Playstation 5 specs leaked! (Project Aerial)

  • 123 results
  • 1
  • 2
  • 3
Avatar image for foxhound_fox
#101 Posted by foxhound_fox (97964 posts) -

Always hilarious when people jerk off over numbers on paper. The PS3 proved to us time and time again that numbers on paper are meaningless, it's what the developers can do with them that counts.

Avatar image for emgesp
#102 Posted by emgesp (7832 posts) -
@foxhound_fox said:

Always hilarious when people jerk off over numbers on paper. The PS3 proved to us time and time again that numbers on paper are meaningless, it's what the developers can do with them that counts.

Numbers are not meaningless. They don't paint the whole picture, but they do give you an idea of potential.

Avatar image for raining51
#103 Posted by Raining51 (1124 posts) -

I feel like allt he big parties are just waiting for the other one to go first and then react to that... it's sad they should definitely be motivated to create something i unqiue that stands out.

I think video games could really end up being more in the excitement of Apple and Amazon and such as opposed to the traditional giants.... with the possible exception of Sony.

Avatar image for foxhound_fox
#104 Posted by foxhound_fox (97964 posts) -

@emgesp said:
@foxhound_fox said:

Always hilarious when people jerk off over numbers on paper. The PS3 proved to us time and time again that numbers on paper are meaningless, it's what the developers can do with them that counts.

Numbers are not meaningless. They don't paint the whole picture, but they do give you an idea of potential.

Okay, potential is great. 4D graphics @ 120 fps.

Avatar image for Martin_G_N
#105 Posted by Martin_G_N (1921 posts) -

@foxhound_fox: The PS3 was a different architecture, and hard for devs to utilize. WIth these X86 AMD specs it's pretty much what you see is what you get. I think these specs could be possible. Remember, with the PS3 Sony took a $300 loss with each PS3 sold, at least. The PS4 was their first console to have a profit at launch. If they are going to release a $400 console with a profit in 2020, we're looking at an overclocked X1X. That won't cut it for a next gen console.

Avatar image for Random_Matt
#106 Posted by Random_Matt (4244 posts) -

Not really possible to change the 64CU limit, unless AMD are prepared to throw tons of money at it. The whole rumour is fake, do not believe they will up the clock speed either.

Avatar image for foxhound_fox
#107 Posted by foxhound_fox (97964 posts) -

@Martin_G_N said:

@foxhound_fox: The PS3 was a different architecture, and hard for devs to utilize. WIth these X86 AMD specs it's pretty much what you see is what you get. I think these specs could be possible. Remember, with the PS3 Sony took a $300 loss with each PS3 sold, at least. The PS4 was their first console to have a profit at launch. If they are going to release a $400 console with a profit in 2020, we're looking at an overclocked X1X. That won't cut it for a next gen console.

Yes, but it was touted by cows for ages as "superior" because of the numbers on the paper being bigger than the Xbox 360.

Avatar image for Pedro
#108 Posted by Pedro (34910 posts) -

@tormentos: WOW! You took all that time to write stuff equivalent to Ron aka bunch of meaningless stuff that has nothing to do with what you are responding to. Good job.

Avatar image for Grey_Eyed_Elf
#109 Posted by Grey_Eyed_Elf (6453 posts) -

@ronvalencia: Again the reason why I say "lucky" to get 10-11... To get 12 TFLOPs like you attempted is based on the idea that AMD is capable of dropping the TPD and heat involved with GCN, remember that its a APU so it won't just be the GPU with direct contact to a cooling solution the entire CPU/GPU solution is on one die and the cooling solution would need to be substantial to get those frequencies with a GCN chip along with a Zen CPU with 16 threads at 3GHz cool.

The theoretical approach of getting 12TFLOPS with GCN at a 200w or lower target is possible but in a console APU in a chassis that won't have great cooling expecting high frequencies from the GPU would be miracle work.

If AMD could get 12TFLOPs from a 150w GPU based on GCN they wouldn't have released a GPU that runs at 300w at 13.8 TFLOPs and costs $700.

Theory speculation is great but the chances of it happening are slim, 10-11 would be the safe area of expectations... The probability of getting 12+ TFLOPS is the same as getting under 10 TFLOPS if they decided to increase VRAM speeds and CPU speeds rather than drop more on the GPU core clock for a more balance system.

Avatar image for ronvalencia
#110 Edited by ronvalencia (28060 posts) -

@Grey_Eyed_Elf said:

@ronvalencia: Again the reason why I say "lucky" to get 10-11... To get 12 TFLOPs like you attempted is based on the idea that AMD is capable of dropping the TPD and heat involved with GCN, remember that its a APU so it won't just be the GPU with direct contact to a cooling solution the entire CPU/GPU solution is on one die and the cooling solution would need to be substantial to get those frequencies with a GCN chip along with a Zen CPU with 16 threads at 3GHz cool.

