Support NeoGAF

[Hyper_hyper]

You can check videos comparing 6700 vs. PS5. Or PS5 Pro vs 9600XT.

You meant 9060XT? In brute numbers 9060XT has 50% more TF compared to PS5 Pro in FP32 and it is a RDNA4 card. It has 16GB/320GB/s and 32MB of Infinity cache. Not so similar. At all. So, like i've said, you still treating games performance on PC and consoles 1 :1.

You are thinking about consoles like Saturn, PS2, PS3 etc. where hardware was exotic. This is not the case since PS4.

It is and always will be. As Carmack said that 15+ years ago. Still applying this day.

 

[Closer]

I never seen a single Resident Evil title at 30 fps on console. Seems extremely unlikely knowing them.

Did you play the classics? Pretty sure those are 30fps.

 

[adamsapple]

Did you play the classics? Pretty sure those are 30fps.

I'm gonna assume he meant any RE engine based Resident Evil games.

 

[Bojji]

You meant 9060XT? In brute numbers 9060XT has 50% more TF in FP32 and it is a RDNA4 card. It has 16GB/320GB/s and 32MB of Infinity cache. Not so similar. At all. So, like i've said, you still treating games performance on PC and consoles 1 :1.


It is and always will be. As Carmack said that 15+ years ago. Still applying this day.

What Carmac said was not accurate when he said it, even less accurate now.

9060XT has fake 2x TF numbers (they use since RDNA3). Real performance is in line with 16TF 6800.

 

[Closer]

I'm gonna assume he meant any RE engine based Resident Evil games.

That or he only played the PS5 games.

 

[Hyper_hyper]

What Carmac said was not accurate when he said it, even less accurate now.

You are dev like him? You understand how console performance works, right? You are a dev over 30+ years of experience on consoles, right?

9060XT has fake 2x TF numbers (they use since RDNA3). Real performance is in line with 16TF 6800.

2x TF numbers is used since 10 years ago. LOL
It isn't fake numbers. 9060 is much more power efficient GPU compared to 6800 in every way ( despite 6800 has a bigger bandwidth), and much higher clock. Since it is RDNA 4 designed GPU, it has FSR4 support ( 6800 doesn't have it, at least not supported by AMD) And it has better RT performance in games.
You used 9060XT vs Pro in post before. Now backpedalling, huh? Anyway, you should stop comparing perofmance in games on PC and consoles 1 :1. Never was like that, never will be.

 

[Gaiff]

You meant 9060XT? In brute numbers 9060XT has 50% more TF compared to PS5 Pro in FP32 and it is a RDNA4 card. It has 16GB/320GB/s and 32MB of Infinity cache. Not so similar. At all. So, like i've said, you still treating games performance on PC and consoles 1 :1.


It is and always will be. As Carmack said that 15+ years ago. Still applying this day.

VOPD TFLOPs. Useless in games.

 

[Gaiff]

You are dev like him? You understand how console performance works, right? You are a dev over 30+ years of experience on consoles, right?


2x TF numbers is used since 10 years ago. LOL
It isn't fake numbers. 9060 is much more power efficient GPU compared to 6800 in every way ( despite 6800 has a bigger bandwidth), and much higher clock. Since it is RDNA 4 designed GPU, it has FSR4 support ( 6800 doesn't have it, at least not supported by AMD) And it has better RT performance in games.
You used 9060XT vs Pro in post before. Now backpedalling, huh? Anyway, you should stop comparing perofmance in games on PC and consoles 1 :1. Never was like that, never will be.

The numbers are useless within a game dev context. Mark Cerny even addressed it when he answered that the PS5 has 16 TFLOPs and not 32 TFLOPs with dual-issue as was previously speculated. 2x TF hasn't been used for 10 years, what are you talking about? This way to calculate them dates back from RDNA3 with AMD and Ampere for NVIDIA.

