sncvsrtoip
Member
yes int8Using Int-8 machine inference? By the way similar results as 3090 would be incredibly good compared to the RDNA2 potatoes we have now, even worse on RDNA1 PS5 totally lacking int-8.
yes int8Using Int-8 machine inference? By the way similar results as 3090 would be incredibly good compared to the RDNA2 potatoes we have now, even worse on RDNA1 PS5 totally lacking int-8.
Perfect. Raster is the last thing we want to be substantially improved from PS5. The improvements will be enough to render native 1080p 60fps in those really demanding games (currently running at 720p).
I didn't believe them eitherOf course the day after I say I think these specs are bullshit.......Tom Henderson confirms them.
I didn't believe them either
Whats worse means MLiD got this right and he went from a booger eater to someone I kind of have to pay attention to
If Tom Henderson says it's real it is.I didn't think there would be a Pro,but I'm day one whenever it releasesOk I'm back to believing the Pro is real lol
I assume Astro's Playroom is going to get an 8K patch for the Pro?
Tom Henderson is saying from what he heard that the devs are really excited about this AI upscaling solution. So that's a good thing.I don't know... Raster is and will be the most important metric for GPU power for quite some time. The most exciting thing about this console is new upscaling tech but developers have to care enough to implement it.
Why didn’t you believe him? We’ve been hearing about those months before he made this video.I didn't believe them either
Whats worse means MLiD got this right and he went from a booger eater to someone I kind of have to pay attention to
The only thing that threw me in all honesty was the overall TF number as I have been told it was basically half that number, for quite sometime nowOnce I saw the paper screenshots he had, it was clear it was right on to something
He even specified the PlayStation Spectral Super Resolution Upscaling
Only an idiot would make that up only to be proven wrong
See aboveWhy didn’t you believe him? We’ve been hearing about those months before he made this video.
The only thing that threw me in all honesty was the overall TF number as I have been told it was basically half that number, for quite sometime now
This coming from the same person (I totally trust) that gave me the 50% raster performance and 2x TF performance
It is a little odd that is 3x the TF number but only 45% boost in raster?
Am I looking at something wrong here?
See above
Well, half that would mean you were told the real-world compute performance and not dual-issue. 17 TFLOPs would be how we'd calculate RDNA2 or older archs' compute because their SIMDs don't have dual-issue compute capabilities like RDNA3.The only thing that threw me in all honesty was the overall TF number as I have been told it was basically half that number, for quite sometime now
This coming from the same person (I totally trust) that gave me the 50% raster performance and 2x TF performance
It is a little odd that is 3x the TF number but only 45% boost in raster?
Am I looking at something wrong here?
See above
I didn't believe them either
Whats worse means MLiD got this right and he went from a booger eater to someone I kind of have to pay attention to
No. I meant to post earlier. Your number is right in line with MiLD it's just different nomenclature for RDNA 3 architecture. The comparable 16TF number is what you want to focus on. Roughly same number you’ve been saying. Thanks for all your service throughout the crazy speculative period. You can rest now, good sir.
See, this shit is way above my paygradeWell, half that would mean you were told the real-world compute performance and not dual-issue. 17 TFLOPs would be how we'd calculate RDNA2 or older archs' compute because their SIMDs don't have dual-issue compute capabilities like RDNA3.
So the information you got sounds accurate.
I wonder if the "45% Raster Boost" is an oversimplification of the GPU front end being 50% bigger (96 ROP vs vs 64 ROPS) minus a small clock reduction.The only thing that threw me in all honesty was the overall TF number as I have been told it was basically half that number, for quite sometime now
This coming from the same person (I totally trust) that gave me the 50% raster performance and 2x TF performance
It is a little odd that is 3x the TF number but only 45% boost in raster?
Am I looking at something wrong here?
See above
did we ever get confirmation Cerny is leading this effort? i know its probably 90% chance but just cuirous.
Pretty disingenuous when his leak is the first to show dev slides. Clearly not a case of throwing mud at the wall until something sticks. Its not his fault if people can't differentiate between speculation and hard information. Whenever I've watched him it has been pretty clear where his leaks stop and his speculation begins, and where it was what he was being told vs what he has seen etc. My advice, pay attention to the detail. He makes his living off this stuff so of course he will stuff content, but that doesn't mean there isn't value in amongst it all.Even a broken clock is right twice a day...
With that said, the confusing part is the low clock speed. 2.18GHz sounds incredibly low. I was expecting it to be at least 2.3GHz since Cerny loves his high clocks. In fact, I was expecting it to be closer to 2.4GHz even.See, this shit is way above my paygrade
Thank you guys, I feel better, need to erase my wtf text I was sending
Pretty disingenuous when his leak is the first to show dev slides. Clearly not a case of throwing mud at the wall until something sticks. Its not his fault if people can't differentiate between speculation and hard information. Whenever I've watched him it has been pretty clear where his leaks stop and his speculation begins, and where it was what he was being told vs what he has seen etc. My advice, pay attention to the detail. He makes his living off this stuff so of course he will stuff content, but that doesn't mean there isn't value in amongst it all.
With that said, the confusing part is the low clock speed. 2.18GHz sounds incredibly low. I was expecting it to be at least 2.3GHz since Cerny loves his high clocks. In fact, I was expecting it to be closer to 2.4GHz even.
