The AMD Execution Thread [2020]

  • Thread starter Deleted member 2197
  • Start date
Status
Not open for further replies.
Mercury Research: AMD Achieves 22.4% Share Of The X86 Market, A High Not Seen Since 2007
November 3, 2020
Mercury Research has revealed (via Tom's Hardware) that AMD has captured a resounding 22.4% of the x86 market. This is a high that hasn't been seen since 2007 and showcases AMD at its prime. These numbers come just before AMD's Ryzen 5000 series are due to hit the market which means we will likely see AMD take a 25% (or higher) share of the x86 market by next year. With Intel's Rocket Lake still stuck on the 14nm node, AMD is in a position to strike while the iron is hot - and it appears to be doing just that.
https://wccftech.com/mercury-resear...of-the-x86-market-a-high-not-seen-since-2007/
 
Mercury Research: AMD Achieves 22.4% Share Of The X86 Market, A High Not Seen Since 2007
November 3, 2020

https://wccftech.com/mercury-resear...of-the-x86-market-a-high-not-seen-since-2007/

Those numbers don't make sense. The report says overall market share is 22.4% but desktop, notebook and client are 20.1%, 20.2% and 20.2% respectively. And server is 6.6% (though if I'm not mistaken AMD had claimed 10% a few quarters back), so unless I'm missing something, how is the overall share 22.4%?

Still great news for AMD and their market share should grow even further with Zen 3 and well into 2022 with Zen 4. AMD should also get $1B+ USD in supercomputer revenue in 2021-2022.
 
Zen 3 is here and it indeed delivered.

Seeing the massive improvements (10-wide INT and 6-wide FP engine!) delivered at the same node, Zen 4 at 5nm sets expectations pretty high.
 
I wish I had a reason to splurge on these massive core CPUs. Alas the only heavy workload I run is gaming and that runs fine on any decent CPU released in the past 5 years.
 
I wish I had a reason to splurge on these massive core CPUs. Alas the only heavy workload I run is gaming and that runs fine on any decent CPU released in the past 5 years.

Gaming performance (especially if you're running on say 1080p) is significantly improved on Zen 3 though. Even a flagship 7700K from 2017 is outclassed by a significant margin.
 
Yes, Zen3 for games is on another level (game dependent on by how much) compared to my Zen 2 3900XT.
One small example is CS:GO where I play in QHD on Radeon VII and was getting around 200FPS on Dust II average and 5900XT power limited to 65W produces easy 300FPS+
In Cinebench R15 when limiting power to 45W I'm getting almost identical performance to my fully tweaked and overclocked to 4.1GHz Threadripper 1920X which was drawing 180W for that task :D
Don't even ask how fast it is at 4.75GHz all cores
 
Good interview with Rick Bergman here - https://www.thestreet.com/investing/amds-rick-bergman-talks-about-current-and-next-gen-cpus-and-gpus

Couple of interesting points:-

AMD’s relationship with TSMC , and whether AMD is still seeing supply tightness for 7-nanometer TSMC wafers.

“In terms of specific product lines, I can't really give specifics on where we're seeing tightness...But yes, we would like to have more wafers going forward, to help the growth that we have in front of us. From a worldwide perspective, there continues to be tightness. But we're working through that with the TSMC, and we’ll see what 2021 brings us.”

How much of the performance gains delivered by AMD’s Zen 4 CPUs, which are expected to use a 5nm TSMC process and might arrive in early 2022, will come from instructions per clock (IPC) gains as opposed to core count and clock speed increases.

Bergman: “[Given] the maturity of the x86 architecture now, the answer has to be, kind of, all of the above. If you looked at our technical document on Zen 3, it was this long list of things that we did to get that 19% [IPC gain]. Zen 4 is going to have a similar long list of things, where you look at everything from the caches, to the branch prediction, [to] the number of gates in the execution pipeline. Everything is scrutinized to squeeze more performance out.”

“Certainly [manufacturing] process opens an additional door for us to [obtain] better performance-per-watt and so on, and we'll take advantage of that as well.”


Whether AMD is aiming for its RDNA 3 GPUs, which will use a more advanced manufacturing process, to deliver performance-per-watt improvements similar to the 50%-plus improvements delivered by its RDNA 2 GPUs, and its future plans for the Infinity Cache technology used by RDNA 2 GPUs.

Bergman: “Let's step back and talk about the benefits of both. So why did we target, pretty aggressively, performance per watt [improvements for] our RDNA 2 [GPUs]. And then yes, we have the same commitment on RDNA 3.”

 
I'm a little bit confused about AMD's roadmap. Has the company said anything about something like a Zen3+ refresh for 2021?
 
Officially, not a thing except Zen 4@5nm being a <= 2022 architecture.

However, there is still that roadmap presenting Warhol as the "Zen3 refresh" for 2021. It kinda makes sense - pipecleaning the AM5 platform before Zen 4, having a new "gen" of Ryzen products for 2021, etc.