It's explained here: https://chipsandcheese.com/p/microbenchmarking-amds-rdna-3-graphics-architecture

Compared to RDNA 2, RDNA 3 obviously has a large advantage in compute throughput. After all, it has a higher WGP count. But potential increases in compute throughput go beyond that, because RDNA 3's SIMDs gain a limited dual issue capability. Certain common operations can be packaged into a single VOPD (vector operation, dual) instruction in wave32 mode. In wave64 mode, the SIMD will naturally try to start executing a 64-wide wavefront over a single cycle, provided the instruction can be dual issued.

A RDNA 3 VOPD instruction is encoded in eight bytes, and supports two sources and one destination for each of the two operations. That excludes operations that require three inputs, like the generic fused multiply add operation. Dual issue opportunities are further limited by available execution units, data dependencies, and register file bandwidth.

 

[Hyper_hyper]

2x TF hasn't been used for 10 years, what are you talking about? This way to calculate them dates back from RDNA3 with AMD and Ampere for NVIDIA.

You haven't followed. 2x TF was mentioned 10 years ago, not since RDNA 3.

 

[Gaiff]

You haven't followed. 2x TF was mentioned 10 years ago, not since RDNA 3.

What the fuck are you talking about man? You said the 9060 XT has 50% more compute than the PS5 Pro's GPU, that's including dual-issue, which has limited use, none of which pertains to gaming. Those computer numbers 2x than the real life numbers have only been started to be used with RDNA3.

 

[MisterXDTV]

Mark Cerny even addressed it when he answered that the PS5 has 16 TFLOPs and not 32 TFLOPs with dual-issue as was previously speculated. 2x TF hasn't been used for 10 years, what are you talking about? This way to calculate them dates back from RDNA3 with AMD and Ampere for NVIDIA.

True but he also said he couldn't use that metric because PS5 Pro GPU is an hybrid that is based on a RDNA 2 "core" to keep "compatibility" with PS5 software

So to make the comparison 1:1 he used, fairly, the same metric as the original PS5

10.3 TF vs 16.7 TF

 

[Hyper_hyper]

You said the 9060 XT has 50% more compute than the PS5 Pro's GPU, that's including dual-issue, which has limited use,

Yes, i did. And did it properly, yes. 25.x TF for 9060XT vs. 16.x TF for PS5 Pro.

Those computer numbers 2x than the real life numbers have only been started to be used with RDNA3.

Looks like you haven't followed. Like I've said, 2x TF was already mentioned 10 years ago, not since RDNA 3

 

[Gaiff]

Yes, i did. And did it properly, yes. 25.x TF for 9060XT vs. 16.x TF for PS5 Pro.

That's 25 with dual-issue, compute that is not usable in gaming. What's so hard to understand?

Looks like you haven't followed. Like I've said, 2x TF was already mentioned 10 years ago, not since RDNA 3

They calculated with dual-issue 10 years ago?

 

[Bojji]

33TF RDNA3 GPU has exactly the same power as 16.5TF RDNA2 GPU. AMD uses those bullshit numbers since 2022.

There was nothing like that before.

 

[Hyper_hyper]

That's 25 with dual-issue, compute that is not usable in gaming. What's so hard to understand?

What is so hard to understand that FP32 is the real TFLOPS?? PS5 Pro in FP32 is 16.7 TF single precision

They calculated with dual-issue 10 years ago?

You don't remember FP32 on PS4 Pro which was AMD's GPU??

 

[Gaiff]

What is so hard to understand that FP32 is the real TFLOPS?? PS5 Pro in FP32 is 16.7 TF. Dual issue/single precision



You don't remember FP32 on PS4 Pro which was AMD's GPU??

Single precision and dual-issue aren't the same thing at all.

