AMD Zen 2 CPU Yields are Potentially Twice that of Intel’s Top Server Chips

naib

naib

[H]ard|Gawd
Joined
Jul 26, 2013
Messages
1,289
https://www.pcgamesn.com/amd/zen-2-ryzen-3000-cpu-yield-70-percent


Combining a new production process with a high-volume new processor architecture is a bit of a gamble, but it looks like it might be paying off with the upcoming AMD Ryzen 3000 series CPUs as early yields are looking good. An anonymous source is stating that AMD’s 7nm processor yields are sitting around the 70% mark, and at this stage in production that’s actually a pretty good figure.

The chip yield is one of the most important metrics in terms of silicon production. If your manufacturing process is delivering high yields that means a greater percentage of the chips on an individual wafer are deemed functional. There will always be defects in such precise manufacturing, so some of the chips on any given wafer will be dead on arrival, but however much you cut that down increases your profitability.


So 35% yield versus 70% yield on both Intel and AMD’s most expensive professional server parts respectively. Guess who’s going to be just printing dollars with those numbers…​
 
auntjemima said:
"anonymous" and "potentially" in the same OP? I'll wait.
Click to expand...
Ah, yes, we won't know until actual numbers are released, if they ever are. News like this could be used to potentially boost AMD stock prices in the short term, and if they don't come true, then see a sinking of the stock price.

Or, there could be some truth to this. AMD's execution in the CPU market the last couple of years has been on a level rarely seen by any manufacturer. That they could do something fantastic with chip yields would not be surprising.
 
jardows said:
Ah, yes, we won't know until actual numbers are released, if they ever are. News like this could be used to potentially boost AMD stock prices in the short term, and if they don't come true, then see a sinking of the stock price.

Or, there could be some truth to this. AMD's execution in the CPU market the last couple of years has been on a level rarely seen by any manufacturer. That they could do something fantastic with chip yields would not be surprising.
Click to expand...

Highly agree. They have made some incredible leaps and bounds coming from bulldozer. I don't think anyone can disagree, honestly that is, with that statement. That alone gives weight to positive potential news.

And again, as you said, it could just be an "anonymous" source used to raise stock prices ahead of Ryzen 3000.
 
Some of these things have nearly no impact on consumers. What it does not say is if the yield is all of the top end chiplets or chiplets that will be usable in the whole Ryzen 3000 range.

The way now is that the number really is trivial it could mean that 69% are all 6 core parts that are good and the last 1% are only good 8 core parts. So this numbers game can mean anything :)
 
Pieter3dnow said:
Some of these things have nearly no impact on consumers. What it does not say is if the yield is all of the top end chiplets or chiplets that will be usable in the whole Ryzen 3000 range.

The way now is that the number really is trivial it could mean that 69% are all 6 core parts that are good and the last 1% are only good 8 core parts. So this numbers game can mean anything :)
Click to expand...

Absolutely, cause if this were Intel or Nv it would become "we have near perfect yields" what they do not say is like you put above, could be far more drastic going by their combined histories (yay crud marketing and bias analyst/relations)

"70% yield on all the product stack exceeding 90% on our top product for the upcoming launch" (truth in the numbers, in fact, only 1% make this "90% top yield" with the rest being thrown into the 70% so actuality is 1% usable at wanted spec and at least 30% are completely unsalvageable with no room to redce this as we see at this time)

marketing will always find a way to spin things, analyst are huge bias by nature of the jobs they do, leaks are leaks and can be truth of a trickle or mountains of sewage sprung from the dike for a number of reasons such as to pump before dump huge amount of stock or cause others to do something they will regret etc, such will never change, that being said, from my understanding (beyond the occasional larger dies)

AMD and Radeon in general for many generations have actually had acceptable to amazing yields in comparison to Nv or Intel, obviously this does not always lead to $$$$$$$$$$$ in their bank account (see the need to fight all the bias, anlyst etc)

All of this for quite a long while AMD had TSMC 7nm "all to themselves" so likely they got some much needed time to "tweak" things before going full out on their Ryzen/Navi etc plans for 2019 on 7nm, Apple and others WERE supposed to be "printing off" but may of them did not see sales they wanted last years and so held off till 2Q/19 if not later before production runs.

Either a very very fortunate set of circumstances for AMD and/or the industry said "we need AMD to get some $ in the bank and lead the charge for beyond 2020"

I know they are all mega business however the "end of the road" is fast approaching and the need and work with each other daily, gpu and cpu wise is a very "small" landscape with massive whales that need hunt the oceans together
:)
 
Last edited:
Lower pricing would be nice.

Wonder what chiplet tech is going to do for SOC's?
 
Last edited:
the source, Bits and Chips, has no credibility

It was hyping the s*** out of Vega.

