AMD's Future Chips & SoC's: News, Info & Rumours.

Page 80 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.

jdwii

Splendid


That's some really good news to hear maybe Amd should focus on IPC and frequency now and lay off increasing the core count alone?
 

goldstone77

Distinguished
Aug 22, 2012
2,245
14
19,965


If you watch AdoredTV's last video about chiplets, here, you will find that the industry demand for more cores is increasing dramatically, so you can expect to see a constantly increasing amount of cores. They will of course continue to push IPC as well with each new uArch, like Zen 2. What we will also see in the future is heterogeneous computing, and chiplets on a package with active interposers. Active interposers offer lower latency over monolithic dies.
NXJ8QV0.png

The increasing demand for core/socket counts for the data center requires many more cores in the near future.
tpileWH.png

Links to sources
https://www.semiconductors.org/clientuploads/Research_Technology/ITRS/2015/0_2015%20ITRS%202.0%20Executive%20Report%20(1).pdf
page 12
http://www.eecg.toronto.edu/~enright/Kannan_MICRO48.pdf
http://www.eecg.toronto.edu/~enright/micro14-interposer.pdf
 

goldstone77

Distinguished
Aug 22, 2012
2,245
14
19,965
I posted this in the Intel thread, but thought I would share it here as well. Ran across this video today, great video if you have the free time 41 minutes long, computer buffs might recognize the name John Hennessy ;)
[video="https://www.youtube.com/watch?v=Azt8Nc-mtKM&feature=youtu.be"][/video]
The future of computing: a conversation with John Hennessy (Google I/O '18)
 

goldstone77

Distinguished
Aug 22, 2012
2,245
14
19,965
New AMD Semi-Custom SOC Combines the Power of AMD Ryzen CPU and AMD Vega GPU for Gamers in China
Posted by jack.huynh Employee in Gaming on Aug 2, 2018 10:10:37 PM

Today, we are excited to detail how we are bringing Ryzen processors and Radeon graphics to even more gamers through our work with Zhongshan Subor to create a new gaming PC and upcoming console for the China market. The PC and console are powered by a semi-custom “Zen” and “Vega”-based AMD SOC. The high-performance AMD chip combines an AMD Ryzen™ CPU (4 core/8 thread running at 3Ghz) with AMD Radeon™ Vega Graphics (24CUs running at 1.3Ghzs) and 8GB of GDDR5 memory onto a single chip.
https://community.amd.com/community/gaming/blog/2018/08/03/new-amd-semi-custom-soc-combines-the-power-of-amd-ryzen-cpu-and-amd-vega-gpu-for-gamers-in-china?sf194826771=1
 

goldstone77

Distinguished
Aug 22, 2012
2,245
14
19,965
Lisa Su is a one smart cookie, hitting segments she knows she can win with the least effort possible. She flipped out a China card from here back pocket. "The Art of War" is probably kept on her night stand.
 

jdwii

Splendid


More than anything I just want AMD to beat or meet Intel's single threaded performance and have full support for AVX2 not even for reasons that I need that but more over the fact that it will finally make Intel 100% irrelevant at least for a while so Intel moves faster to compete.

 

goldstone77

Distinguished
Aug 22, 2012
2,245
14
19,965
Intel has no chance in servers and they know it
Numbers straight from the horses mouth
Aug 7, 2018 by Charlie Demerjian

SemiAccurate has laid out a pretty stark picture of Intel’s performance and market competitiveness over the next 3-4 years. We are highly confident in the information presented because the majority of it comes from Intel’s internal outlook and documentation that was shown to us. They know they have no chance in their most lucrative core market, and are trapped between raising prices to keep margins up and cratering marketshare. Either way they lose because they aren’t close in performance
https://www.semiaccurate.com/2018/08/07/intel-has-no-chance-in-servers-and-they-know-it/

I originally posted this link, here.
 


There's the Charlie we know and love. But in three years from now, when Intel share is about what it is now, no one will remember this.

Seriously, I read this same exact article in 2015.
 

goldstone77

Distinguished
Aug 22, 2012
2,245
14
19,965


Add to that the sure clock increases AMD will bring to their Zen 2 parts - you can bet the power consumption and top frequency benefits for the 7 nm manufacturing process (some estimated 1.6X logic density, ~20% speed improvement, and ~40% power reduction compared to TSMC's 10 nm process) won't go to waste when they can be used for an even greater performance uplift.

It's hard to say what exactly the performance will be from GF 14nm LPP to TSMC 7nm, assuming Desktop processors are made at TSMC. Density and lower consumption/per core are going to be the expected advantages. Lisa Su promised some performance numbers at the end of the year, but she seemed excited about Epyc.
 

jdwii

Splendid


Basically once Amd does that and Intel brings their usual 2-3% IPC improvement on average with some being 10% for operations not too many people use, Amd will simply be king. I already hear it from almost all the major tech-tubers Paulshardware, Bitwit to name a few.

