Forums - Sony Discussion - Digital Foundry believes they may have found PS5's chipset

shikamaru317 said:
HollyGamer said:

SO you contradicted your own original post??? 

How so? If AMD can manage to pull off a 1.5x improvement again, that would put Navi 10 at 10.6 tflops, which is within the 10-12 tflop range that Digital Foundry says to expect for PS5 in the video. If they can pull off a 1.6x improvement, which they have never done before to my knowledge, but might be possible since they have been working on Navi for nearly 3 years since Polaris released, that would put Navi 10 at 11.4 tflops (very close to the 11.1 tflop 48 compute unit number in the video). 

All I'm saying is, keep your expectations in check. Better to be pleasantly surprised if AMD does pull off a big improvement for Navi 10 over it's predecessor, than to be disappointed because you are expecting high end performance numbers from a mid-range chipset. 

11.1TF would put it just above the 10.7TF of Stadia and wouldn't be far off a 12TF Anaconda. For a typical console form factor, even with a more advanced vapor chamber cooling solution like XB1X, 11.1TF would be pushing the limits. Plus 111 just has a nice ring to it.



Around the Network

This is probably not it since the 95 watt TDP gives it away that this is not the PS5 chip. I would imagine that the power consumption is at least roughly double that ...



fatslob-:O said:
This is probably not it since the 95 watt TDP gives it away that this is not the PS5 chip. I would imagine that the power consumption is at least roughly double that ...

XB Lockhart maybe?



EricHiggin said:

XB Lockhart maybe?

What was XB Lockhart supposed to be again ? 

Is it also going to be used for Microsoft's xCloud too as well ?



shikamaru317 said:
HollyGamer said:

The range is between 8 to 14, 7 teraflop they said, but Digital foundry said it's hard to believe so 13 teraflop is possible  

The 14.7 was for the the max CU's possible on the GCN architecture, 64 CU's. Highly unlikely we would ever see a console with 64's CU's, issues with yields alone would drive the price through the roof. Radeon 7 has 60 CU's and it costs $700 right now. 

Vega 64 has more CU's than Vega 7 and is cheaper.

shikamaru317 said:

I can't see that happening. Navi 10 is a replacement for Polaris 30. Polaris 30 is a 7 tflop chip with 36 CU's, so 13 tflops and 60 CU's would be nearly double it's predecessor. AMD has never managed to release a new GPU generation with a performance improvement of more than 1.5x over the previous generation afaik. For instance, Polaris 30, RX 590, offered a 1.4x improvement in flops over the previous generation GPU it replaced, the R9 390.

I am thinking around Vega 56 levels of performance but with a substantially lower CU count thanks to clockrates.

It should loose out in terms of flops and memory bandwidth but see gains in other areas which should hopefully result in competitive gaming performance with Vega 56. - But Vega 56 should still win in compute/flops.

JEMC said:

Unless AMD has somehow massively improved their GCN architecture with Navi, I find it ver, very hard to believe that it will reach 1.8GHz on something as limited by airflow as a home console.

So far, their fastest graphics processor based on the GCN architecture is the Radeon VII, based on Vega 20 and made on 7nm, with a base clock of 1,400MHz and a boost clock of 1,750MHz. And that thing needs three fans to cool its 295W TDP.

So, as I said, unless AMD has done magic with Navi, I don't see how PS5 can have a 1.8GHz GPU.

7nm is maturing now, so clockrates can go up.
But in also saying that... Navi 10 is likely to be a smaller chip than Vega 7, so clockrates should again be able to be a step up.

1.8Ghz clockrate is entirely within the realm of possibility in my opinion if the SoC is a conservative size and not monolithic with a massive TDP to match.

HollyGamer said:

Then also "  Tum Apisak "  leak that mentioned by Digital Foundry and Adored TV guy, they  agree about the performance of navi 10 will be around  Vega 64 + 15% , While Digital foundry said 1,8 Ghz with 64 CU  will be around 14,7 teraflop, 60 Cu will be around 13,8 teraflop, 56 Cu will be 12 teraflo etc etc until 32 CU. All Navi price will be around 430 USD to 130 USD. 

As in fact I am agree with your post "12 teraflop" but i am slightly more optimist that PS5 will be around 13 teraflop (60 CU) from 64 CU disable.  

I doubt greater-than Vega 64 levels of performance is going to happen.
Just like how Polaris never managed to beat Hawaii, let alone Fiji.

People need to keep their expectations in check... Navi is still Graphics Core Next... And that includes all the Pro's and Con's of the architecture.

shikamaru317 said:

Also, I just found this in a PC Gamer article about Navi:

"Looking at the Vega 64 vs. Radeon VII core size, we can get at least a reasonable estimate of what can be done. Vega 10 (Vega 64) at 14nm is a 486mm^2 die with 12.5 billion transistors, Vega 20 (Radeon VII) at 7nm is a 331mm^2 die with 13.2 billion transistors. That's 32 percent smaller with 6 percent more transistors. With similar scaling, AMD could end up with a 200mm^2 die with 40-48 CUs (2560 to 3072 streaming cores). Toss in higher clockspeeds and lower power consumption, with more memory bandwidth thanks to GDDR6, and the top Navi 10 part could be competitive with RTX 2070."

PC Gamer says they are expecting between 40 and 48 CU's for Navi 10. Based on the scale in Digital Foundry's video, that would be 9.2 tflops for 40 CU's and 11.1 tflops for 48 CU's. So yeah, I wouldn't expect more than 11.1 tflop for PS5 if I was you, unless Sony decides to wait on Navi 20's release in 2020 instead of using Navi 10. 

