• 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.

*** AMD "Zen 4" thread (inc AM5/APU discussion) ***

Permabanned
Joined
2 Sep 2017
Posts
10,490
We're starting this before we have Zen 3?

Exactly what I was thinking. There should be a rule that you can't make threads like this when the previous generation isn't even out. This will just turn into another mess of back and forward quarrels like the current 4000 thread is which makes finding pertinent information harder than it should be.
 
This thread shouldn't confuse anyone who's paying attention though.
AT already have both a Zen 3 and a Zen 4 speculation thread, and that seems to work fine there.
Keep thinking that going full DDR5 for Zen 4 could be risky depending on price and availability.
If Zen 3 and Zen 4 stay with a separate IO die, making separate ones for both might make more sense.
The last thing they'd want is a wonderful new platform totally dependent on memory manufacturers having DDR5 out in volume.
Unlike with Intel trying to ram Rambus down everyone's throat way back when, third-party vendors cannot do a VIA again as the memory controller is no longer on the motherboard or chipset.
 
There is one very important detail here:


https://wccftech.com/amd-ryzen-2020-2022-cpu-apu-roadmap-leak-zen-2-3-4-families-unveiled/

So, Vermeer is MSDT CPU with Zen 3, Warhol is MSDT CPU with Zen 3+, Raphael is MSDT big APU with Zen 4 and Navi of some generation.
So, probably AMD is cooking a very bad surprise for us.

What is Warhol? Just another +1-2% ala the Ryzen 3000 XT series? :confused:

Vermeer only 15% IPC improvement, Warhol +1-2% if any, Raphael - no core count improvement.

:eek:
 
My guess is that AMD want to cover their bases and release a Zen3+ on 7nm for AM4 while also releasing a Zen4 5NM on AM5 as they know people will be hesitant to jump on a new platform and ram prices will be quite high to start with so they can make up extra sales from those looking to upgrade on the large AM4 base.
 
My guess is that AMD want to cover their bases and release a Zen3+ on 7nm for AM4 while also releasing a Zen4 5NM on AM5 as they know people will be hesitant to jump on a new platform and ram prices will be quite high to start with so they can make up extra sales from those looking to upgrade on the large AM4 base.

I don't know what to think. If Raphael has an iGPU, it may be two options - the same chiplet design + iGPU chiplet or a monolithic die all-in-one.
The first option - might turn out to be too power hungry, the second option - the performance improvement won't be there because the core count will remain the same.

If they want to use the iGPU for heterogeneous computing, it's decades late - this was the initial goal of the Fusion project and Llano and still AMD Ryzens don't accelerate anything meaningful (benchmarks, Adobe, etc) with the iGPU.


I read that Vermeer can have 45-watt versions, which if true, might mean that Warhol should go back to the 95-105-watt TDPs.
 
There is one very important detail here:


https://wccftech.com/amd-ryzen-2020-2022-cpu-apu-roadmap-leak-zen-2-3-4-families-unveiled/

So, Vermeer is MSDT CPU with Zen 3, Warhol is MSDT CPU with Zen 3+, Raphael is MSDT big APU with Zen 4 and Navi of some generation.
So, probably AMD is cooking a very bad surprise for us.

What is Warhol? Just another +1-2% ala the Ryzen 3000 XT series? :confused:

Vermeer only 15% IPC improvement, Warhol +1-2% if any, Raphael - no core count improvement.

:eek:
It’s hard to gauge what AMD’s architectural refresh will bring in terms of IPC improvement. Zen to Zen+ has architectural improvements as well as node shrink. I don’t think Warhol will be like what the recent XT series CPUs. I think it will be a more aggressive than just better binned parts.
 
What is Warhol? Just another +1-2% ala the Ryzen 3000 XT series?
The XT's were not a new family arch, whereas Warhol seems to be. As @pc-guy points out, it is more likely to be like a 1000 to 2000 jump.

Vermeer only 15% IPC improvement,

They are aiming for 15% IPC, and a clockspeed jump, so performance may be around 20%.

Raphael - no core count improvement.

There are no details about Rapheal or its core count, so why are you saying no improvement?
 
The XT's were not a new family arch, whereas Warhol seems to be. As @pc-guy points out, it is more likely to be like a 1000 to 2000 jump.

Which, technically, is an IPC improvement in the statistical error margin. 1%.


https://www.guru3d.com/articles_pages/amd_ryzen_7_2700x_review,9.html



There are no details about Rapheal or its core count, so why are you saying no improvement?

I am saying so because the slide above says Raphael will be an APU, which means that we, with 90% certainty or more, can predict that it won't have 16 cores slapped to an iGPU of some Navi type.
 
Which, technically, is an IPC improvement in the statistical error margin. 1%.
IPC isn't the only measure of performance when comparing arch progressions tho? The boost from that change was higher clocks, reduced latency and the better memory controller. And the step up was significant.

I am saying so because the slide above says Raphael will be an APU, which means that we, with 90% certainty or more, can predict that it won't have 16 cores slapped to an iGPU of some Navi type.
The certainty isn't that high given they already have a 8 core, 16 thread igpu enabled processor in a 35W package, and companies are already putting 3950X's in laptops. You are talking about a future APU in 2 years, possibly under 5nm that we have absolutely no details on, so it seems bizarre to claim no increase in two years time.
 
Exactly. Lots of people saying PCIE 4.0 isn't needed yet whilst others saying PCIE 5.0 is a must have. They can't both be right!

I guess it might be worth it for that niche group that are graphic artists or something transferring massive files or people with huge 8K or 16K video files. For GPUs it doesn't seem necessary (unless, I suppose, we go to multiple GPUs per card). There was a Hardware Unboxed video yesterday that showed even PCIe 3.0 x8 didn't affect GPU performance too badly, especially as resolutions increased.
 
Back
Top Bottom