PS5 Pro codenamed ‘PROJECT TRINITY’ details and release date

Loy310

Veteran
14 Aug 2022
1,561
1,832
I cant wait to see the game they got lined up to market this thing with.
PS4 Pro had HZD to co-market the pro if i recall.
GTA6 maybe no loading 4k/60, 8k/30, comes bundled?

If Jim can pull that off with R* then PS5 will make sales history for sure. Even more so considering the cod marketing agreement end in 2024 sony will be looking for something/someone huge to co market with in its place next.
Next year and a half will be nuts on ps5 thats for sure.
 
24 Jun 2022
3,949
6,878
60 CUs feels kind of low, no?

Not really. Remember, RDNA3 uses dual-issue shaders. So even if the new PS5 has the same GPU clock as PS5, that's 34.25 TF.

But as we should know by now, TF is not a realistic measurement of game performance. I'm more interested in what changes there are to the primitive units (it's 12 triangle cullings per cycle now, up from 8 with RDNA2, so at the same GPU clock poly perf should jump to 26.76 Gpolys/sec culling > 13.38 Gpolys/sec rasterization), ROPs, TMUs, RT acceleration, GPU caches (if any form of Infinity Cache is present, for example, which I almost feel 100% is), etc.

The RAM bandwidth seems like it'll be 576 GB/s if it sticks with a 256-bit interface. That's proportionally about similar a bandwidth jump as PS4 Pro had over PS4, which was actually a point of contention.. So either the GDDR clocks go up, or they have infinity cache with MCDs. I wouldn't expect a ton of L3$ for PS5 Pro; maybe something like 48-64 MB. But that would be enough.

About memory capacity..again if they're on a 256-bit bus, they can either only do 16 GB or up to 32 GB with GDDR6W. But if they use GDDR6W they'd have to double the chip bandwidth to hit 576 GB/s since GDDR6W is 64-bit per chip module. The tradeoff is you get more capacity per chip. However I don't think there's going to be a memory capacity increase for the Pro, so they are probably aiming for something like this:

GPU (minimum): 1 GCD, 2 SE, 2 SA, 5 WGP, 8 SIMD, 24 ALUs (3,840 shader cores) @ 2.23 GHz (34.25 TF). 48-64 L3$ (remember, RDNA3 and assuredly RDNA4 use dual-issue shaders)

Memory (likely): 256-bit interface, 8x 2 GB 18 Gbps GDDR6 modules (576 GB/s), 16 GB capacity (depending on what's cheaper or works better for production logistics, either they'll go with faster RAM or a bit more L3$ on the GPU).

Storage (possible): 2 TB internal SSD m.2 NVME 2.1 PCIe 5.0 (supports drives up to 16 GB/s bandwidth; default SSD probably closer to base PS5's bandwidth) with some tweaks/improvements to SSD I/O. Private NAND block (for firmware updates and OS restore partitions) non-accessible by user (single-lane PCIe 4.0 interface)

That's all I can really speculate on based upon what's being shared. I do think the ROP and TMU counts will increase, though not by 2x, and the shader logic might be a bit higher (2.5 GHz vs 2.23 GHz). The frontend logic might stay at base PS5's GPU clock speed, though.

PS5 GPU has 18WGP.
PS5 GDDR6 is 14000 MT/s.

Memory is not a issue because the bus will dictate the final bandwidth so a 320bits bus can do wonders even without a big increase in MT/s.

But the WGP seems a small jump.
30 WGP at same speeds as actual PS5 will give you around 17TFLOPS.

And I have the feeling the clocks will decrease due the more WGPszzz maybe 15-16TFs.

No no, you have to remember that RDNA3 (or what PS5 Pro is probably based on, RDNA4) uses dual-issue shaders. I'm not 100% sure how they work, as they may only dual-issue in specific conditions (i.e the doubled increase in compute is not 100% always present), but that's how they function.

So theoretically PS5 Pro's TF perf should be at 34.25 TF if the shader clock is the same as base PS5's.
 
