Why no Zen 3 in PS5 and XSX?

The silicon was taped out many months ago. Zen 3 hasn't even shipped. Can't use something that seat of the pants. As far as the GPU, they look at AMDs pipeline and pull in RDNA 2 features they want, bit different and more modular than a CPU.
 
Last edited:
this consoles are designed in years not days. the margins for power, heat, and budget arte very very small. You just cannot throw something very new into the mix.
 
I mean zen 3 CPUs, rdna 2 GPUs and the consoles launch around the same timeframe. Why didn't they go for zen 3 then? It kinda is a missed opportunity since gaming performance is much higher than zen 2.

They use a 1 yr old CPU arch and the newest edge GPU architecture available !? Makes no sense to me.

Please someone explain it to me.

  • 105 watts is too high for a console application for the CPU portion

  • Console have to design an APU, so CPU and GPU together, needs more time

  • Timing, Consoles have allot to test with both CPU, GPU, IO and drivers / apis and need more time.

It would be interesting to see if either has a unified cache as we do nnot know yet do we
 
Last edited:
Up to 4MB data width = low latency, which means everything is in L3 cache or lower (L1/L2).
8MB data width = half in RAM, half in cache so latency is at the halfway point.
16MB data width = ~75% RAM latency, 25% cache latency.
>32MB data width = ~100% RAM latency, basically entirely in RAM.
Okay i see now, thanks a lot for explaining. Someone should inform Paul about this, there is no need to speculate anymore.
 
Okay i see now, thanks a lot for explaining. Someone should inform Paul about this, there is no need to speculate anymore.

A tip is to not listen to these youtube clowns (RedGamingTech, Moore's Law is Dead, Coreteks etc). Mostly the just make shit up or speculate.
 
As much as MS boasts about Server class config, Zen 2 is ready for APU's now and has been for a while. Zen 3 has just premiered now for high end desktops, it would have not fit the timelines either of the console makers had.

And how do they have RDNA2 GPUs?

They are not customers, they don't need to wait for the products to hit the market. I mean maybe the timeline for Zen3 was much tighter and overall it wasn't worth it because realistically that "little bit extra" compared to Zen 2 just doesn't make sense either way.

If they really wanted to they probably could have.
 
There's no low-power variant of Zen3 available.. You can't compare a full fledged standalone CPU with a SoC package.
And also, Zen2 is tried & tested, they know what they get, so easier to "design" the GPU portion of the SoC
Youre wrong, zen 2 in the consoles is the full desktop version with lower clocks.
 
Because it didn't exist months ago when they did start to produce the consoles, and years before when they started to make the games.

And well, because its price is too expensive, right now doesn't make sense for a console or for the mainstream PC market.

The consoles are getting custom GPU'd based on a hybrid of RDNA 1 and 2.
Fake news, both are RDNA 2 with custom stuff on top of it.
 
Last edited:
PS5 has Zen 3 because Cerny custom built some of the features and will allow AMD to use the custom features for Zen 3 possibly even Zen 4
 
Last edited:
And how do they have RDNA2 GPUs?

They are not customers, they don't need to wait for the products to hit the market. I mean maybe the timeline for Zen3 was much tighter and overall it wasn't worth it because realistically that "little bit extra" compared to Zen 2 just doesn't make sense either way.

If they really wanted to they probably could have.

They are not the customers, but customers can buy the console (or pre-order it) from mid September for a November launch, MS and Sony need to start volume production a lot earlier on. I do not think 6 months or more ago Ryzen 3 was as in an advanced state as RDNA2 based cores are.

I get your , but even on the consoles most of the power budgets would afford a higher power RDNA2 based GPU (without perhaps requiring as deep mobile oriented low power optimisations).

Still, I think RDNA2 has been finalised earlier than Ryzen 3 has and Ryzen 2, which already received its mobile oriented variant with the 4000 line, was possibly the first thing they settled on to ensure developers had a stable semi final DevKit as soon as possible.

The performance boost from 1.6 GHz Jaguar cores to the Ryzen 2 cores (same variant used in the 4000 based APU's) was already big enough for both console manufacturers to hit their CPU performance target and thus saving money on not even trying to get Ryzen 3 rushed out for the consoles earlier made little sense.
 
Last edited:
Youre wrong, zen 2 in the consoles is the full desktop version with lower clocks.
I disagree, it is likely the revision they ended up using for the 4000 line APU's the Renoir core:


https://www.google.co.uk/amp/s/www....-4000-renoirs-memory-controller-is-scary-good (I think they already confirmed a reduced L3 cache for the console Zen 2 cores too).
 