 

[IbizaPocholo]

• PS5 and Xbox Series X only have a 60fps display mode without Ray Tracing.
• PS5 Pro is the only console with Ray Tracing at 60fps. We can disable it to have a variable frame rate up to 120fps.
• The PC version runs at maximum settings.
• The most noticeable differences are on PC thanks to the use of Path Tracing with more realistic global lighting, more accurate shadows, and better reflections.

 

[Hyper_hyper]

Single precision and dual-issue aren't the same thing.

Yes, they aren't. Mixed it. My bad. But anyway, FP32 is used 10 years ago

 

[Gaiff]

Yes, they aren't. Mixed it. My bad. But anyway, FP32 is used 10 years ago

FP32 has been used for way more than 10 years. It's been the standard to measure compute for decades, since the start of programmable shaders.

What I'm telling you is that the way RDNA3 reaches such gaudy FP32 numbers is through dual-issue, something that isn't leveraged in gaming and has been introduced with RDNA3. Thus, those 25 TFLOPs you see don't mean a thing in a gaming context. So it isn't a FP32 vs non-FP32 thing. It's a FP32 with dual-issue vs FP32 without it. As such, the 9060 XT doesn't actually have 50% more compute performance than the PS5 Pro, since dual-issue isn't used anyway.

 

[Hyper_hyper]

As such, the 9060 XT doesn't actually have 50% more compute performance than the PS5 Pro, since dual-issue isn't used anyway.

I've used official specs,yes. From their website. Totally 1 :1. Yes, 9060XT has 50% more TF in single precision just like PS5 Pro GPU has 16.7 TF in single precision. Don't blame me.

 

[Gaiff]

I've used official specs,yes. From their website. Totally 1 :1. Yes, 9060XT has 50% more TF in single precision just like PS5 Pro GPU has 16.7 TF in single precision. Don't blame me.

Now you're just a troll or willfully ignorant. Those numbers aren't 1 to 1. Dual-issue isn't used in games. It's therefore useless in this comparison.

 

[viveks86]

Grace looks like a goblin without path-tracing.

Quite true actually. RT improves on raster quite a bit so she looks...ok. Still very videogame-y. It appears that all specular reflections on her jacket are gone without RT

But with PT, she looks fantastic. Like CGI. Probably the biggest difference I've seen with PT in this game

And then there is this with complex lighting that needs to bounce around, which only looks good with PT

 

[Hyper_hyper]

Now you're just a troll or willfully ignorant. Those numbers aren't 1 to 1. Dual-issue isn't used in games. It's therefore useless in this comparison

Who ever mentioned that 9060XT 25.6TF is reallu dual issue ( i've said btw. that I've mixed it) of 16.7 TF for PS5 Pro?

Em.... me trolling using official specs? Ok. PS5 Pro isn't 16.7 TF in FP 32 nor 9060XT isn't 25.6 TF in FP32. Happy now? PS5 Pro is 8.35 TF then.:/

 

[Gaiff]

Em.... me trolling using official specs? Ok. PS5 Pro isn't 16.7 TF in FP 32 nor 9060XT isn't 25.6 TF in FP32. Happy now? PS5 Pro is 8.35 TF then.:/

Pro doesn't use dual-issue, so why would you halve the compute performance? That result is what you get in the real world. Yeah, so a troll it is.

 

[Badlucktroll]

 

[Hyper_hyper]

Pro doesn't use dual-issue, so why would you halve the compute performance? That result is what you get in the real world. Yeah, so a troll it is.

Never ever claimed it is a dual issue since PS5 Pro doesn't have that nor 9060XT. I've used official specs. I've used brute numbers in comparison 1 : 1. And I've used this emoji :/ since you've asked why i halved numbers. LOL

Provide your numbers since maybe you know better. :/

 

[Gaiff]

Never ever claimed it is a dual issue since PS5 Pro doesn't have that nor 9060XT. I've used official specs. I've used brute numbers in comparison 1 : 1. And I've used this emoji :/ since you've asked why i halved numbers. LOL

Yes, the 9060 XT has dual-issue compute. Every AMD card since RDNA3 does.