Last edited:
  • brain
Reactions: Kokoloko

historia

Veteran
29 Jun 2023
2,818
2,720
I cant wait to see the game they got lined up to market this thing with.
PS4 Pro had HZD to co-market the pro if i recall.
GTA6 maybe no loading 4k/60, 8k/30, comes bundled?

If Jim can pull that off with R* then PS5 will make sales history for sure. Even more so considering the cod marketing agreement end in 2024 sony will be looking for something/someone huge to co market with in its place next.
Next year and a half will be nuts on ps5 thats for sure.
RDR2 got marketing with PS4 pro so there's that.

They seems hitting jackpots when it comes to marketing, this year Hogwarts Legacy pull big numbers
 
  • they're_right_you_know
Reactions: Jim Ryan

historia

Veteran
29 Jun 2023
2,818
2,720
Not really. Remember, RDNA3 uses dual-issue shaders. So even if the new PS5 has the same GPU clock as PS5, that's 17.13 TF.

But as we should know by now, TF is not a realistic measurement of game performance. I'm more interested in what changes there are to the primitive units (it's 12 triangle cullings per cycle now, up from 8 with RDNA2, so at the same GPU clock poly perf should jump to 26.76 Gpolys/sec culling > 13.38 Gpolys/sec rasterization), ROPs, TMUs, RT acceleration, GPU caches (if any form of Infinity Cache is present, for example, which I almost feel 100% is), etc.

The RAM bandwidth seems like it'll be 576 GB/s if it sticks with a 256-bit interface. That's proportionally about similar a bandwidth jump as PS4 Pro had over PS4, which was actually a point of contention.. So either the GDDR clocks go up, or they have infinity cache with MCDs. I wouldn't expect a ton of L3$ for PS5 Pro; maybe something like 48-64 MB. But that would be enough.

About memory capacity..again if they're on a 256-bit bus, they can either only do 16 GB or up to 32 GB with GDDR6W. But if they use GDDR6W they'd have to double the chip bandwidth to hit 576 GB/s since GDDR6W is 64-bit per chip module. The tradeoff is you get more capacity per chip. However I don't think there's going to be a memory capacity increase for the Pro, so they are probably aiming for something like this:

GPU (minimum): 1 GCD, 2 SE, 2 SA, 5 WGP, 8 SIMD, 24 ALUs (3,840 shader cores) @ 2.23 GHz (17.126 TF). 48-64 L3$

Memory (likely): 256-bit interface, 8x 2 GB 18 Gbps GDDR6 modules (576 GB/s), 16 GB capacity

Storage (possible): 2 TB internal SSD m.2 NVME 2.1 PCIe 5.0 (supports drives up to 16 GB/s bandwidth; default SSD probably closer to base PS5's bandwidth) with some tweaks/improvements to SSD I/O. Private NAND block (for firmware updates and OS restore partitions) non-accessible by user (single-lane PCIe 4.0 interface)

That's all I can really speculate on based upon what's being shared. I do think the ROP and TMU counts will increase, though not by 2x, and the shader logic might be a bit higher (2.5 GHz vs 2.23 GHz). The frontend logic might stay at base PS5's GPU clock speed, though.
SSD probably gonna be 1.65TB 12 channel PCIE 4.0 again, why should they cannabalize cheap PCIE 4.0 2280 SSD support. CPU could be Zen 2 or better Zen 4 like ROG Ally, if Sony wants to reduce bottlenecks.


GPU:

RDNA 3: 30 WGP so 4 SIMD64 blocks, 128 ALUs running at >2.5Ghz because the architecture allow for it. So probably bigger or more efficient cooling. So somewhere around 35-40 TFLOPs FP32

RDNA 4: probably improvements to BVH tree traversal, especially it will not rely on ALUs. We haven't even seen Navi 41 die yet so can't say anything.

Memory:

GDDR6X: cooler at 576 GB/s bandwidth at 384 bit? If it has 24GB then that number gonna higher
 
Last edited:

historia

Veteran
29 Jun 2023
2,818
2,720
Also I hope that people are not stupid enough to bite the TFLOPs number equal % of performance delta again.
😅

For me a consistent 70% performance uplift in rasitization already justify the purchase.