Up to 4MB data width = low latency, which means everything is in L3 cache or lower (L1/L2).
8MB data width = half in RAM, half in cache so latency is at the halfway point.
16MB data width = ~75% RAM latency, 25% cache latency.
>32MB data width = ~100% RAM latency, basically entirely in RAM.

That does not account for latency between the 2 halves, that is just latency by size of CPU cache.
 
Everbody saying zen 3 was not ready.
How can we have RDNA2 with your logic guys? Zen 3 launching as we speak.

my guess is budget. Zen 2 was cheaper and gpu was priotized.
 
I mean, the 8 core processer costs more than the entire console. That probably has something to do with it. The better question is why do you care? You are on a console for a reason.
 
Everbody saying zen 3 was not ready.
How can we have RDNA2 with your logic guys? Zen 3 launching as we speak.

my guess is budget. Zen 2 was cheaper and gpu was priotized.
RDNA 2 GPUs are launching at the same time as the consoles. Where are those Zen 3 laptop CPUs?
 
the price and it would be useless.

Zen 2 already has very good game performance, especially for a console environment.
Zen 3 only gets interesting if you are on PC and you want crazy framerates in every game.

a well optimised game on a Zen 2 CPU will run perfectly fine. you don't need CS:GO at 400fps on a console
 
Probably: It wasn't ready in time. It's too expensive.

A $500 console that has an underclocked Ryzen 7 3700 is already a big enough miracle.

It is however very unfortunate for the consoles that Zen 3 focuses primarily on gaming, and they have to miss out on that.

On one hand, if they are sticking almost always to higher resolutions and lower fps (60fps/30fps), then the GPU would probably be the bottleneck almost always in most cases.

However, Zen 3 can still been very useful for the consoles, if for instance it was used to achieve the same performance as the current CPUs in XB/PS5, but at lower clocks/power/thermals.
 
Last edited:
The CPUs should be more than enough for the GPUs and FPS targets. For console that's all that matters - the minimum spec needed to eliminate bottlenecks.
 
My man, last time you got 2 4-core jaguars duct taped together. Something that can't even come close to competing with a really budget desktop PC CPU at the time.

A 8c/16t Zen 2 is already insanely good for the price.
 
Last edited:
I mean zen 3 CPUs, rdna 2 GPUs and the consoles launch around the same timeframe. Why didn't they go for zen 3 then? It kinda is a missed opportunity since gaming performance is much higher than zen 2.

They use a 1 yr old CPU arch and the newest edge GPU architecture available !? Makes no sense to me.

Please someone explain it to me.

There are some rumors about PS5 using infinity caches in the CPU, similar to ZEN3. Nothing confirmed so better wait as we might hear more from both after RDNA2 official reveal.
 
I mean zen 3 CPUs, rdna 2 GPUs and the consoles launch around the same timeframe. Why didn't they go for zen 3 then? It kinda is a missed opportunity since gaming performance is much higher than zen 2.

They use a 1 yr old CPU arch and the newest edge GPU architecture available !? Makes no sense to me.

Please someone explain it to me.

They would have to wait one more year, pay more for the newest tech and this would force them to use bigger/faster GPUs because the CPU would be overkill for the GPUs that they have now.
 
the price and it would be useless.

Zen 2 already has very good game performance, especially for a console environment.
Zen 3 only gets interesting if you are on PC and you want crazy framerates in every game.

a well optimised game on a Zen 2 CPU will run perfectly fine. you don't need CS:GO at 400fps on a console
I think we will have same problems this gen with framerate as games get higher fidelity with ray tracing but both sony and microsoft made compromises again this gen. Yes we will see improvement on what we have not but thats not enough.
Fucking give me a 600 paunds console 6-7 year life
People buy phones every 2 years with that amount of money
 
I think we will have same problems this gen with framerate as games get higher fidelity with ray tracing but both sony and microsoft made compromises again this gen. Yes we will see improvement on what we have not but thats not enough.
Fucking give me a 600 paunds console 6-7 year life
People buy phones every 2 years with that amount of money

the framerate issues this gen will be GPU bound not CPU bound.

there's almost no way any game can't run at 60fps on these CPUs.
if games will go below 60fps it will definitely be the GPU that can't keep up.

we already see this, games like Spider-Man and Ratchet & Clank have 30fps modes with raytracing and 60fps modes without raytracing.

we will see this a lot next gen, but that's also a good thing, because now the developers actually can give you working high performance modes.
that was impossible on current gen in many games because no matter how low they would push the graphics, the CPU just couldn't run the game fast enough.
 
becouse of this:

masterrace.png
 
Desktop variant of Zen 2 CPUs are available since over a year, but the mobile variant has just now arrived. And that's what PS5/XSX are using. Mobile Zen 2 with 8MB L3 cache (8MB confirmed for SX). Additionally it's a "server grade" chip i.e. enhanced security.

