THE:MILKMAN
Member
Forgive my ignorance, but does this have any implications on the CU count?
I believe it all points it being 54/60 CUs? I'm sure more informed posters will correct me if wrong.
Forgive my ignorance, but does this have any implications on the CU count?
In a console I'm assuming it's going to be close to the 7800XT if the clock is at 2.5 GHz and the buswidth is 256 as the 7800XT would only have a 8% bandwidth advantage if the Pro has 18GBs ram speed.so PS5 pro will be a 7700XT? That should be pretty decent right?
Yes. But better yet,is that PS5pro would be based on RDNA3+. It should have stuff that RDNA3 (aa 7xxs series) lacks.so PS5 pro will be a 7700XT? That should be pretty decent right?
Technically speaking, the 7700XT.... IS the 7800XT. But yeah, it would be based on this and go from there.Too early to say. Most probably it will be a very custom piece of hardware based on the RX 7800 XT.
Yes.Forgive my ignorance, but does this have any implications on the CU count?
Also18Gbps19.5Gbps memory (reading the 7700xt, whoops)
I'm guessing the "[8 7] [8 7]" tweet was memory timings
No, I'm not sure if it was in this thread but the Pro going to have two shader engines with 8 WPG equaling 16+16+16+16 for 64 total CUs or 32 WPGs. If true, Redgaming Tech suggested the PS5 pro could have 64,96 or 128 ROPs.I believe it all points it being 54/60 CUs? I'm sure more informed posters will correct me if wrong.
No, I'm not sure if it was in this thread but the Pro going to have two shader engines with 8 WPG equaling 16+16+16+16 for 64 total CUs or 32 WPGs. If true, Redgaming Tech suggested the PS5 pro could have 64,96 or 128 ROPs.
We're still missing the fine details and also we don't know how much Sony can get AMD to customise the design or whether for example a full chip (no deactivating CUs for yields) is possible/likely.
Hopefully more solid answers come before years end.
I don't expect a repeat of the NEO developer documents PDF leak but hope the likes of DF start to drop crumbs. Maybe when Sony finally reveal the revised PS5 we get a hint or indications about a Pro?
Probably the greatest PDF ever released
Don't mess with Sony Ninja. They took the memory of it outta my head.I had a copy on my system but it has now disappeared. I'm convinced Cerny hacked my PC and took it back.....
No, I'm not sure if it was in this thread but the Pro going to have two shader engines with 8 WPG equaling 16+16+16+16 for 64 total CUs or 32 WPGs. If true, Redgaming Tech suggested the PS5 pro could have 64,96 or 128 ROPs.
Yeah I thought so - so Viola based off 7700xt (however we know that it will at least have the 256bit bus of the bigger brother as the ps5 already has a 256bit bus and gimping bandwidth to a 192bit bus wouldn't be a sensible move). Also will have RDNA3+ features / sony patented rt acceleration.No, Henderson leak talked about 18 Gbps chips (actually 18.000 MT/s but it's the same) just like the 7700 XT.
AMD should announce both cards at Gamescom in about 3 hours....
I'm curious about the prices
Thanks for the explanation.Yes. But better yet,is that PS5pro would be based on RDNA3+. It should have stuff that RDNA3 (aa 7xxs series) lacks.
Technically speaking, the 7700XT.... IS the 7800XT. But yeah, it would be based on this and go from there.
Yes.
Basically... RDNA makeup goes like this.
CU > WG (CU pair) > SA (shader array) > Shader engine (SA pair) > GPU (combination of SEs)
The PS5 has 5 WGS in its SA. The SA is the first level of GPU component grouping. Among other things, WGs in a SA share the same l1 cache. And has 2 SAs per SE. So that 36CU it has is made up of 4 SAs spit between 2 SEs. Oh, and then a WG is disabled in each SE that's how we end up with only 36CU instead of 40.
What that post suggests (or is rumored to suggest), is that the PS5pro would have 8 WGs/SA. Using the same PS5 layout, that would bring the total to 32 WGs. 8/SA, 16/SE and x2SE =32WG total. Disable one per SE and you end up with 30 WG active = 60CU.
Yeah I thought so - so Viola based off 7700xt (however we know that it will at least have the 256bit bus of the bigger brother as the ps5 already has a 256bit bus and gimping bandwidth to a 192bit bus wouldn't be a sensible move). Also will have RDNA3+ features / sony patented rt acceleration.
IDK,the only thing said about the tempest engine by Mark Cerny is that the TE is a repurposed CU without any cache and DMA'ing data.[16+14 ]+ [16+14]
With the missing WGP from each SA used for Tempest ?