Keep in mind that you need an 8-core Zen CPU, Audio processor, memory controllers, caches and other pieces of logic like North Bridge/South Bridge and such all in that SoC. Expecting the entire package to be 200mm2 is being disingenuous...

Flops aren't important anyway. The sooner gamers realize this the better. It's a theoretical ceiling not representational of a parts gaming capabilities.

Bofferbrauer2 said:
I posted that chip before, as Gonzalo was rumored to be the Chip for the next gen consoles.

However, if the 95W TDP is correct, this won't be it, that's just gonna be a higher-end APU (all Ryzen APU to date were 65W or less).

With 95W, even in 7nm, that wouldn't be enough to even compete with the PS4 Pro. This chip could reach 2.5-3TFlops, far below even the 4.2TFlops of the PS4 Pro

You can get 4x Zen cores @ 3.3Ghz, 11 CU's for only 35w TDP in the Ryzen 2800H @14nm.

shikamaru317 said:

Yeah, the 95 watts seems too low for sure. Sure it's a 7nm chipset, two die shrinks down from the 16nm Xbox One X chipset, but I don't see how they could get the power usage down from 175 watts on Xbox One X to 95 watts on PS5, just from 2 die shrink's, while increasing the tflops from 6 to 10-12, with a more powerful CPU as well. Something seems off there for sure. Maybe the 95 watts is for the CPU in the APU only, with an extra 150 watts from the GPU (Navi 10 is rumored to have a 150 watt TDP)? 

One die shrink. The Xbox One X isn't made at 28nm or larger you know.

The reason being is that 7nm is being refined, is getting mature... So chips can run faster at the same power level.
All comes down to those pesky clockrates rather than a sheer magnitude of functional units. (Balancing act and all that.)

drkohler said:

One of the millions of rumours was that the base clock was 1GHz with a boost max of 1.8GHz. There is no way an AMD gpu at 1.8GHz blows 95 Watts only. At 1Ghz, it is entirely possible for the gpu.

1Ghz isn't going to happen.
1Ghz @ 64CU's would equate to 8 Teraflops... And we know it's not going to have 64 CU's.

If we were to take the Xbox One X's 40 CU count... Clock it at 1.8ghz, we are probably looking at a good 50% performance boost (9.2 Teraflops), drop it from 16nm to 7nm and it would run at roughly the same powerlevel.

Add in further refinements to the 7nm manufacturing process, the architectural improvements that Vega and Navi will bring... And the overall performance improvements from Scorpio start to look rather exciting... And that is with the same number of CU's.



--::{PC Gaming Master Race}::--

Around the Network
fatslob-:O said:
EricHiggin said:

XB Lockhart maybe?

What was XB Lockhart supposed to be again ? 

Is it also going to be used for Microsoft's xCloud too as well ?

Around 4.0TF apparently, maybe up to 6.0TF like XB1X. I think One SADE is mostly for the cloud and Lockhart is probably for cloud, digital, and disc and Anaconda is mostly for digital and disc.



fatslob-:O said:
EricHiggin said:

XB Lockhart maybe?

What was XB Lockhart supposed to be again ? 

Is it also going to be used for Microsoft's xCloud too as well ?

There was a leak saying that Lockhart is a streaming console with a 4 tflop GPU, which is designed to run latency intensive tasks locally, with the rest of the data pulled from the cloud. So yeah, a 95 watt TDP could work for Lockhart maybe. 



I also don't think that the PS5 will be built on 7nm DUV like Gonzalo probably is. I believe it will be built on TSMC's 7nm EUV logic node which will already be in use if PS5 launches in 2020 ...



shikamaru317 said:
fatslob-:O said:

What was XB Lockhart supposed to be again ? 

Is it also going to be used for Microsoft's xCloud too as well ?

There was a leak saying that Lockhart is a streaming console with a 4 tflop GPU, which is designed to run latency intensive tasks locally, with the rest of the data pulled from the cloud. So yeah, a 95 watt TDP could work for Lockhart maybe. 

Richard did point out it's a little odd it's running on Windows. It wouldn't be odd if it were for XB though.

fatslob-:O said:
I also don't think that the PS5 will be built on 7nm DUV like Gonzalo probably is. I believe it will be built on TSMC's 7nm EUV logic node which will already be in use if PS5 launches in 2020 ...

Well unless EUV has stellar yields by then I can't see PS5 using it. 7nm DUV yields should be at the point that consoles have no worries in terms of available capacity and price by 2020. Third gen Ryzen and Navi should be coming off the line near flawless by then on 7nm DUV so it makes a lot of sense for PS5 to just continue from there.



EricHiggin said:

Well unless EUV has stellar yields by then I can't see PS5 using it. 7nm DUV yields should be at the point that consoles have no worries in terms of available capacity and price by 2020. Third gen Ryzen and Navi should be coming off the line near flawless by then on 7nm DUV so it makes a lot of sense for PS5 to just continue from there.

It's coming in use this year with the new mobile SoCs. 7nm EUV will be a significant improvement in costs because it's using the most advanced scanners to achieve shorter wavelengths to be able to create higher resolution features of the chip without resorting to multiple patterning. Won't have to deal with quadruple patterning anymore like with 7nm DUV or double patterning with 16/14nm ... 

Just being able to do double patterning for 7nm will massively reduce the amount of masks used into making these chips ...