Hopefully there will be a 200% performance uplift in Ray-Tracing, and we could even get closer to full path Ray-Tracing.

Next PS exclusive game running at 30FPS/reconstructed 4K/full-path Ray Tracing would be amazing.
 
  • Shake
Reactions: SuperDuper
24 Jun 2022
3,949
6,878
SSD probably gonna be 1.65TB 12 channel PCIE 4.0 again, why should they cannabalize cheap PCIE 4.0 2280 SSD support. CPU could be Zen 2 or better Zen 4 like ROG Ally, if Sony wants to reduce bottlenecks.

My understanding was that PCIe 5.0 is BC with 4.0, and the current NVME spec supports the 5.0 interface. So it was more about future-proofing the system for faster SSDs to release on the market, not cutting support for 4.0. m.2 form factor should be the same.

I was just thinking Sony may switch from internal soldered NAND chips for the SSD and go with an m.2 form factor, but maybe have two slots. Meanwhile use a small NAND block soldered to the system interfacing with the I/O that has restore partitions and is used for firmware updates, but wouldn't need a 5.0 lane connection.

I do think they would do 12 channels for the internal drive even if they go to a standard m.2 form factor, though, and your size sounds about right.

GPU:

RDNA 3: 30 WGP so 4 SIMD64 blocks, 128 ALUs running at >2.5Ghz because the architecture allow for it. So probably bigger or more efficient cooling. So somewhere around 35-40 TFLOPs FP32

RDNA 4: probably improvements to BVH tree traversal, especially it will not rely on ALUs. We haven't even seen Navi 41 die yet so can't say anything.

Agreed. I'm also curious if RDNA4 will support multi-GCDs; that would probably enable further chipletization, if AMD can pull it off.

Memory:

GDDR6X: cooler at 576 GB/s bandwidth at 384 bit? If it has 24GB then that number gonna higher

Nah, I don't see Sony going 384-bit wide. They can technically do 24 GB on 384-bit with a mix of 4 GB GDDR6W (4x) and 2 GB GDDR6 chips (4x), I guess it's going to come down to what works best for Sony's logistics.

The benefit with GDDR6W would be simplifying the layout of the PCB as you need less chips physically on the board for the capacity requirements. It's also capable of speeds closer to 6X than basic 6.
 
  • Like
Reactions: Killer_Sakoman

Johnic

Veteran
24 Mar 2023
3,777
6,263
Outer Heaven
Me looking at all these specs

Confused Rooster Teeth GIF by Achievement Hunter


The series S will still be a bottleneck this entire gen. But this might help devs favor the PS5 even more. Make developing even easier.
 
  • haha
Reactions: Kokoloko

historia

Veteran
29 Jun 2023
2,818
2,720
My understanding was that PCIe 5.0 is BC with 4.0, and the current NVME spec supports the 5.0 interface. So it was more about future-proofing the system for faster SSDs to release on the market, not cutting support for 4.0. m.2 form factor should be the same.

I was just thinking Sony may switch from internal soldered NAND chips for the SSD and go with an m.2 form factor, but maybe have two slots. Meanwhile use a small NAND block soldered to the system interfacing with the I/O that has restore partitions and is used for firmware updates, but wouldn't need a 5.0 lane connection.

I do think they would do 12 channels for the internal drive even if they go to a standard m.2 form factor, though, and your size sounds about right.


Nah, I don't see Sony going 384-bit wide. They can technically do 24 GB on 384-bit with a mix of 4 GB GDDR6W (4x) and 2 GB GDDR6 chips (4x), I guess it's going to come down to what works best for Sony's logistics.

The benefit with GDDR6W would be simplifying the layout of the PCB as you need less chips physically on the board for the capacity requirements. It's also capable of speeds closer to 6X than basic 6.
No not the split memory configuration again, three consoles has suffered because of it, probably four if you count the ESRAM bs.

I still surprise they are going for 30 WGPs instead of 36 WGPs or 18. Probably this is about Ray-Tracing after all.