This is why I am saying that the whole 64CU 8WGs/SA thing makes no sense and doesn't go in line with what Sony has been doing.I still think you all aren't paying attention to the actual 7800 XT chip.
![]()
![]()
![]()
![]()
I may be wrong but this makes much more sense to the actual chip. This is done to scale with the 7900XTX.
![]()
3 Shader Engines
2 Shader Arrays per Shader Engine
5 WGP per Shader Array
This lines up with the leaked specs from here:
AMD's RDNA 3 Graphics
Navi32
- gfx1101 (Wheat Nas)
- Chiplet - 1x GCD + 4x MCD (0-hi)
- 30 WGP (60 legacy CUs, 7680 ALUs)
- 3 Shader Engines / 6 Shader Arrays
- Infinity Cache 64MB (0-hi)
- 256-bit GDDR6
- GCD on TSMC N5, ~200 mm²
- MCD on TSMC N6, ~37.5 mm²
And it also matches how Navi 22 is half of Navi 21.
![]()
![]()
The PS5pro is probably going to be built on a 4nm node not a 5nm one. So it's not going to be pulling that power anyway.How would the 7800xt ~300W TDP workout?
As I said before, they are basically the exact same card. Off the wafer they are probably identical, one is just made up of the defective chips of the other. That and that during SOC packaging, the 7700xt would have only 3 instead of 4 MCDs. Each MCD adds x2 bus channels and 16MB of infinity cache. And that is exactly what the differences between the two GPUs are.Only $50 difference between the two cards....
![]()
IDK,the only thing said about the tempest engine by Mark Cerny is it because a CU without any cache and DMA'ing data.
I don't know how to post images with my phone but here from the PS5 thread.
![]()
Sony Principal Graphics Engineer confirms PS5 has custom architecture that is based on RDNA2. (Update: Read OP)
Rosario Leonardi, Principal Graphics Engineer at Sony Interactive Entertainment Europe, had this interesting tweet... He also confirmed there is no machine learning... You are confusing actual info with some bullshit. There is a full GCN support for hw BC. That's why it's "RDNA1".www.neogaf.com
Were you expecting less? Maybe something like Vega 64/56 or Radeon 7?
This rules out Red Gaming Tech rumors.Were you expecting less? Maybe something like Vega 64/56 or Radeon 7?
Nah 2 people asked if it was PS5 pro and he liked those tweetsYh my hunch is that's it's related to something totally different, Keplar is primarily a leaker in the PC space.
Unless Sony was going with 4 SEs,it seemed like a stretch for the pro to have 128 ROPs.This rules out Red Gaming Tech rumors.
It's not 64 ROPs / 2 Shader Engines or 128 ROPs / 4 Shader Engines.
When Mark Cerny said this,Unless Sony was going with 4 SEs,it seemed like a stretch for the pro to have 128 ROPs.
Also,I was thinking why couldn't Sony go for a "Navi dense" GPU similar to the Series X? If I recall correctly,the PS5 APU is 308mm and the Series X APU are 360mm when both were on 7nm so why couldn't the Series X design work better on a smaller node?
I'm just spit balling ideas, maybe Navi dense in an APU form factor makes it's difficult to reach high clocks so it makes sense to space out the APU components enough.
I'm very skeptical about rgt but that would be the case if we assume a 7800xt aka navi3 will be in the pro no? We can only make assumptions at this point really.This rules out Red Gaming Tech rumors.
It's not 64 ROPs / 2 Shader Engines or 128 ROPs / 4 Shader Engines.
I'm very skeptical about rgt but that would be the case if we assume a 7800xt aka navi3 will be in the pro no? We can only make assumptions at this point really.
How are you getting 25TF? The PS5prowoud be at best 18TF (native)by my estimates. Using native to describe the actual non-dual issue TF.7800 XT? So even with lower TDP etc, this thing could be 25TF or thereabouts? If so, then...jeez.
I don't know what's going on, but the idea of Sony using anything other than a 7700XTX on the PS5pro makes no sense at all. Sony has also always based their PS console on very definitive AMD hardware. ie, you can look at AMD GPU range and see the exact match of every AMD-based PS GPU. The 60CU interpretation of that tweet is suggesting something happens that Sony has never done, does not have to do, and would be stupid of them to do.Nah 2 people asked if it was PS5 pro and he liked those tweets
Yes assumptions until anything concrete comes out these are all rumors so far, we got only the most basic of spec outlines and so far we are all making assumptions like rgt etc, the specs can change also and leaks can detail an earlier concept for example the ps5 also had a vega based initial prototype I believe. I don't think they are putting a 7800xt in the pro, the tdp, heat and performance numbers dont matchup not to mention it would still have the pathetic rt perfrormance, people have been looking at the 7800xt specs as some eureka moment as if its confirmation that its definitely the chip thats going to be in the pro.Assumptions? Henderson said 60 CUs and Navi 32 is a 60 CU chip.... Just a coincidence?
