certain bottlenecks.
Go build a PC with a similar setup of the PS4/PS4 Pro and see how it handles the same games.
PS5 Pro will take a lot of those games with unlocked 40fps modes above the 48fps VRR threshold making for a much better experience.
Last edited:
Brigandier
Member
Honestly if it is to be 24tf I'd be extremely happy lol...
I'm a bit out the loop with GPU CPU tech these days but is the latest RDNA potentially being used in PS5Pro? How does that double it's TF output? I know I could Google it but what would be the point of discussion!!!
You think Pro will use a newer CPU???
SegaSnatcher
Member
17.2 is pretty close to my expectation of 18Tflops, so I guess I'd only be mildly dissapointed if they couldn't reach 18.
Trust me, he knows absolutely zero. Always just combining various rumors.
Mr.Phoenix
Member
Yes. That is about the only thing to be certain of. At worst, PS5pro will be using RDNA3, which is already out now, so a safer bet will be to say it will use RDNA3+. Basically something in between 3 and 4.
Since you say you are out of the loop, I wi give a simple explanation of TF doubling in RDNA3. In RDNA3, each compute unit can process two instructions instead of one. But only if the GPU driver can sort out which instructions best fit into this get-two-per-clock thing. That is the most basic way I can explain it.
A slightly more detailed (and complex way), each CU has 64 shaders aka. Alu`s. The SIMD units that cater to those 64 shaders have the ability to handle two instructions simultaneously. The problem is that its still based on the og CU foundation. So you have hardware that can handle two instructions per clock now but with the underpinnings of something that was designed to handle one per clock. This means that at best, in real-world performance, you will never get 2x the TF number with RDNA3. Maybe you et like 60-70% more, but not 100% more. Right now, however, you would be hard-pressed o even see ike 20-30% improvements, that is because the ability to dual issue is currently very driver dependent. The SIMD has to basically be good at finding instances that would favor dual issue utilization. And AMD is not known for its great drivers.
But it's something that would ikey increase over time, or even better, something that would be easier to implement in a console.
Not sure about this one. I am sure we all would love it to use Zen 4 and all that. But in truth, I personally see them doing no more than increasing the CPU clock and increasing the CPU cache. Its lack of cache is the reason its underperforming. If they were to throw in Zen 4 cores with its current cache structure,it would be just as bad as it is now.
SegaSnatcher
Member
Naa, they will just use the smaller die shrink to slightly boost clock speeds. The Zen 2 core is fine, but it wouldn't hurt to boost the clock speed to 3.8Ghz or higher.
Brigandier
Member
Thanks for that I understood you clearly
sounds like it has huge potential if Cerny and other engineers nail it which I think they will.Do you think the Zen 2 CPU with a bigger cache / clock speed will be competitive even in late 2024?
Last but not least what is Cernys secret sauce with the RT patent should we be expecting huge improvements to RT capability or not to get carried away?.
Sounds like the PS5 Pro has big potential to be a worthy mid gen refresh!!!
Physiognomonics
Member
As PS5 runs at fixed 1080p and mostly locked 60fps we have no idea how high it could run.
I think it should have been pretty self explanatory that I was judging sony first parties efforts so far this gen, in the future they might drop some gems but looking at spiderman2 yeah lets keep our expectations grounded. Compared to prior gens thir output has been underwhelming to say the least, we are half way into next gen and so far there has been nada in terms of next gen showcases really. Everyone is more than happy to stay safe and stagnant in the cross gen bubble slightly upgrading their lastgen pipelines and calling it a day. Ratchet is nice and great in terms of io usage but in terms of visuals its a good starting point but not like the technical toure de forces we had come to expect in the halfway points of prior gens.
We had killzone2, uncharted 2, gow3 etc and so many jawdroppers on ps3 by that point and even on the ps4 which was a relatively conservative upgrade we had a easily distinguashable next gen output within a few years of the gen meanwhile look at the shit we had so far on the ps5.
Last edited:
TheRedRiders
Member
You're saying the dual-issue compute system is driver dependent? this would mean it requires no input from developers/programmers.