 

[Bojji]

Yes, the 9060 XT has dual-issue compute. Every AMD card since RDNA3 does.

That's why 7700XT

Has almost the same performance as 6800

And 9060XT

And 9060XT is more or less on par with PS5 Pro:

That shows that what Carmack said is bullshit. He developed Rage for consoles and had to drop all dynamic lighting from Doom 3 to make it 60fps. There was no hidden secret performance there...

Modern GPUs perform very close to parts found in consoles H Hyper_hyper

 

[Hyper_hyper]

That's why 7700XT

Has almost the same performance as 6800

And 9060XT

And 9060XT is more or less on par with PS5 Pro:

That shows that what Carmack said is bullshit. He developed Rage for consoles and had to drop all dynamic lighting from Doom 3 to make it 60fps. There was no hidden secret performance there...

Modern GPUs perform very close to parts found in consoles H Hyper_hyper

You are comparing 7700XT RDNA 3 vs 9060XT which is RDNA 4. I will just switch 6800 with 7700XT

9060 is much more power efficient GPU compared to 7700 in every way ( despite 7700 has a bigger bandwidth), and much higher clock. Since it is RDNA 4 designed GPU, it has FSR4 support ( 7700 doesn't have it, at least not supported by AMD) And it has better RT performance in games.

• RX 7700 XT (RDNA 3): Rated at 245W maximum Total Board Power (TBP).
• RX 9060 XT (RDNA 4): Reported at 150W - 160W maximum TBP.

Also, you are comparing 9060XT 50% more TF to PS5 Pro GPU which proves Carmack theory from 15+ years ago.

Now show comparison from 7700 ( or 6800) vs PS5 Pro

 

[ThroatPie]

 

[Gaiff]

You are comparing 7700XT RDNA 3 vs 9060XT which is RDNA 4. I will just switch 6800 with 7700XT

9060 is much more power efficient GPU compared to 7700 in every way ( despite 7700 has a bigger bandwidth), and much higher clock. Since it is RDNA 4 designed GPU, it has FSR4 support ( 7700 doesn't have it, at least not supported by AMD) And it has better RT performance in games.

• RX 7700 XT (RDNA 3): Rated at 245W maximum Total Board Power (TBP).
• RX 9060 XT (RDNA 4): Reported at 150W - 160W maximum TBP.

Also, you are comparing 9060XT 50% more TF to PS5 Pro GPU which proves Carmack theory from 15+ years ago.

Now show comparison from 7700 ( or 6800) vs PS5 Pro

Bojji yep, got ourselves a troll. Just ignore him. Not worth engaging.

 

[Bojji]

You are comparing 7700XT RDNA 3 vs 9060XT which is RDNA 4. I will just switch 6800 with 7700XT

9060 is much more power efficient GPU compared to 7700 in every way ( despite 7700 has a bigger bandwidth), and much higher clock. Since it is RDNA 4 designed GPU, it has FSR4 support ( 7700 doesn't have it, at least not supported by AMD) And it has better RT performance in games.

• RX 7700 XT (RDNA 3): Rated at 245W maximum Total Board Power (TBP).
• RX 9060 XT (RDNA 4): Reported at 150W - 160W maximum TBP.

Also, you are comparing 9060XT 50% more TF to PS5 Pro GPU which proves Carmack theory from 15+ years ago.

Now show comparison from 7700 ( or 6800) vs PS5 Pro

Me and Gaiff talking to you:

 

[Gaiff]

Quite true actually. RT improves on raster quite a bit so she looks...ok. Still very videogame-y. It appears that all specular reflections on her jacket are gone without RT

But with PT, she looks fantastic. Like CGI. Probably the biggest difference I've seen with PT in this game

And then there is this with complex lighting that needs to bounce around, which only looks good with PT

I really wanted to see a PT 30fps mode on the Pro. They got 120fps, come on...