Navi 42 is not coming out in 2024
I think it would be a mistake to look at the 7800/7700 and say, yes, that's what is going into the Pro. So I agree with you on that.Yes assumptions until anything concrete comes out these are all rumors so far, we got only the most basic of spec outlines and so far we are all making assumptions like rgt etc, the specs can change also and leaks can detail an earlier concept for example the ps5 also had a vega based initial prototype I believe. I don't think they are putting a 7800xt in the pro, the tdp, heat and performance numbers dont matchup not to mention it would still have the pathetic rt perfrormance, people have been looking at the 7800xt specs as some eureka moment as if its confirmation that its definitely the chip thats going to be in the pro.
The consoles launched before rdna2 retail cards as well and had many of the same features, so lets see how this turns out.
It could be done without impacting performance if they increase the amount of cache by SA.Taking a look at RDNA3.
32WGPs looks so unrealistic.
It would have to be 30WGPs for 54CUs or 36WGPs for 60CUs.
![]()
Yes but that card is releasing this year, not next year. I think for PS5 Pro we may get a RDNA4 architecture as a blueprint meaning they may change the number of WGP by SA if they change the structure (like cache amount) by SA. Things are getting spicy indeed! Now I am starting to understand the cryptic Kepler tweet about that nothing would prevent Sony from using 60CUs. And the Henderson leak is still right.AMD RX 7700 XT has 54 CUs, 6 less than the full Navi 32 Chip...
Just like we expect from PS5 Pro
![]()
AMD announces Radeon RX 7800 XT 16GB and RX 7700 XT 12GB graphics cards - VideoCardz.com
AMD Radeon RX 7800 XT & RX 7700 XT revealed, designed for 1440p gaming At Gamescom AMD announces its mid-range RX 7000 GPUs. AMD has unveiled two highly anticipated graphics cards: the RX 7800 XT and RX 7700 XT, filling a crucial gap in AMD’s RDNA3 desktop lineup. Although we had some...videocardz.com
It could be done without impacting performance if they increase the amount of cache by SA.
Yes but that card is releasing this year, not next year. I think for PS5 Pro we may get a RDNA4 architecture as a blueprint meaning they may change the number of WGP by SA if they change the structure (like cache amount) by SA. Things are getting spicy indeed! Now I am starting to understand the cryptic Kepler tweet about that nothing would prevent Sony from using 60CUs. And the Henderson leak is still right.
They could if they use a 8/7 WGP structure by SA. And they will use a new chip if they use RDNA4 as a blueprint. I'd say the real PS5 Pro GPU could be a possible RX 8700 AMD GPU releasing next year.If they use the full 60 CU chip, there's nothing left to improve yields....
The alternative is for Sony to ask AMD for a completely new chip.
I don't see that happening
Any opportunity to wear "Sex Panther" (by odeon) and I am so fuckin there.Like I said.
18.5 tf and we seeing a party in my pants!
Yes but that card is releasing this year, not next year. I think for PS5 Pro we may get a RDNA4 architecture as a blueprint meaning they may change the number of WGP by SA if they change the structure (like cache amount) by SA. Things are getting spicy indeed! Now I am starting to understand the cryptic Kepler tweet about that nothing would prevent Sony from using 60CUs. And the Henderson leak is still right.
I think it would be a mistake to look at the 7800/7700 and say, yes, that's what is going into the Pro. So I agree with you on that.
However, it would be smart to look at the 7700/7800 as a blueprint for what the PS5pro could be. Then coupe that blueprint with a couple of educated guesses, and as usual, we end up being very accurate to what ends up in the box.
Eg. PS5pro would be using 3rd gen AMD RT, 2nd Gen dual-issue compute, second-gen AI cores...etc. All things that exist in the current RDNA3 and would be improved in RDNA4. PS5pro would also likely be on a 4nm process, not 5nm, so it would be drawing even less power. This is why I say it would be based on RDNA3+.
And if you ask me, I strongly believe that the focus of the PS5 would be on better RT and AI-based upscaling. People think the PS5 Pro is angling to be a 4k 60fps machine. Nope, it's not going to be, no more than Sony made the PS4pro a 4k 30fps machine (which it wasn't). The PS4pro was a 4k 30fps CBR machine. The Ps5pro, in my opinion, is going to be a 1440p 60fps AI reconstructed to 4k machine in quality mode. Dropping to 1080p 100fps+ reconstructed to 4k in performance mode.
