• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Rembrandt 6nm APUs - Will they use DDR5 RAM?

Yes, no reason to think they won't. They won't be 4ghz chips (limiting to 4ghz is just useful to compare products of different clock speeds, plus they don't have to commit themselves to higher clocks in advance). I think they will probably have similar clock speeds to the Ryzen 5000 series. AMD don't need a lot of cores to compete with Alder Lake in tasks like games.

8 is all Alder Lake needs, the E-Cores make only a small performance difference. The prototype Zen3D chip had 12 cores. Performance shown here:
https://cdn.mos.cms.futurecdn.net/od7uMMTWeNWpUxNzikPa5Y-970-80.png

So, I'd expect a 10-15% performance boost in in Directx12/Vulkan games, maybe less in the majority of DX11 games (maybe just 5% in some).

It's a useful upgrade for people with B550/B570 AM4 boards, lots of people don't seem willing to upgrade their whole system to DDR5/Alder Lake. Don't know about support for older AM4 chipsets though.

I think one factor maybe some have missed with Zen3D is the difficulty of adding compatibility to older boards, it's presumably difficult to update the firmware of tons of older AM4 chipsets to add support for Zen3D. If they don't get this right, the launch could be seen as a failure, so I think that's why it didn't release in 2021. It wouldn't surprise me if they have to remove support of some older (less common) CPU models.

Didn’t AMD show some games gaining 26% with the prototype chip at 4ghz?

Im talking Zen 3D. Do you mean 15% from the 6nm Zen+ APU?
 
For Zen3D, I think it will be 10-15% gain on average in DX12/Vulkan games. AMD will put always put in some high outliers in the promotional material, that's just good marketing. It's only extra L3 cache, only so far that can help with performance. Maybe more L2/L1 cache would boost performance further in the future.
 
What if they tweaked architecture of Zen3d little bit to gain couple % IPC like they did with Zen+? Or memory controller on 7nm node so every sample can work at 2000mhz+ IF, combined with extra l3 cache it will be very competitive, in some games better than alder lake, in some little weaker, but in range of ADL.
 
For Zen3D, I think it will be 10-15% gain on average in DX12/Vulkan games. AMD will put always put in some high outliers in the promotional material, that's just good marketing. It's only extra L3 cache, only so far that can help with performance. Maybe more L2/L1 cache would boost performance further in the future.

L1 and L2 cache is part of the core pipeline. Essentially if you double the L1 and L2 pools you would triple the size of the chip and probably be looking at 5-10x the cost.
 
That's interesting. Sounds like it could be possible with TSMC's 5nm or Intel's 4/3 process technologies, as these would both offer around double the transistor density of existing AMD/Intel CPUs.

I guess they'd need to analyse if the benefits are worth the additional cost, which new architecture designs could help to reduce.
 
Last edited:
I doubt redesigning the core around individual cache would gain much performance outside of some niche work types. With todays massively multi core approach and everything focused on large scale parallelism unified pools of cache is likely a better approach. I think AMD are on the right path.

IIRC IBM have pretty much the CPU you’re describing. Someone has anyway.
 
I had a look on Wikipedia, and it looks like each AMD microarchitecture generation has included both CPUs (desktop) and APUs since Excavator in 2015 (followed by each Zen generation). Before 2015, it varied a bit with each generation. Link:
https://en.wikipedia.org/wiki/Excavator_(microarchitecture)#Processors

So, if AMD announce 6nm APUs on the 4th of January, it seems a fairly safe bet that there will be 6nm desktop CPUs of the same generation released too.
 
I had a look on Wikipedia, and it looks like each AMD microarchitecture generation has included both CPUs (desktop) and APUs since Excavator in 2015 (followed by each Zen generation). Before 2015, it varied a bit with each generation. Link:
https://en.wikipedia.org/wiki/Excavator_(microarchitecture)#Processors

So, if AMD announce 6nm APUs on the 4th of January, it seems a fairly safe bet that there will be 6nm desktop CPUs of the same generation released too.

Maybe. I could see Zen+ AMD 6nm all in one and embedded systems. Not sure AMD will release AM4 retail APU’s though.
 
LPDDR5 was always on the cards for the newer Zen3 mobile parts, especially useful given the bandwidth advantages when moving to an RDNA2 based GPU on the die. 7nm or 6nm is the only unanswered question, but given the alleged clock speeds, core counts and TDP's I'd imagine that 6nm would be advantageous to keeping the power usage down, and more dies per wafer being monolithic.
 
I suppose if they do some desktop CPUs on 6nm EUV, at least they will probably be able to get higher clock speeds out of Zen 3, without increasing the temps much. Maybe 5ghz+ for the 8 core and above models. A 10% increase in clock speeds should be enough for Zen 3 to compete with Alder Lake, AMD isn't that far behind. Might need slightly higher clocks than this though, it's hard to say. AMD has the advantage on core count so, you'd expect that to put them ahead, if could manage 5ghz or above on all CPU cores.

Clock speed is unfortunately still one of the main weaknesses of AMD's Ryzen CPUs, vs Intel. With the Ryzen 5950X, the average clock frequencies with all threads enabled can fall to 4.4ghz with SSE instructions and 4.2Ghz on AVX:
https://www.techpowerup.com/review/amd-ryzen-9-5950x/21.html

I wonder if this improves with SMT / 'Hyperthreading' disabled?

Combine that with a better DDR5 memory controller than Alder Lake, and they'd be onto a winner.

I think they have some catching up to do in this area, 3733mhz is the current 'infinity fabric limit' on DDR4, correct?

Ideally, they'd aim for at least DDR5 4800Mhz RAM (dual channel), with the memory controller running at 1:1, to see an improvement over DDR4.
 
Last edited:
I think they will quickly hit diminishing returns by increasing DDR5 frequency if they can't have a more linear increase in CAS latency, compared to say DDR4000 modules at CL19.
 
Very exciting stuff:

AMD-Ryzen-5800X3D-Raphael-Zen4.jpg


Will watch the whole thing in a bit.

It's exceeded my expectations tbh (thought it would take longer for Zen 4), Intel is in trouble in 2022. Although, Zen 4 /AM5 might not release until November / December 2022

Agree, can close this thread iyw, pretty obvious what I should recommend for my dad now.
 
Last edited:
Back
Top Bottom