The theoretical approach of getting 12TFLOPS with GCN at a 200w or lower target is possible but in a console APU in a chassis that won't have great cooling expecting high frequencies from the GPU would be miracle work.

If AMD could get 12TFLOPs from a 150w GPU based on GCN they wouldn't have released a GPU that runs at 300w at 13.8 TFLOPs and costs $700.

Theory speculation is great but the chances of it happening are slim, 10-11 would be the safe area of expectations... The probability of getting 12+ TFLOPS is the same as getting under 10 TFLOPS if they decided to increase VRAM speeds and CPU speeds rather than drop more on the GPU core clock for a more balance system.

My comments are based on TSMC's 16 nm vs 7 nm improve gains with the same logic complexity.

Radeon VII's 1750 Mhz to 1800 Mhz clock speed amounts to factory overclock.

https://www.reddit.com/r/Amd/comments/ao43xl/radeon_vii_insanely_overvolted_undervolting/

Radeon VII is overvolted, console GPU would have little provision for PC overclocking.

RTX 2080 has 215 watts TDP. Undervolting enabled VII to rival RTX 2080's performance per watt.

X1X GPU has per motherboard GPU power delivery tuning.

VII has sloppy/lazy power delivery configuration from AMD i.e. craftsmanship is a little lacking. One wonders if AMD is even trying to compete against NVIDIA with their best craftsmanship.

Your "The theoretical approach of getting 12TFLOPS with GCN at a 200w" statement...

The following graph has 7nm++ 10 percent improvements over 1st generation 7 nm

https://www.anandtech.com/show/13923/the-amd-radeon-vii-review/18

VII 1500Mhz vs V64 at the same 1500Mhz clock speed

VII has IPC improvements with ray-tracing .

Avatar image for Grey_Eyed_Elf
#111 Posted by Grey_Eyed_Elf (6453 posts) -

@ronvalencia said:
@Grey_Eyed_Elf said:

@ronvalencia: Again the reason why I say "lucky" to get 10-11... To get 12 TFLOPs like you attempted is based on the idea that AMD is capable of dropping the TPD and heat involved with GCN, remember that its a APU so it won't just be the GPU with direct contact to a cooling solution the entire CPU/GPU solution is on one die and the cooling solution would need to be substantial to get those frequencies with a GCN chip along with a Zen CPU with 16 threads at 3GHz cool.

The theoretical approach of getting 12TFLOPS with GCN at a 200w or lower target is possible but in a console APU in a chassis that won't have great cooling expecting high frequencies from the GPU would be miracle work.

If AMD could get 12TFLOPs from a 150w GPU based on GCN they wouldn't have released a GPU that runs at 300w at 13.8 TFLOPs and costs $700.

Theory speculation is great but the chances of it happening are slim, 10-11 would be the safe area of expectations... The probability of getting 12+ TFLOPS is the same as getting under 10 TFLOPS if they decided to increase VRAM speeds and CPU speeds rather than drop more on the GPU core clock for a more balance system.

It's based on TSMC's 16 nm vs 7 nm improve gains with the same logic complexity.

I added 7nm++ 10 percent improvements

According to that estimate the Radeon VII which is 7nm GCN with its 1750MHz boost should sit at 160-170w. It does not... Infact its peak is at 330w and average is closer to 270w and it uses 60CU's rather than 64 in the Vega 64.

64 CU's with GCN at 1600MHz with 130w?... is that idle?... That's less power than a RX 570.

You can't be this gullible ron.

GCN is power hungry just look at what 7nm did when you compare the Vega 64 to VII... a die shrink to 7nm and 4 less CU's and its still eating more power than a 1080 Ti. You believe that Navi will somehow make GCN at high CU counts with the same die size as VII drop its power consumption by 1/2?...

Avatar image for ronvalencia
#112 Edited by ronvalencia (28060 posts) -

@Grey_Eyed_Elf:

Refresh your page. I added under voltage information for VII vs RTX 2080.

Avatar image for Grey_Eyed_Elf
#113 Posted by Grey_Eyed_Elf (6453 posts) -

@ronvalencia: Yes more performance but that has nothing to do with power draw. Even under-volted that Radeon VII Still pulls over 200w... at 13.8 TFLOPS. 225w in FC5:

That is just GPU power draw... the TOTAL system has to be under 200w.

You would have to underclock that VII to 1200-1300Mhz and cut 4-8 CU's to get it to run at 130-150w and then you would no longer have a 13.8TFLOP GPU.

Undervolted 225w... Good god ron, your own posts bite you back in the ass. Let me guess Navi will turn that 225 w into 130w and you will still be able to get 12+TFLOPS on GCN?... Please. I'll believe it when I see it, so far AMD is incompetent.

