• 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.

PlayStation 5 Pro Unboxed, 16.7 TFLOPs GPU Compute Confirmed

damidu

Member
yikes-gif.gif



lol yeah good times.

don't envy the poor souls, who listened to this moron and still trying to soldier through the gen with a 1060 build
 

SKYF@ll

Member
Some of you guys get way too worked up about this. Being curious about the discrepancy isn’t some crusade against Sony. If the Pro does include VOPD is it the instruction set from RDNA 3 or RDNA 4? If it’s either what’s the performance differences between the two?
Or are these questions also not allowed to be asked?
Even when the PS5 Pro runs RDNA3 VOPD, the game benchmarks do not increase much.
However, leaked developer documents indicate that the PS5 Pro has 67 TFLOPS in FP16, which means 33.5 TFLOPS in FP32, so it can be expected that the PS5 Pro is equipped with a dual issue compatible GPU.
PS5-Pro-PSSR.jpg
 

Lysandros

Member
EDIT: yes I did remember it right. PS5 was similar to a 1060 according to that guy in DMC5 using the highly CPU bottlenecked 120hz mode. Such a clown.
Honestly that video was the worse example of the PS5 hardware misinformation campaign to intentionally to show it in the worse light possible in comparison to XSX. And of course, he didn't do such a 'comparison' piece for XSX, so unfortunate... He put even DF/Dictator to shame with his efforts to undermine the machine, clown of the generation indeed. Sony was way too tolerant on that one.
 
Last edited:

winjer

Gold Member
Honestly that video was the worse example of the PS5 hardware misinformation campaign to intentionally to show it in the worse light possible in comparison to XSX. And of course, he didn't do such a 'comparison' piece for XSX, so unfortunate... He put even DF/Dictator to shame with his efforts to undermine the machine, clown of the generation indeed. Sony was way too tolerant on that one.

People focus too much on the 1060 he used, but the reality is that at the 120Hz mode, with 1080p or lower and with low detail settings, with crossgen titles, it becomes a CPU test.
It would not matter much if he used a 1060 or a 3090. Most often than not the CPU would be the limiting factor.
So it becomes a comparison between the 3300X and the PS5 CPU.
The 300X does have fewer cores, but has higher clocks and much more L3 cache per CCD.
And this shows, as the 3300X gest better 1% lows.
 

Lysandros

Member
People focus too much on the 1060 he used, but the reality is that at the 120Hz mode, with 1080p or lower and with low detail settings, with crossgen titles, it becomes a CPU test.
It would not matter much if he used a 1060 or a 3090. Most often than not the CPU would be the limiting factor.
So it becomes a comparison between the 3300X and the PS5 CPU.
The 300X does have fewer cores, but has higher clocks and much more L3 cache per CCD.
And this shows, as the 3300X gest better 1% lows.
He did say that the machine "this thing" was equivalent to "a mid to high end PC GPU of two-three years ago" in 2020... So mid to high of 2017-18. Emphasis on the 'outdated' in a sense, which is blatantly false of course. That was the final messaging, his assessment. That was a monumental and intentional fuckup, no need to soften it.
 
Last edited:

winjer

Gold Member
He did say that the machine "this thing" was equivalent to "a mid to high end PC GPU of two-three years ago" in 2020... So mind to high of 2017-18. That was the final messaging, his assessment. That was a monumental and intentional fuckup, no need to soften it.

That is true. But the fact remains, that he did a CPU benchmark.
The comparison to the 1060 was pointless. But the comparison to the 3300X was a lot more accurate and interesting.
 

Gaiff

SBI’s Resident Gaslighter
That is true. But the fact remains, that he did a CPU benchmark.
The comparison to the 1060 was pointless. But the comparison to the 3300X was a lot more accurate and interesting.
He did, but that’s not the way he framed it. Steve damn well knew this was a CPU test, yet he kept going back to the GPU.
 
Last edited:

winjer

Gold Member
He did, but that’s not the way he framed it. Steve damn well knew this was a CPU test, yet her kept going back to the GPU.

His conclusion is not wrong. For 120 fps mode, with low settings and resolution, a 1060 is enough for those games.
And the 3300X does a better job, than the Zen2 CPU on the PS5.
But you are right, that he over focused on the 1060. Had he focused on the 3300X, and how those tests are mostly CPU bound, the reception to that video would have been better.
It's not that the information and benchmarks in that video are wrong, it's just the framing that is.
 
Last edited:
His conclusion is not wrong. For 120 fps mode, with low settings and resolution, a 1060 is enough for those games.
And the 3300X does a better job, than the Zen2 CPU on the PS5.
But you are right, that he over focused on the 1060. Had he focused on the 3300X, and how those tests are mostly CPU bound, the reception to that video would have been better.
It's not that the information and benchmarks in that video are wrong, it's just the framing that is.
Obviously. Most of those **cough journalists **cough doing hit pieces against Playstation are using this methodology.
 

damidu

Member
His conclusion is not wrong. For 120 fps mode, with low settings and resolution, a 1060 is enough for those games.
And the 3300X does a better job, than the Zen2 CPU on the PS5.
But you are right, that he over focused on the 1060. Had he focused on the 3300X, and how those tests are mostly CPU bound, the reception to that video would have been better.
It's not that the information and benchmarks in that video are wrong, it's just the framing that is.
yeah almost like he cherry picked the whole setup in such way, so he can come to that clickbait grossly misleading framing. lol

its classic insecure pc dork hit piece.
no need to defend these idiots
 

winjer

Gold Member
Obviously. Most of those **cough journalists **cough doing hit pieces against Playstation are using this methodology.