...and we all know how Vega turned out
 
Last edited:
dragonstongue said:
Absolutely, cause if this were Intel or Nv it would become "we have near perfect yields" what they do not say is like you put above, could be far more drastic going by their combined histories (yay crud marketing and bias analyst/relations)

"70% yield on all the product stack exceeding 90% on our top product for the upcoming launch" (truth in the numbers, in fact, only 1% make this "90% top yield" with the rest being thrown into the 70% so actuality is 1% usable at wanted spec and at least 30% are completely unsalvageable with no room to redce this as we see at this time)

marketing will always find a way to spin things, analyst are huge bias by nature of the jobs they do, leaks are leaks and can be truth of a trickle or mountains of sewage sprung from the dike for a number of reasons such as to pump before dump huge amount of stock or cause others to do something they will regret etc, such will never change, that being said, from my understanding (beyond the occasional larger dies)

AMD and Radeon in general for many generations have actually had acceptable to amazing yields in comparison to Nv or Intel, obviously this does not always lead to $$$$$$$$$$$ in their bank account (see the need to fight all the bias, anlyst etc)

All of this for quite a long while AMD had TSMC 7nm "all to themselves" so likely they got some much needed time to "tweak" things before going full out on their Ryzen/Navi etc plans for 2019 on 7nm, Apple and others WERE supposed to be "printing off" but may of them did not see sales they wanted last years and so held off till 2Q/19 if not later before production runs.

Either a very very fortunate set of circumstances for AMD and/or the industry said "we need AMD to get some $ in the bank and lead the charge for beyond 2020"

I know they are all mega business however the "end of the road" is fast approaching and the need and work with each other daily, gpu and cpu wise is a very "small" landscape with massive whales that need hunt the oceans together
:)
Click to expand...

For the cpu the 7nm process seems to be very good sadly that is due to other circumstances on the gpu side it is a whole different story that might work out this year but next year AMD would need the improved/fixed gpu architecture.
 
Meeho said:
Of course they will be better with chiplets vs large monolithic dies.
Click to expand...


Right, when you need 4 of the chiplets to make the same performance processor, it's a lot easier to have high yields.

I think the question that must be answered is how many equivalent packaged 4-chiplet processors is AMD getting from that set of yields, in comparison to Intel's 28-core processors?
 
Pieter3dnow said:
For the cpu the 7nm process seems to be very good sadly that is due to other circumstances on the gpu side it is a whole different story that might work out this year but next year AMD would need the improved/fixed gpu architecture.
Click to expand...

Yes, absolutely, even if Navi turns out to be as expected from leaks "a shocker" that will NOT last and something has to be that much better to replace it because Nv and Intel are right there waiting licking their chops after a bad lunch or a few previously.

AMD did excellent on Zen architecture, AM4 in general, TR, EPYC, Navi has to be a true "stop gap" (current per shr prce (far too low in my opinion at this point considering turn around) ~$27.68 .. "great" that things "worked as they did" the little stumbles such as R 2xxx not "as good as should have been" (with massive inventory buy back problems and excessive amount) and Intel while "potent" has had numerous problems with yield as everyone knows, very unlikely mass market 10nm high performance before end of 2019 (unless I am wrong)

this has helped AMD make some much needed bank on a very great company turn around, but, they NEED to keep the pressure on, any stumble and the blade will bite in their neck deep, they cannot afford that price unlike Nv or Intel can (can sell portions of company or go red many quarters/years and live (especially intel as they have many billion dollar fabs they can sell at moments notice) AMD what 1 year maybe 2 (now) red tops with next to nothing more to "sell away" (frying pan vs walls thick as castles)

2Q on Apr 30 will show good horizons for AMD "value" and unless ppl are truly stupid should be ~$30, summer and into fall will be a "test" ($35 by August my estimate)
everything after that is still very much up in the air not a person alive can dispute that (well, unless higher ups know this lul was designed on purpose)
 
What includes the 70%? If a chip has defects but is still usable with reduced cores/features is that a 70% passed chip? Or is that 30% failed chip?
 
whateverer said:
Right, when you need 4 of the chiplets to make the same performance processor, it's a lot easier to have high yields.

I think the question that must be answered is how many equivalent packaged 4-chiplet processors is AMD getting from that set of yields, in comparison to Intel's 28-core processors?
Click to expand...
It's not quite as important, though, as they still have all those usable chiplets for the whole product stack. And there is much higher chance of getting 4 excellent small parts than one large one, all else being equal, and TSMC's 7 nm seems to be relatively mature at this point. Going with chiplets was genious and might prove crucial for AMD's survival.
 
I believe it. Having smaller die sizes allows em to get more out of a wafer. less waste from si deformities. it's honestly pretty brilliant.
 
whateverer said:
Right, when you need 4 of the chiplets to make the same performance processor, it's a lot easier to have high yields.

I think the question that must be answered is how many equivalent packaged 4-chiplet processors is AMD getting from that set of yields, in comparison to Intel's 28-core processors?
Click to expand...

Not sure I get your math here ? You need more chiplets yes but you also produce more with the same components, the only change is the yield. So basically let's say you produce 100 Intel using 1 wafer but would produce 400 chiplets. You only get 30 workings Intel CPU while you have 280 chiplets / 4 = 70 working CPUs. This is grossly explained because you also need to factor 14+++++ nm VS 7nm which probably means even more CPU on the same wafer ;)
 
You must log in or register to reply here.
Back
Top