Intel IPC is basically the only thing they have left as their main advantage over Amd and AVX2 support.

 

jdwii

Splendid


Knew they would be the first one to go under haha. Basically what do they do now?
 


Focus on other nodes, which non-high performance CPUs will continue to use. There's a reason why Intel still has low rate production on its 450 and 300nm nodes after all.

But yeah, it basically comes down to ROI, and the foundries are starting to realize the ROI really isn't there.

Does explain why AMD went to TSMC though.
 

jaymc

Distinguished
Dec 7, 2007
614
9
18,985
AMD’s next-gen CPUs, GPUs will all be built on TSMC’s 7nm process:

"The new manufacturing agreement leaves open the possibility that AMD could turn to Samsung for some amount of its 7nm manufacturing. While GloFo was used for all the shipping Zen-based processors, AMD did validate Samsung's process and successfully build Ryzen chips on it, retaining Samsung as a fallback if needed. A similar situation could emerge for 7nm, though there are wrinkles."

Looks like Samsung will be another option... also looks like no fee's from GF for going elsewhere for 7nm which is great... WSA is more or less void for next year. Currently being renegotiated. This could be a massive blessing in disguise guy's... no penalties for going elsewhere an no quota to meet next year at GF or much less, maybe just some 12 / 14nm chips that are already in production. Numbers will be drastically reduced I'd say as AMD starts to move away from GF an become a proper freelance Semi Company. Free from the shackles of their missteps an delays sounds great (in theory anyway)... lets see how it plays out in practice..

AMD is finally free all thanks to Dr Su's modification to the WSA agreement. They are breaking the deal an voiding the contract... I initially thought it was a disaster but hey If they can get capacity from TSMC it may even bring desktop chips to market sooner as well. Actually look like Lisa Su was expecting this to happen an was well prepared... well remains to be seen I guess. But we shall see.

https://arstechnica.com/gadgets/2018/08/amd-to-use-tsmc-to-make-its-7nm-cpus-gpus-as-glofo-abandons-7nm-development/
 

aldaia

Distinguished
Oct 22, 2010
533
18
18,995
Intel can’t supply 14nm Xeons, HPE directly recommends AMD Epyc
https://semiaccurate.com/2018/09/07/intel-cant-supply-14nm-xeons-hpe-directly-recommends-amd-epyc/
 

aldaia

Distinguished
Oct 22, 2010
533
18
18,995
AMD Ryzen 7 2800X spotted in the wild
Ryzen-7-2800X-Cinebench-740x458.jpg


However I have some concerns with that rumor. We all know that the zeppelin die has 8 cores. The alternatives in decreasing order of plausibility are:


    ■ Its a Fake
    ■ AMD will use 2 Zeppelin dies inside an AM4 package. But why only 10 cores? That means 5 cores per die, so the number of cores per CCX is unbalanced. I will find more plausible if it was 12c
    ■ It's a new 12 nm die. I find it hard to believe that AMD evelopped a new 10 c die in 12 nm just for bragging rights.
    ■ It's actually a 7nm die. Again if find it hard to believe that 7nm is labeled 2800x and not 3600x or something similar.
    ■ Any other possibility i didn't thought of?

 
There was an old rumour about AMD making 6-core CCX'es, so this could be their testing mules for that and trying them out in 12nm for the consumer market and 7nm parts will be 8core CCX'es for Rome and other EPYC CPUs. As in, stop gap model (one of the potential reasons they left the 2800X name out of this line up) before moving onto 3K-X series.

EDIT: They being 10-core parts, could be because they wouldn't enable the 6th core reliable in the CCX due to manufacturing reasons? That is, they could be (like you say) 5 core CCX'es.

On the other hand the image, although pretty damn good, has some artifacts in it, so the chances of it being fake are still high.

Cheers!
 

jaymc

Distinguished
Dec 7, 2007
614
9
18,985
It could be fake alright... but at the same time it could be a 6c CCX paired with a 4c CCX....for the 2800x

Could they eventually release a 2x 6c CCX as a 2900x in that case, would that be possible power wise ?

I know lots of if's and maybe's right...
 

jaymc

Distinguished
Dec 7, 2007
614
9
18,985
That makes sense I guess. I have a feeling we might see 6c CCX's for some reason. I wonder will they end up doing a 12c desktop part just to screw with Intel, and also just because they can l guess.
 

aldaia

Distinguished
Oct 22, 2010
533
18
18,995
Acording to adoredTV EPYC 2 Rome will be 9 dies. 8 tiny core dies (with 8 cores each) + 1 uncore die
If that is true, expect one of those tiny dies to be the basis of ryzen 3 series processors.

Annother interesting rumor is that Zen2 core is not an evolution of Zen core, but a total new design from the ground up. So it is even possible that the new design dosn't even have CCXs

https://www.youtube.com/watch?v=KVXDOWy4vTU