If they manage to pull full-path Ray-Tracing games on this console then it should lay groundwork nicely for next gen
 

Banana

Well-known member
10 Jan 2023
322
219
If they manage to pull full-path Ray-Tracing games on this console then it should lay groundwork nicely for next gen
AMD doesn't have the ability to do that yet, so unless they suddenly develop that for RDNA4, probably not.
 

historia

Veteran
29 Jun 2023
2,818
2,720
AMD doesn't have the ability to do that yet, so unless they suddenly develop that for RDNA4, probably not.
They can, just their implementation of RT is just pathetic, compared to RTX 4090 which is 200% faster.
 

ethomaz

Rebolation!
21 Jun 2022
11,779
9,601
Brasil 🇧🇷
PSN ID
ethomaz
60 CUs RDNA3, running at 2.2 Ghz, that gonna be at least 35 TFLOPs FP32, great marketing number Sony (y)

Just hope for at least 100% performance uplift.
Ohhhhh AMD is doing the TFLOPS bullshit of nVidia now?
Cards with 15-20TFs being marketed as 30-40TFs?
 
24 Jun 2022
3,949
6,878
No not the split memory configuration again, three consoles has suffered because of it, probably four if you count the ESRAM bs.

I still surprise they are going for 30 WGPs instead of 36 WGPs or 18. Probably this is about Ray-Tracing after all.

If they manage to pull full-path Ray-Tracing games on this console then it should lay groundwork nicely for next gen

That's the neat thing about GDDR6W; it functions exactly like GDDR6, it's not like it would have to be treated like a completely different memory type.

But I agree insofar as, to keep design simple, they'd go with a flush capacity rather than mixing the capacities between two different size types, for reasons you reference.
 

historia

Veteran
29 Jun 2023
2,818
2,720
Ohhhhh AMD is doing the TFLOPS bullshit of nVidia now?
Cards with 15-20TFs being marketed as 30-40TFs?
It is 30-40 TFs... On paper. They simple double SIMD blocks per compute units, give it another 50% of ALUs.

Reality is that it diminishing return because IPC doesn't increase. RDNA4 approach will be to increase node from 7nm FF/5nm to 4nm FF which manage to incrase transistors count therefore improving IPC.
 

ethomaz

Rebolation!
21 Jun 2022
11,779
9,601
Brasil 🇧🇷
PSN ID
ethomaz
It is 30-40 TFs... On paper. They simple double SIMD blocks per compute units, give it another 50% of ALUs.

Reality is that it diminishing return because IPC doesn't increase. RDNA4 approach will be to increase node from 7nm FF/5nm to 4nm FF which manage to incrase transistors count therefore improving IPC.
These double SIMD are not independent, shares the same control units, and have a lot of restrictions to be used simultaneously… in 90% of cases it is will run like a single SIMD.

It is basically what nVidia did with some GeForce arch in the past.

Performance wise that 30TFs RDNA3 will run more in close to 15TFs RDNA2 than 30TFs.
That for me is just marketing.

7900XT 84CU 51TFs > 6900XT 80CU 23TFs
Difference in performance ~25%

PS. the writepaper for these interested: https://www.amd.com/system/files/TechDocs/rdna3-shader-instruction-set-architecture-feb-2023_0.pdf
 
Last edited:
  • Informative
Reactions: Entropi

Entropi

Veteran
Icon Extra
22 Jan 2023
3,417
4,667
IMO releasing a Pro console gives MS the opportunity to release their next gen console a year or more before the PS6, if MS skips a pro console themselves. Given the state of things, that’s not a wise decision by Sony.
 

Ezekiel

Veteran
21 Jun 2022
784
1,129
IMO releasing a Pro console gives MS the opportunity to release their next gen console a year or more before the PS6, if MS skips a pro console themselves. Given the state of things, that’s not a wise decision by Sony.
Worked great last gen
 

historia

Veteran
29 Jun 2023
2,818
2,720
IMO releasing a Pro console gives MS the opportunity to release their next gen console a year or more before the PS6, if MS skips a pro console themselves. Given the state of things, that’s not a wise decision by Sony.
There is more Xbox SeS then SeX. MS already tried to skip a gen and they failed at it