 

[Hyper_hyper]

Bojji yep, got ourselves a troll. Just ignore him. Not worth engaging.

Me and Gaiff talking to you:

Both of you ignoring the very fact that PS5 Pro has much better RT capabilities than these 6800 and 7700. 9060 is comparable to PS5 Pro GPU, in RT especially. Both of you can laugh as much as you want.

 

[Zathalus]

Also, you are comparing 9060XT 50% more TF to PS5 Pro GPU which proves Carmack theory from 15+ years ago.

9060XT is 12.82TF. the 25.6TF is derived from using VOPD. VOPD = Vector Operation Dual-Issue. It's an RDNA 3/4 feature that allows a single SIMD unit to execute two vector instructions in one cycle. But games never meet the conditions needed for VOPD to trigger. It requires instructions to be independent and neatly pairable, and support from the compiler, something you basically never get in games.

So the real TF figure for the 9060 XT that applies to games is half of what AMD advertises. You can slap the same VOPD on the PS5 Pro and gains will be extremely tiny.

Note, this is different than the dual FP32 units that Nvidia has had since Ampere.

 

[Bojji]

Both of you ignoring the very fact that PS5 Pro has much better RT capabilities than these 6800 and 7700. 9060 is comparable to PS5 Pro GPU, in RT especially. Both of you can laugh as much as you want.

Yeah, why not. Pro has RDNA4 RT performance. That still makes it perform close to 9060XT like I said already... and this GPU is 25TF vs. 16TF on Pro.

Only confirms that RDNA3/4 TF numbers are bullshit.

 

[Mr Moose]

Both of you ignoring the very fact that PS5 Pro has much better RT capabilities than these 6800 and 7700. 9060 is comparable to PS5 Pro GPU, in RT especially. Both of you can laugh as much as you want.

And 9060XT is more or less on par with PS5 Pro

 

[Hyper_hyper]

9060XT is 12.82TF. the 25.6TF is derived from using VOPD. VOPD = Vector Operation Dual-Issue. It's an RDNA 3/4 feature that allows a single SIMD unit to execute two vector instructions in one cycle. But games never meet the conditions needed for VOPD to trigger. It requires instructions to be independent and neatly pairable, and support from the compiler, something you basically never get in games.

Official from AMD? Or any other tech site?

 

[Gaiff]

Official from AMD? Or any other tech site?

Yes.

https://docs.amd.com/v/u/en-US/rdna3-shader-instruction-set-architecture-feb-2023_0

The VOPD instruction encoding allows a single shader instruction to encode two separate VALU operations
that are executed in parallel. The two operations must be independent of each other. This instruction has
certain restrictions that must be met - hardware does not function correctly if they are not. This instruction
format is legal only for wave32. It must not be used by wave64's. It is skipped for wave64.
The instruction defines 2 operations, named "X" and "Y", each with their own sources and destination VGPRs.
The two instructions packed into this one ISA are referred to as OpcodeX and OpcodeY.
• OpcodeX sources data from SRC0X (a VGPR, SGPR or constant), and SRC1X (a VGPR);
• OpcodeY sources data from SRC0Y (a VGPR, SGPR or constant), and SRC1Y (a VGPR).
The two instructions in the VOPD are executed at the same time, so there are no races between them if one
reads a VGPR and the other writes the same VGPR. The 'read' gets the old value.
Restrictions:
• Each of the two instructions may use up to 2 VGPRs
• Each instruction in the pair may use at most 1 SGPR or they may share a single literal
◦ Legal combinations for the dual-op: at most 2 SGPRs, or 1 SGPR + 1 literal, or share a literal.
• SRC0 can be either a VGPR or SGPR (or constant)
• VSRC1 can only be a VGPR
• Instructions must not exceed the VGPR source-cache port limits
◦ There are 4 VGPR banks (indexed by SRC[1:0]), and each bank has a cache
◦ Each cache has 3 read ports: one dedicated to SRC0, one dedicated to SRC1 and one for SRC2
▪ A cache can read all 3 of them at once, but it can't read two SRC0's at once (or SRC1/2).
◦ SRCX0 and SRCY0 must use different VGPR banks;
◦ VSRCX1 and VSRCY1 must use different banks.
▪ FMAMK is an exception : V = S0 + K * S1 ("S1" uses the SRC2 read port)
◦ If both operations use the SRC2 input, then one SRC2 input must be even and the other SRC2 input
must be odd. The following operations use SRC2: FMAMK_F32 (second input operand);
DOT2ACC_F32_F16, DOT2ACC_F32_BF16, FMAC_F32 (destination operand).
◦ These are hard rules - the instruction does not function if these rules are broken
• The pair of instructions combined have the following restrictions:
◦ At most one literal constant, or they may share the same literal
◦ Dest VGPRs: one must be even and the other odd
◦ The instructions must be independent of each other
• Must not use DPP
• Must be wave3