Yes I remember this. Technically both PS5 and XSX are RDNA1.5 GPUs. Stating anything else is just spinning. Even Sony never stated PS5 was true RDNA2. Only RDNA2 based GPU.I'm the opposite with specs,Audiophile . I'm getting more and more convinced we're getting pretty much another PS4 Pro. It is a mid-gen bump not a new gen and if they went ham with Zen 3 and RDNA3.5/4 it would directly affect how much better PS6 can be especially with ever diminishing returns. Also I'm starting to wonder if it would be possible PS5 Pro comes out as early as Spring 2024....
Zen 2 @4.0GHz+ (maybe increase cache?)
60CU/54CU @ ~ 2.5GHz RDNA 2 customised with cherry-picked RDNA3/4 features and Cerny sauce
16GB 18Gbps GDDR6
Talking about blueprints, this morning I was reading the Tweeter thread by Locuza from 2 years ago discussing the console APU die shots and he says they basically employ a RDNA1 design for the render frontend and joked they're all RDNA 1.5!
![]()
Given this I would expect PS5 Pro to have a base of the PS5 GPU architecture and then backport whatever new features they want from the newer ones as they did with PS4 Pro.
Seems AMD is designing some of their GPUs in partnership with Sony so we shouldn't be surprised if PS5 Pro is again co-designed by Sony and AMD. So it's likely going to be a new product in 2024 and not using one year old AMD card (the RX 7800) as its blueprint.
This implies PS5 chip was finalized in 2019 and released in 2020.It could be done without impacting performance if they increase the amount of cache by SA.
Yes but that card is releasing this year, not next year. I think for PS5 Pro we may get a RDNA4 architecture as a blueprint meaning they may change the number of WGP by SA if they change the structure (like cache amount) by SA. Things are getting spicy indeed! Now I am starting to understand the cryptic Kepler tweet about that nothing would prevent Sony from using 60CUs. And the Henderson leak is still right.
The 7800XT chip has being designed in 2022 for this year release. PS5 Pro should get what AMD is designing this year for their 2024 cards: RDNA4 (which Kepler said it would release in 2024).This implies PS5 chip was finalized in 2019 and released in 2020.
![]()
It also resembles RDNA1 but is based on RDNA2.
RDNA1
![]()
PS5
RDNA2![]()
![]()
If PS5 Pro is releasing next year, it'll be finalized this year, 2023.
Rumors say the chip will be RDNA3.5, so it'll look like RDNA3 but is based on RDNA4.
One could look at the 7800XT chip and see the possible layout of the PS5 Pro chip.
They could if they use a 8/7 WGP structure by SA. And they will use a new chip if they use RDNA4 as a blueprint. I'd say the real PS5 Pro GPU could be a possible RX 8700 AMD GPU releasing next year.
See Kepler tweet about the 8/7 thing:
I'd gladly take that, tbh.The Ps5pro, in my opinion, is going to be a 1440p 60fps AI reconstructed to 4k machine in quality mode. Dropping to 1080p 100fps+ reconstructed to 4k in performance mode.
The 7800XT chip has being designed in 2022 for this year release. PS5 Pro should get what AMD is designing this year for their 2024 cards: RDNA4 (which Kepler said it would release in 2024).
We would have had the 7800 XT chip as a baseline IF Pro was released this year. But I bet Sony didn't think that card was good enough for them specifically in regards to RT performance.
Sony and Microsoft use custom chips and they aren't off the shelf parts, the XboneX had a total of 44 CUs and 40 active, AMD released the Vega series of GPUs in the same year as the XboneX. I bring that up because the reveal of the XboneX, AMD at that time didn't have a 44/40 CU GPU.That tweet could just be him hypothesizing. Or it could be true.
Just doesn't make sense to me though, cause that would mean they are asking AMD for a new Chip. Unless AMD has a GPU using 64CUs next year, I don't see that happening. Just all seems unnecessary. lie why do that if they have a perfectly fine and capable chip already?
Also, consider that there are more relevant things to spend extra die space on than ore CUs. Eg. Mor cache for both CPU and GPU. A smaller chip also means less power-hungry. And of course... cheaper.
I think it would be smart to always lean on the assumption that whatever ends up in the PS5pro being the smallest most performant Chip Sony can make. I would be pleasantly surprised if they end up going for 64CU, just do not see that happening. Just seems unnecessary.
I believe it all points it being 54/60 CUs? I'm sure more informed posters will correct me if wrong.