Avatar image for 04dcarraher
#114 Posted by 04dcarraher (23196 posts) -
@Grey_Eyed_Elf said:

Yes more performance but that has nothing to do with power draw. Even under-volted that Radeon VII Still pulls over 200w... at 13.8 TFLOPS. 225w in FC5:

That is just GPU power draw... the TOTAL system has to be under 200w.

You would have to underclock that VII to 1200-1300Mhz and cut 4-8 CU's to get it to run at 130-150w and then you would no longer have a 13.8TFLOP GPU.

Undervolted 225w... Good god ron, your own posts bite you back in the ass. Let me guess Navi will turn that 225 w into 130w and you will still be able to get 12+TFLOPS on GCN?... Please. I'll believe it when I see it, so far AMD is incompetent.

And this is with the fact that Radeon VII is using 7nm node.... I would expect NAVI with a 130 TDP to be around VEGA 56 level of TFLOP performance . Highly doubt 1st gen NAVI is going to be coming close to VEGA 64 or RVII performance wise.

Avatar image for ronvalencia
#115 Edited by ronvalencia (28060 posts) -

@Grey_Eyed_Elf said:

@ronvalencia: Yes more performance but that has nothing to do with power draw. Even under-volted that Radeon VII Still pulls over 200w... at 13.8 TFLOPS. 225w in FC5:

That is just GPU power draw... the TOTAL system has to be under 200w.

You would have to underclock that VII to 1200-1300Mhz and cut 4-8 CU's to get it to run at 130-150w and then you would no longer have a 13.8TFLOP GPU.

Undervolted 225w... Good god ron, your own posts bite you back in the ass. Let me guess Navi will turn that 225 w into 130w and you will still be able to get 12+TFLOPS on GCN?... Please. I'll believe it when I see it, so far AMD is incompetent.

Add 7nm+ 10 percent improvement with next year's process tech.

My estimate didn't ask for VII's 1750 to 1800 Mhz clock speed e.g. try 1550 Mhz UV range instead.

I didn't specifically state 13.8 TFLOPS GPU for consoles.

VII's workstation features needs to be cut e.g. 64bit math, High Bandwidth Cache.

I gave you some raw data, work on your estimate. Hint: This is a discussion.

Avatar image for ronvalencia
#116 Posted by ronvalencia (28060 posts) -

@04dcarraher said:
@Grey_Eyed_Elf said:

Yes more performance but that has nothing to do with power draw. Even under-volted that Radeon VII Still pulls over 200w... at 13.8 TFLOPS. 225w in FC5:

That is just GPU power draw... the TOTAL system has to be under 200w.

You would have to underclock that VII to 1200-1300Mhz and cut 4-8 CU's to get it to run at 130-150w and then you would no longer have a 13.8TFLOP GPU.

Undervolted 225w... Good god ron, your own posts bite you back in the ass. Let me guess Navi will turn that 225 w into 130w and you will still be able to get 12+TFLOPS on GCN?... Please. I'll believe it when I see it, so far AMD is incompetent.

And this is with the fact that Radeon VII is using 7nm node.... I would expect NAVI with a 130 TDP to be around VEGA 56 level of TFLOP performance . Highly doubt 1st gen NAVI is going to be coming close to VEGA 64 or RVII performance wise.

At the same clock speed, VII (60 CU) has better ray tracing results over Vega 64

Avatar image for aia89
#117 Posted by aia89 (2827 posts) -

The Navi architecture is aimed at mid-range gpu's, so by those specs it won't be able to outperform the current best rigs.

Avatar image for heirren
#118 Posted by Heirren (1496 posts) -

Surely sony understands people are willing to spend more for technology, today, especially if it is going to last 8 years.

Avatar image for tormentos
#119 Posted by tormentos (29191 posts) -

@Grey_Eyed_Elf said:

According to that estimate the Radeon VII which is 7nm GCN with its 1750MHz boost should sit at 160-170w. It does not... Infact its peak is at 330w and average is closer to 270w and it uses 60CU's rather than 64 in the Vega 64.

64 CU's with GCN at 1600MHz with 130w?... is that idle?... That's less power than a RX 570.

You can't be this gullible ron.

GCN is power hungry just look at what 7nm did when you compare the Vega 64 to VII... a die shrink to 7nm and 4 less CU's and its still eating more power than a 1080 Ti. You believe that Navi will somehow make GCN at high CU counts with the same die size as VII drop its power consumption by 1/2?...

While i don't think it will be like the old days the PS3 fat did pull more than 200 watts and had a 380 watts psu.

But i think newer consoles will not go that power hungry again.