 

[cripterion]

AS i said in another topic: Japan carrying the industry

Watching Capcom transforming during the past decade has been amazing to witness. Instead of chasing trends they decided to stick to their guns and do what they were always good at in the first place. And look at it paying off.

It's great to see a decent current gen game competent on a technical level but Japan carrying the industry? By their own admittance they can't keep up with China.

Need I remind you how Capcom managed Dragon's Dogma 2 ? And shit the bed with MH : Wilds (performance and DLC fest).

 

[Hyper_hyper]

Yes.

https://docs.amd.com/v/u/en-US/rdna3-shader-instruction-set-architecture-feb-2023_0

Not that. I've found that. I want to see that VOPD 12 TF is applied to 9060XT for example, because it is not in official specs for 9060XT

 

[Gaiff]

Not that. I've found that. I want to see that VOPD 12 TF is applied to 9060XT for example, because it is not in official specs for 9060XT

The documentation tells you what is required to leverage VPOD with the very first line I posted and games don't meet those requirements. Since it "doubles" your FP32 compute throughput, you halve it if said conditions aren't met, or do you need them to spell it out for you more clearly?

 

[Zathalus]

Official from AMD? Or any other tech site?

Microbenchmarking AMD’s RDNA 3 Graphics Architecture

Editor’s Note (6/14/2023): We have a new article that reevaluates the cache latency of Navi 31, so please refer to that article for some new latency data.

chipsandcheese.com

You can read up about VOPD there.

AMD Technical Information Portal

docs.amd.com

The ISA clearly lists the limitations regarding VOPD.

Another point of simple evidence. The 6950XT and 9060XT are very similar TF wise (as reported by AMD), but the 6950XT being 25% faster in regular compute workloads, despite being two generations older and RDNA4 having a whole host of GPU architecture improvements. A pretty big clue on the uselessness of VOPD in gaming.

 

[FireFly]

Not that. I've found that. I want to see that VOPD 12 TF is applied to 9060XT for example, because it is not in official specs for 9060XT

"The VOPD instruction encoding allows a single shader instruction to encode two separate VALU operations that are executed in parallel. The two operations must be independent of each other. This instruction has certain restrictions that must be met - hardware does not function correctly if they are not. This instruction format is legal only for wave32. It must not be used by wave64's."

AMD Technical Information Portal

docs.amd.com

 

[PaintTinJr]

That's why 7700XT

Has almost the same performance as 6800

And 9060XT

And 9060XT is more or less on par with PS5 Pro:

That shows that what Carmack said is bullshit. He developed Rage for consoles and had to drop all dynamic lighting from Doom 3 to make it 60fps. There was no hidden secret performance there...

Modern GPUs perform very close to parts found in consoles H Hyper_hyper

The GPU being a discrete part rather than an APU is the only reason a RX 9060XT has the die space for a 30MB L3 cache, when the the Pro merely has an 4MB L2 cache, which the 9060XT also has.