I don't think he did a hit piece against the PS5, it's more like he was out of his depth, doing console analysis.
Don't blame something on malice, that can easily explained by incompetence.
 

winjer

Gold Member
16.2 vs 16.7 Teraflops, RDNA 2 base, it's an RX6800, you know the saying if it walks like a duck

Non-sense.
The closest we have now would be a 7700XT. Similar cache sizes, and feature set, of RDNA3.
And we still have to see what changes with the shaders in RDNA4.
 

Gaiff

SBI’s Resident Gaslighter
I don't think he did a hit piece against the PS5, it's more like he was out of his depth, doing console analysis.
Don't blame something on malice, that can easily explained by incompetence.
Nah, don’t agree this time. Steve is an intelligent man. This was just him dunking on the console with a flawed methodology.
 

TBiddy

Member
Even when the PS5 Pro runs RDNA3 VOPD, the game benchmarks do not increase much.
However, leaked developer documents indicate that the PS5 Pro has 67 TFLOPS in FP16, which means 33.5 TFLOPS in FP32, so it can be expected that the PS5 Pro is equipped with a dual issue compatible GPU.
PS5-Pro-PSSR.jpg

The way that image is presented reminds me of that guy on Twitter. What was his name. The one with the claims of a hidden GPU in the power brick in the OG XB1. The guy that writes gibberish constantly.
 

winjer

Gold Member
Nah, don’t agree this time. Steve is an intelligent man. This was just him dunking on the console with a flawed methodology.

It's not like he has never made mistakes.
And although he is intelligent, he often gets out of his depth.
Just take a look at his interviews, with Intel or Nvidia engineers, talking about low level GPU specs, and he quickly loses his footings.
And don't forget, that he made the best test and tear down of the PS5. No one did better tests with thermals and power, than GN.
 

Gaiff

SBI’s Resident Gaslighter
It's not like he has never made mistakes.
Everyone makes mistakes, but you, I, and anyone with just passive knowledge of hardware could see the flaw in his methodology. This wasn’t some slight error that could have flown under the radar. This was basic benchmarking 101 and he knows how to do that. The methodology doesn’t change all that much from console to PC.
And although he is intelligent, he often gets out of his depth.
Just take a look at his interviews, with Intel or Nvidia engineers, talking about low level GPU specs, and he quickly loses his footings.
Totally different things. These are engineers talking in-depth about things he has no training on. Benchmarking is nowhere near as complex.
And don't forget, that he made the best test and tear down of the PS5. No one did better tests with thermals and power, than GN.
Yes, I saw those, but this doesn’t change the test he did was terrible and it was easy to tell for anyone, including Steve himself. That’s why I don’t think it was due to incompetence. He saw an easy way to shit on the console and took it.
 
the GPU is clearly an RX 6800 with RDNA 4 RT features added
I am aware as above!

You quoted my response to somebody who said it wasn't like an RX 6800 - it's clear its an RDNA 2 RX 6800 with RDNA 4 RT
Although the purebred RDNA2 RX6800 uses the same floating-point calculations as the PS5 Pro, its RT method is old and inefficient, and it does not have AI upscaling functionality.

dLjZpxi.jpg
 
Last edited:

SKYF@ll

Member
The way that image is presented reminds me of that guy on Twitter. What was his name. The one with the claims of a hidden GPU in the power brick in the OG XB1. The guy that writes gibberish constantly.
Indonesians who love to fantasize...blueisviolet?
 

winjer

Gold Member
it's clear its an RDNA 2 RX 6800 with RDNA 4 RT

It's not. The PS5 Pro has the same cache sizes has RDNA3 (except the L3).
It has WMMA, either the RDNA3 or RDNA4 version. But I think there were leaks that said it's the RDNA4 version.
And it has VOPD, that despite being useless in RDNA3, it's still a feature that is not on RDNA2.
 
I'm not disputing there are changes winjer winjer it's slightly higher in teraflops than a 6800, there's clearly changes, it's added RDNA 4 RT and ML, WMMA, VOPD, will all be RDNA 4 based I'd imagine and come from the fact it's got RDNA 4 touches to it, but it's clear the base design of PS5 Pro was built around the RX 6800.
 

winjer

Gold Member
I'm not disputing there are changes winjer winjer it's slightly higher in teraflops than a 6800, there's clearly changes, it's added RDNA 4 RT and ML, WMMA, VOPD, will all be RDNA 4 based I'd imagine and come from the fact it's got RDNA 4 touches to it, but it's clear the base design of PS5 Pro was built around the RX 6800.