Avatar image for 04dcarraher
#120 Edited by 04dcarraher (23196 posts) -

@ronvalencia said:

At the same clock speed, VII (60 CU) has better ray tracing results over Vega 64

NAVI is not VEGA, which means there will be more downsizing on compute power. NAVI is about efficiency per watt for "gaming" not a compute focused gpu like VEGA. R VII(ie M150/160) extended FP16 instructions and packing 16 instructions per ALU per clock. Also the better HBCC/HBC that allows the gpu use vram as last level onboard cache helps. But the major improvement for VII vs VEGA 64 is the wider memory bus(bandwidth).

So comparing VII vs 64 with " Raying Tracing" is flawed since of the 2x memory bus.

Avatar image for pmanden
#121 Edited by pmanden (665 posts) -

@fedor: Cos they are cow fanboys at their finest, blinded by their belief that Sony is almighty and can deliver 24 teraflops, 20 gb ram for 400 USD. They will probably start crying when they realise that the PS5 will be in the 10 teraflops range.

Avatar image for TheEroica
#122 Posted by TheEroica (18414 posts) -

OnE of many thousands of leaks sure to vamp the flames of speculation and intrigue... This happens every gen.

Avatar image for ronvalencia
#123 Posted by ronvalencia (28060 posts) -

@Random_Matt said:

Not really possible to change the 64CU limit, unless AMD are prepared to throw tons of money at it. The whole rumour is fake, do not believe they will up the clock speed either.

Adding CUs are not a problem when Fury X is just Hawaii 44 CU with another 20 CUs. GCN's main problem is the upgrade for existing quad raster engines and 64 ROPS configuration i.e. the real GPU hardware that separates GPUs from DSPs.

Avatar image for ronvalencia
#124 Edited by ronvalencia (28060 posts) -

@04dcarraher said:
@ronvalencia said:

At the same clock speed, VII (60 CU) has better ray tracing results over Vega 64

NAVI is not VEGA, which means there will be more downsizing on compute power. NAVI is about efficiency per watt for "gaming" not a compute focused gpu like VEGA. R VII(ie M150/160) extended FP16 instructions and packing 16 instructions per ALU per clock. Also the better HBCC/HBC that allows the gpu use vram as last level onboard cache helps. But the major improvement for VII vs VEGA 64 is the wider memory bus(bandwidth).

So comparing VII vs 64 with " Raying Tracing" is flawed since of the 2x memory bus.

VII (GFX906) has extra quad rate 8bit integer and half-rate double precision on top of VII's extra AI related instruction set. PS4 Pro already has double rate FP16.

HBC is useless for most games since it has higher latency when compare to L2 cache.

The gap between theoretical vs practical memory bandwidth (before DCC is applied ) with HBM v2 shows it's less efficient when compared to GDDR5

V64's 303 / 485 = 62.4 percent efficient, VII scaled estimate is 624 GB/s practical

RX580's 193 / 256 = 75.39 percentefficient,

The next X1X's 384 bit with GDDR6-14000 has estimate 506.6 GB/s practical from 672 GB/s theoretical.

The next X1X's 384 bit with GDDR6-13000 has estimate 470 GB/s practical from 624 GB/s theoretical.

There's a technical reason why NVIDIA didn't jump on HBM v2 for gaming.

For RX-580 replacement with NAVI, 256 bit x GDDR6-14000 = 448 GB/s theoretical or estimate 337.7 GB/s practical which is more than a match against V64's 303 GB/s practical. No brainier on RX-580 replacement landing somewhere Vega 56/64 range.

Stock Vega 56 HBM v2 settings(409 GB/s theoretical), Vega 56 at 1709 Mhz OC already beats or rivals RTX 2070. My point, quad stack HBM v2 on VII has diminishing returns.

Avatar image for Zero_epyon
#125 Posted by Zero_epyon (13289 posts) -

What if that 8K GT Sport demo Sony showed off was running on a PS5 devkit? Now I'm not suggesting PS5 will do 8K natively, but it could do 4K natively, and perhaps do 8K via checkerboard rendering. Since the PS4 has dedicated checkerboard rendering hardware, one can imagine they'd improve this hardware and include it in the PS5.

Avatar image for ronvalencia
#126 Edited by ronvalencia (28060 posts) -

@Zero_epyon said:

What if that 8K GT Sport demo Sony showed off was running on a PS5 devkit? Now I'm not suggesting PS5 will do 8K natively, but it could do 4K natively, and perhaps do 8K via checkerboard rendering. Since the PS4 has dedicated checkerboard rendering hardware, one can imagine they'd improve this hardware and include it in the PS5.

Vega 56 at 1548 Mhz OC and HBM v2 at 950Mhz can run FM7 at 8K at 60 fps

Loading Video...

I assume Sony's GT Sport is optimized like FM7 for AMD hardware i.e. making AMD TFLOPS to rival NVIDIA's TFLOPS.

Stock Vega 56 target was factored in for Crytek's ray-tracing 4K 30 fps demo.