LLC is a major leveller or booster in this situation, especially in situations where the 9060XT could be compute bound without using the Dual Issue. But the biggest advantage the 9060XT has, that makes it nothing like the Pro APU's GPU, is that it can clock to +3000Mhz as a discrete GPU chip, which in turn gives an extra 50Gpixels or rather a whole XSS in fillrate more.

So no, unless you are putting a PC APU with the same area restrictions on the chip - eliminating the L3 and needing conservative clocking - the parts are nothing alike in benchmarking and it doesn't mean Carmack was wrong, even if it is irrelevant given the lack of optimisation on console today.

 

[Bojji]

The GPU being a discrete part rather than an APU is the only reason a RX 9060XT has the die space for a 30MB L3 cache, when the the Pro merely has an 4MB L2 cache, which the 9060XT also has.

LLC is a major leveller or booster in this situation, especially in situations where the 9060XT could be compute bound without using the Dual Issue. But the biggest advantage the 9060XT has, that makes it nothing like the Pro APU's GPU, is that it can clock to +3000Mhz as a discrete GPU chip, which in turn gives an extra 50Gpixels or rather a whole XSS in fillrate more.

So no, unless you are putting a PC APU with the same area restrictions on the chip - eliminating the L3 and needing conservative clocking - the parts are nothing alike in benchmarking and it doesn't mean Carmack was wrong, even if it is irrelevant given the lack of optimisation on console today.

Carmack said:

consoles offer twice the performance of a PC for the same "paper spec"

And this is bullshit. 6800 has the same number of CUs, almost the same clock and pretty much on par TF numbers. But less raw memory BW (512GB/s), yet it performs very close to PS5 Pro:

L3 cache HELPS with memory BW limited scenarios but it can't make up to 2x performance bullshit from Carmack. Age of exotic console hardware is over and now we have x86 PCs in small packages that surprisingly perform very similar to standard PC parts.

 

[yogaflame]

Looks like this 2026 and onwards, ps5 pro will be truly worth it. It will be a cheaper alternative for those who want decent tech but more affordable than expensive pc rig.

 

[Gaiff]

Carmack said:



And this is bullshit. 6800 has the same number of CUs, almost the same clock and pretty much on par TF numbers. But less raw memory BW (512GB/s), yet it performs very close to PS5 Pro:

L3 cache HELPS with memory BW limited scenarios but it can't make up to 2x performance bullshit from Carmack. Age of exotic console hardware is over and now we have x86 PCs in small packages that surprisingly perform very similar to standard PC parts.

Didn't Carmack say that like 20 years ago? I don't think it applies with the multiplatform environment of today anymore.

 

[Bojji]

Didn't Carmack say that like 20 years ago? I don't think it applies with the multiplatform environment of today anymore.

He said it ~PS3 times I think. Even then it wasn't true.

 

[PaintTinJr]

Carmack said:



And this is bullshit. 6800 has the same number of CUs, almost the same clock and pretty much on par TF numbers. But less raw memory BW (512GB/s), yet it performs very close to PS5 Pro:

L3 cache HELPS with memory BW limited scenarios but it can't make up to 2x performance bullshit from Carmack. Age of exotic console hardware is over and now we have x86 PCs in small packages that surprisingly perform very similar to standard PC parts.

He said that when the PS3/360 to PS4/X1 transition happened, and it was fair comment back then with him coming off the back of optimising on 360/PS3 where each had discrete CPUs and GPUs.

Now you are holding up discrete parts with twice the combined memory paired with CPUs above the consoles with far more cache and with unoptimized games and holding up his comment out of context.

He wouldn't make that same comment now, because it isn't true now mainly because of the software and area constraints on an APU within that wattage budget.

I clearly showed your 9060XT vs a Pro GPU was completely unfair with the discrete part having the most vital spec: fillrate as an entire console from this gen extra.