I don't know if we'll actually see DDR5 on consumer platforms in 2021. Given the likely high initial price, servers are more likely to be the early adopters. Warhol might just be an XT style refresh of Vermeer. Though there were some rumours of Zen3 on 5nm which might indicate the possibility of a simple shrink and pipecleaner for 5nm/AM5 as you suggest.
 
I don't know if we'll actually see DDR5 on consumer platforms in 2021. Given the likely high initial price, servers are more likely to be the early adopters. Warhol might just be an XT style refresh of Vermeer. Though there were some rumours of Zen3 on 5nm which might indicate the possibility of a simple shrink and pipecleaner for 5nm/AM5 as you suggest.
Yea, DDR5 initial release will probably be the classic story of slow, expensive, and non-available memory. However, Intel aggressively pushes it's DDR5/PCIe5 consumer platform (LGA 1700/Alder Lake) for H2 2021. AMD might simply follow that.

Also there are really many possibilities for Warhol:
* no Warhol at all
* just laughable +100MHz XT models
* shrunk IO die to 7nm but still with DDR4
* shrunk IO die to 7nm with DDR5
* shrunk IO die to Zen 4's 5nm with DDR5
* shrunk compute dies to a better variant of 7nm
* shrunk compute dies to Zen 4's 5nm
 
Yea, DDR5 initial release will probably be the classic story of slow, expensive, and non-available memory. However, Intel aggressively pushes it's DDR5/PCIe5 consumer platform (LGA 1700/Alder Lake) for H2 2021. AMD might simply follow that.

Also there are really many possibilities for Warhol:
* no Warhol at all
* just laughable +100MHz XT models
* shrunk IO die to 7nm but still with DDR4
* shrunk IO die to 7nm with DDR5
* shrunk IO die to Zen 4's 5nm with DDR5
* shrunk compute dies to a better variant of 7nm
* shrunk compute dies to Zen 4's 5nm
I can pretty much guarantee there won't be 5nm IO die anytime soon. 7nm maybe, but definitely not smaller, and I wouldn't be be holding my breath for 7nm either.
 
I can pretty much guarantee there won't be 5nm IO die anytime soon. 7nm maybe, but definitely not smaller, and I wouldn't be be holding my breath for 7nm either.

Yea the earliest we'll see a 5nm IOD is when we see a 3nm chiplet. For volume and cost reasons, the IOD will always be at least a node behind the chiplets. A 6nm/7nm IOD in 2022 might be feasible as 5nm capacity will be much higher by then. If AMD is using TSMC 3D packaging technology for Zen 4, I'm not sure what other options they would have.
 
I can pretty much guarantee there won't be 5nm IO die anytime soon. 7nm maybe, but definitely not smaller, and I wouldn't be be holding my breath for 7nm either.
Aren't the IO dies made on Globalfoundries' 12nm the way that AMD found to not break the supply agreement they have with the foundry?
I don't remember how far the agreement goes, but I think it was something like 2023.

And IIRC GF isn't going below 12nm any time soon, at least according to their public roadmaps.
 
Aren't the IO dies made on Globalfoundries' 12nm the way that AMD found to not break the supply agreement they have with the foundry?
I don't remember how far the agreement goes, but I think it was something like 2023.

And IIRC GF isn't going below 12nm any time soon, at least according to their public roadmaps.

The overall WSA is valid till March 2024 but the current purchase commitments are only through 2021 with further wafer purchases from 2022-24 to be negotiated. AMD did renegotiate the WSA favorably when GF stopped development of 7nm. Couple of other points such as not being required to make any payments to GF for use of 7nm and smaller nodes and also reducing the penalties payable to GF in case they don't meet the agreed wafer volumes were also agreed. Full details here - https://www.anandtech.com/show/1391...th-globalfoudries-set-to-buy-wafers-till-2021

The consumer IO dies are on 12nm and the server dies on 14nm, but AMD will continue to require significant amounts of these wafers through 2021 and beyond. Zen 3 will sell for a while due to the DDR5 transition, at least into 2023. And they have a bunch of earlier gen long-life pro and embedded parts which they will continue to manufacture. So while 2022 shouldn't be an issue, by 2023 the volumes will certainly start coming down I guess.

If they do start using TSMC 3D stacking tech, they will necessarily have to use TSMC processes for the IO dies. The added power benefits of having an IO die on a more advanced process are also compelling, especially for the server IO dies which are pretty large and power hungry. As I've mentioned, I could see them going for a 6nm/7nm IO die for Zen 4 in 2022 as adequate 7nm capacity (and lower pricing) should be available by then. For the consumer parts they could conceivably continue using 2.5D tech and possibly GF's newer 12LP+ process which offers some of the benefits of a 7nm class process. AMD have not yet negotiated those agreements though and I'm sure they'll factor all of these possibilites into their plans when they negotiate the volumes for 2022-24. Even if they don't meet the volumes though, as mentioned, the penalties are smaller than they were in the earlier WSA so overall I don't see the WSA being a major problem. The wafer volume they can secure from TSMC is the bigger concern.
 
Status
Not open for further replies.
Back
Top