This isn't as wide and as slow as the SX but definitely not as narrow and as fast as the PS5. It's more balanced. 45% performance increase in rasterization isn't much and this could be attributed to the lower than expected clocks. 60 CUs at 2.18GHz means a lot of the components affected by clocks won't see that much of an improvement which I find odd.
Kind of holding out hope that those specs aren't final and that Sony will bump up the clocks. Otherwise, RT and the new upscaling solution will have to do the heavy lifting, which is fine, but it never hurt to have a lot of raw power just to make sure incompetent devs don't fuck it up.
That's how I arrived at 16.75 TFLOPs or whatever that figure was. Could probably rework out some napkin table maths and get a higher number but I think I also saw Kepler mention it?Are we sure the 60 CUs are all active?
If they disable some for yields, the clock will be higher
I think the 33 number comes because of "dual issue". I think it is up to the devs to use it (or maybe??? just recompile for PS5 Pro). Divide the 33 by 2 to compare apples to apples with the base PS5.The only thing that threw me in all honesty was the overall TF number as I have been told it was basically half that number, for quite sometime now
This coming from the same person (I totally trust) that gave me the 50% raster performance and 2x TF performance
It is a little odd that is 3x the TF number but only 45% boost in raster?
Am I looking at something wrong here?
See above
This isn't just guessing numbers. He has the Sony docs to prove it.Even a broken clock is right twice a day...
I like your numbers as thats the range I hinted at for quite awhile now as this is the numbers I was gettingThat's how I arrived at 16.75 TFLOPs or whatever that figure was. Could probably rework out some napkin table maths and get a higher number but I think I also saw Kepler mention it?
That's how I arrived at 16.75 TFLOPs or whatever that figure was. Could probably rework out some napkin table maths and get a higher number but I think I also saw Kepler mention it?
With that said, the confusing part is the low clock speed. 2.18GHz sounds incredibly low. I was expecting it to be at least 2.3GHz since Cerny loves his high clocks. In fact, I was expecting it to be closer to 2.4GHz even.
This isn't as wide and as slow as the SX but definitely not as narrow and as fast as the PS5. It's more balanced. 45% performance increase in rasterization isn't much and this could be attributed to the lower than expected clocks. 60 CUs at 2.18GHz means a lot of the components affected by clocks won't see that much of an improvement which I find odd.
Kind of holding out hope that those specs aren't final and that Sony will bump up the clocks. Otherwise, RT and the new upscaling solution will have to do the heavy lifting, which is fine, but it never hurt to have a lot of raw power just to make sure incompetent devs don't fuck it up.
64CU, 4 is disabledAre we sure the 60 CUs are all active?
64CU, 4 is disabled
Looks at SX, there 52CU, 320 bus etc, console gpu not related DS variantThat would require a chip bigger than the current Navi 32, I'm not so sure
The only thing that threw me in all honesty was the overall TF number as I have been told it was basically half that number, for quite sometime now
This coming from the same person (I totally trust) that gave me the 50% raster performance and 2x TF performance
It is a little odd that is 3x the TF number but only 45% boost in raster?
Am I looking at something wrong here?
See above
This would make a lot more sense and previous leaks did suggest 60 CUs with 6 disabled for improved yields.If we take the 7800 XT and the 7700 XT as reference:
60 CUs @ 2.18 Ghz = 33.5 TF
54 CUs @ 2.42 Ghz = 33.5 TF
Was this confirmed? Well, not "confirmed" but I don't recall seeing the CU count in the leaks.64CU, 4 is disabled
Was this confirmed? Well, not "confirmed" but I don't recall seeing the CU count in the leaks.
If we take the 7800 XT and the 7700 XT as reference:
60 CUs @ 2.18 Ghz = 33.5 TF
54 CUs @ 2.42 Ghz = 33.5 TF
I don't see much use in comparing to RDNA 3 because of the RT and AI upscaling, which are the highlights of the Pro console. As I said elsewhere, from a final image quality/output perspective, I'd be surprised if even one game with RT implementation looks better on flagship RDNA 3 compared to PS5 Pro.
Kepler?Was this confirmed? Well, not "confirmed" but I don't recall seeing the CU count in the leaks.
That would require a chip bigger than the current Navi 32, I'm not so sure
The CU count?Kepler?
YesThe CU count?
You're putting a lot of pressure on that upscaling. The 7900 XTX would still be about 80% faster than the Pro if those leaks are true. There will be no path-traced games where the PS5 Pro would have (maybe relatively speaking) an enormous advantage. They will all be hybrid workloads where the 7900 XTX is to the Pro what the 3080 is to the PS5. Even assuming Sony's upscaling tech is better than FSR2, the 7900 XTX would still start from a much higher base resolution.I don't see much use in comparing to RDNA 3 because of the RT and AI upscaling, which are the highlights of the Pro console. As I said elsewhere, from a final image quality/output perspective, I'd be surprised if even one game with RT implementation looks better on flagship RDNA 3 compared to PS5 Pro.
I'll have to agree with you.I saw it as a rumor a couple of months ago, but never actually confirmed
But the point is with PS4 being 18 CUs, PS4 Pro and PS5 being 36 CUs...
54 CUs makes too much sense to me for PS5 Pro. It's just a clever approach
Yeah something's not rightThe only thing that threw me in all honesty was the overall TF number as I have been told it was basically half that number, for quite sometime now
This coming from the same person (I totally trust) that gave me the 50% raster performance and 2x TF performance
It is a little odd that is 3x the TF number but only 45% boost in raster?
Am I looking at something wrong here?
See above