It has all the features of RDNA3 and the RT of RDNA4 and maybe, also the WMMA of RDNA4, yet you still insist it's a 6800?
Seriously, wtf?
 

Gaiff

SBI’s Resident Gaslighter
It has all the features of RDNA3 and the RT of RDNA4 and maybe, also the WMMA of RDNA4, yet you still insist it's a 6800?
Seriously, wtf?
Gooners: It’s a dog.

Winjer: No, it’s a leopard here is why.

Gooners: Well, it’s an orange spotted feline dog that can climb trees and lives in the African savana.
 
Last edited:

Mr Moose

Member
I'm not disputing there are changes winjer winjer it's slightly higher in teraflops than a 6800, there's clearly changes, it's added RDNA 4 RT and ML, WMMA, VOPD, will all be RDNA 4 based I'd imagine and come from the fact it's got RDNA 4 touches to it, but it's clear the base design of PS5 Pro was built around the RX 6800.
That's 2 gens behind (RDNA2).
 

Bojji

Member
It would be hard for AMD to add features to 4 year old GPU.

It's clearly some hybrid of RDNA3/3.5/4

On the other hand, they still use 2019 CPU on this thing...
 
If it's RDNA 3 16.7 teraflops makes it around even maybe slightly worse than the RDNA 2 based standard PS5, RDNA 3.0 7600 XT has 22.57 Tflops, which is more than the RX 6800 but the 16.17 Tflops 6800 has 36% more relative performance in game, it's not just an off shelf RX 6800 but it's RDNA2.0 RX 6800 based with RDNA 4 RT and ML
 

winjer

Gold Member
If it's RDNA 3 16.7 teraflops makes it around even maybe slightly worse than the RDNA 2 based standard PS5, RDNA 3.0 7600 XT has 22.57 Tflops, which is more than the RX 6800 but the 16.17 Tflops 6800 has 36% more relative performance in game, it's not just an off shelf RX 6800 but it's RDNA2.0 RX 6800 based with RDNA 4 RT and ML

Like it has been said so many times in this forum, including this thread, AMD presents RDNA3 TFLOPs while accounting for VOPD.
But VOPD is almost useless for games, so even Sony decided to present the value without it.
 

Bojji

Member
Like it has been said so many times in this forum, including this thread, AMD presents RDNA3 TFLOPs while accounting for VOPD.
But VOPD is almost useless for games, so even Sony decided to present the value without it.

Maybe they even turned off VOPD in their software.
 

Lysandros

Member
Sony could even tell AMD to remove VOPD and save on some transistors.
A bit like they did with the FP units in the Zen2 CPU of the PS5.
How do we achieve 300 TOPs in this case, is it (nearly) all coming from the custom ML silicone?
 
Last edited:

Mr Moose

Member
If it's RDNA 3 16.7 teraflops makes it around even maybe slightly worse than the RDNA 2 based standard PS5, RDNA 3.0 7600 XT has 22.57 Tflops, which is more than the RX 6800 but the 16.17 Tflops 6800 has 36% more relative performance in game, it's not just an off shelf RX 6800 but it's RDNA2.0 RX 6800 based with RDNA 4 RT and ML
Isn't that 11.28TF? 2048 x 2 x 2755?
 

winjer

Gold Member
How do we achieve 300 TOPs in this case, is it all coming from the custom ML silicone?

As far as we know, the Pro is using WMMA. I don't think it has a dedicated NPU.
But at this point, it might not take long until some one photographs and annotates the Pro's chip. Then we can have a better idea of what is in there.
 

Valonquar

Member
The neckbeards creating this round of unboxing videos have been some of the least exciting videos I have ever had the displeasure of cringing through.
 

Lysandros

Member
As far as we know, the Pro is using WMMA. I don't think it has a dedicated NPU.
But at this point, it might not take long until some one photographs and annotates the Pro's chip. Then we can have a better idea of what is in there.


At 3:42, "We added custom hardware for machine learning."



At 1:03, "Have an actual custom silicon AI upscaler."

As far as we know, they can't be any clearer.
 
Last edited:


At 3:42, "We added custom hardware for machine learning."



At 1:03, "Have an actual custom silicone AI upscaler."

As far as we know, they can't be any clearer.

I remember reading that a dedicated NPU at 300 TOPS would take up a lot of die space, I think it was K KeplerL2 who said that but I could be mistaken.
 

Lysandros

Member
I remember reading that a dedicated NPU at 300 TOPS would take up a lot of die space, I think it was K KeplerL2 who said that but I could be mistaken.
Is Kepler running the PlayStation engineering department or something? Why should we believe his words instead of the actual chief architect of the machine? No one is saying the entire 300 TOPs is coming from that officially mentioned custom block by the way, i also do believe that PS5 PRO has RDNA 3 dual issue fp32 like the leaked developer documentation mentioned. This leaked Brazilian manual spec list doesn't invalidate it.
 
Last edited:
Top Bottom