Anyway, we have desktop Zen 3 available in a few weeks from now but the mobile variant of Zen 3 will not arrive until 2021 Q3/Q4. Consoles would have to be delayed by 2021 if they wanted Zen 3. So yeah, availability is the issue first and foremost, IMO. Next would be cost.
 
I mean zen 3 CPUs, rdna 2 GPUs and the consoles launch around the same timeframe. Why didn't they go for zen 3 then? It kinda is a missed opportunity since gaming performance is much higher than zen 2.

They use a 1 yr old CPU arch and the newest edge GPU architecture available !? Makes no sense to me.

Please someone explain it to me.
Price, availability, necessity (I can't think of any console games that would be bottlenecked by an 8/16 Zen2 chip. People aren't playing CIV6, Cities:SL, Total War, Ashes, etc. on consoles)

Would be overkill
 
How does this graph proves this? I am not saying that's not the case, i am asking genuinely.
Unified L3 8MB cache would delay the steep upward curve until it reaches the 8 MB point.

Latency to the second 4MB L3 cache is lower than the external memory.
 
Last edited:
Desktop variant of Zen 2 CPUs are available since over a year, but the mobile variant has just now arrived. And that's what PS5/XSX are using. Mobile Zen 2 with 8MB L3 cache (8MB confirmed for SX). Additionally it's a "server grade" chip i.e. enhanced security.

Anyway, we have desktop Zen 3 available in a few weeks from now but the mobile variant of Zen 3 will not arrive until 2021 Q3/Q4. Consoles would have to be delayed by 2021 if they wanted Zen 3. So yeah, availability is the issue first and foremost, IMO. Next would be cost.
There is two Zen 3 based APUs i.e. Zen 3 + Vega 8 CU at 2Ghz aka Cezanne and Zen 3 with RDNA 2 IGP aka Rembrandt.
 
There are some rumors about PS5 using infinity caches in the CPU, similar to ZEN3. Nothing confirmed so better wait as we might hear more from both after RDNA2 official reveal.
PS5's APU will not have a 128 MB Infinity Cache L3 cache.

Reminder, XBO's 32 MB ESRAM can handle 1600x900p frame buffer without delta color compression.

128 MB is four times of 32 MB. 4K is four times of 1080p resolution.

128 MB can handle 4K frame buffer with delta color compression (DCC) i.e. RX 6800 has greater 4K optimizations when compared to game console RDNA 2 GPUs

RX 6800's 128 MB Infinity Cache is attempting to reduce external memory access hit rates.

Smaller GPU SRAM storage must do micro-tile cache render optimization methods.
 
We've known for over a year that this isn't the case. Latency graphs from when AMD/Sony benchmarked the chip:

EADW-CJVAAA7sEz


The CPUs in XSX/PS5 are exactly the same except for clock speeds.

For CPU,

Whats that graph got to do with a unified cluster 8 MB vs 2 x 4 MB ?

If data is in wrong half.....
 
For CPU,

Whats that graph got to do with a unified cluster 8 MB vs 2 x 4 MB ?

If data is in wrong half.....

If it was a unified L3 cache a core would have access to 8MB cache rather than 4MB, so you'd have 8MB data width at roughly the same ~20ns as 4MB rather than ~90ns (i.e. you have 50% of your data in L3 and half in RAM).
 
Also, let's nip this PS5 has Infinity Cache nonsense in the bud.

We know the SoC is roughly ~300mm^2 from Sony's disassembly video.

* CPU portion for Renoir (8xZen2 with 2x4MB LM3) is roughly ~40mm^2.
* Navi 10 ~250mm^2, and the PS5 GPU is similarly specced (36CU, 256b).
* Looking at Navi 21 die shots, the 128MB LLC is approximately ~100mm^2.

There just isn't die space for large LLC.
 
For CPU,

Whats that graph got to do with a unified cluster 8 MB vs 2 x 4 MB ?

If data is in wrong half.....



Matt didn't confirm PS5's Zen 2 unified L3 cache.

On PC's Zen 2 non-APU with two CCD chiplets, if CCX0 (on CCD0) needs data from CCX3 (on CCD1), then the data transfers will travel via the chiplet IO. On PC Zen 2 APU, this problem is minimized since the two CCX modules are on the same chip.


b958c5fb-1187-46b6-9643-aa8ddc560fc9.PNG




There are two issues with CCX's communication links and these are

1. CCX's link within CCD.
2. CCX's link with another CCD.

For point 1, Zen 3 removes the Infinity Fabric link between the two CCX modules hence the link between CPU cores is done via the L3 cache.
 
Last edited:
Top Bottom