I'd be curious to see what kind of implications this would have on console, as the hardware is fixed as well as the drivers.
Returnal is an outlier and in almost all other benchmarks the gimped bandwidth makes the 6600xt fall behind the ps5 and most other competitors. The card is terribly designed.
Mr.Phoenix
Member
Yup. It woudbe more than adequte. Especially when you consider that all these consoles are expected to do is peak at 120fps when possible.Thanks for that I understood you clearly
Do you think the Zen 2 CPU with a bigger cache / clock speed will be competitive even in late 2024?
To put it into perspective, the PC equivalent of the PS5 CPU is the Ryzen 7 3700x. That CPU paired with a 2080ti can run Fortnite in 1080p at over 210fps. Metro [email protected]. And the only thing that is different from that CPU to that in the PS5, is the PS5s is clocked lower, and has only like I think 8MB of cache vs the 32MB in PC version of the chip.
No, you can get carried away. It however is not that groundbreaking, which is a good thing because it means it's more realistic.
Cerny's patent basically would make an AMD GPU handle RT in the exact same way Nvidia and IntelGPUs handle RT. Currently, AMD GPUs use shaders (CUs) to handle BVH traversal. Whereas, Nvidia and ntelGPUs have BVH traversal handled by the RT cores too. That is why AMD GPUs are crap at RT. This patent basically catches AMD up. Some context of how excited you should at least be?
Take Hogwarts. An extreme case, but shows how messed up it is.
No RT. 1440p.
7900XTX = 112fps
4080 = 109fps
3070 = 54fps
Intel A770 = 42fps
RT Ultra. 1440p
4080 = 52fps
3070 = 24fps
Intel a770 = 20.5fps
7900xtx = 15fps
Yup, this is why we need BVH acceleration. RT at ultra cuts the performance of every other GPU by nearly exactly half... but it completely decimates that of the AMD GPU. Dropping performance by over 80%.
Yes. It's driver dependent. Devs could obviously help by using specific kinda code, which makes it easier for the driver to identify what would work with dual issue and what wouldn't. But it's primarily driver dependent. And yes, that's great for console development. Would be easier to implement in consoles than say in PCs.
As it stands, RDNA3 dual issue requires the compiler to find instances where this thing works. And how well the compiler can do that is driver based. But even if the compiler is working at its peak, and the drivers are perfect, unless certain things are changed in RDNA3+ and above, dual issue compute is never going to really give you 100% compute improvement. But more like 50-70%..... at times.... if you are lucky... maybe.
Last edited:
Loxus
Member
A more extreme case, path tracing.
AMD RT performance is not bad at all, they're just one gen behind Nvidia.
4080 - 87 fps
3080TI - 55 fps
3080 - 51 fps
7900XTX - 50 fps
PS5 Pro should be using RDNA4 RT. So, that Sony RT patent is more likely to be a API/software patent to utilize RDNA4 RT, since Sony doesn't used DirectX Ray Tracing.
When it come to RDNA4 RT, AMD has a RT patent which features a Traversal Engine. I expect this to be featured in RDNA4.
GRAPHICS PROCESSING UNIT TRAVERSAL ENGINE
This is what AMD lacked in comparison to Nvidia's RT implementation.
In terms of Dual-issue, I wouldn't base that on what's happening on PC to what will happen on consoles.
Tqaulity
Member
No you're right there. This is MUCH slower than previous gens and we haven't had that significant showpiece title. That's more of a reflection of the state of game development today and how difficult (and time consuming) it is. Obviously Sony isn't alone here. Remasters and Remakes are all the rage because they are easier and faster (and cheaper) than building a game from scratch. We're seeing devs that used to pump out new games every 2-3 years now taking twice as long to even have something to show in some cases. It's sad and unfortunate.
But just pointing out that based on what has actually released, we don't have much to judge in terms of how far Sony studios will go in terms of pushing the technology envelope. The few examples we do have show them pushing further than anyone else, but as a whole it's still very much TBD 3 years in. Hopefully Spider Man 2 will offer some hints once we get the full game, but yeah I wouldn't expect anything revolutionary there either.
