But there is stories that the size from different makes don't always equal the same on another makers process.
I not an chip technical manufacturer ( even though I did take some classes in micro code in college ) - but maybe a better indication of process is taking the density of transistors per square area.
Agree HStewart, density would be a better measure as the nm feature size is mostly marketing since only some features are at that size and with these "3D" chips there is so much more to the resulting density that I would much rather they market it that way. At the end of the day they can call it whatever they want, the people paying to have chips made on this process will know exactly what kind of densities they will get even if none of us get to see that data.
If you look at above link - the die size is pretty close - but it also appears that AMD has a lot of wasted space - not referring to Zen modules not in use. I don't know the exact transistor counts - but it looks like Intel has more transistors for each related size.
To AMD's merit, I would think it cost less to engine modules to fit together instead having more cores - it would be cheaper processed - to me it kind of like multi-cpu server box using single chip. So a dual cpu 32 core AMD Server chip is kind of like a 8 cpu 8 core MP system.
I believe Microsoft had to special changes to it operating systems so zen architecture does not count as multi-cpu setup.
We live in a world where TSMC is upset because their "16" is higher than their competitor's "14", so they renamed it to "12" because "12" is less than "14".
Yes ... also change frequency to be higher than competitors thinking that more powerful and ignoring features on chip.
Also lately it seams to be related to number of core - ignoring the speed of actual cores. Quad core in one manufacturer does not always equal quad core in competitor or even dual core - ie ARM vs x86 CPU.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
11 Comments
Back to Article
wiineeth - Thursday, September 14, 2017 - link
Is Nvidia volta going to be based on 7nm?edzieba - Thursday, September 14, 2017 - link
GV100 is on 12nm FFN, so smaller Volta dies are likely to use the same process.tipoo - Thursday, September 14, 2017 - link
And 12nm is an extension of their 16nm fwiw. In intel terms it would probably be the ++ node.Qwertilot - Thursday, September 14, 2017 - link
Not likely - V100 isn't. Son of Volta presumably will.tipoo - Thursday, September 14, 2017 - link
If 14nm is actually a 20nm back end, what exactly does "7nm" mean?HStewart - Thursday, September 14, 2017 - link
Using what you just wrote ( 20/14 * 7 ) = 10But there is stories that the size from different makes don't always equal the same on another makers process.
I not an chip technical manufacturer ( even though I did take some classes in micro code in college ) - but maybe a better indication of process is taking the density of transistors per square area.
FreckledTrout - Thursday, September 14, 2017 - link
Agree HStewart, density would be a better measure as the nm feature size is mostly marketing since only some features are at that size and with these "3D" chips there is so much more to the resulting density that I would much rather they market it that way. At the end of the day they can call it whatever they want, the people paying to have chips made on this process will know exactly what kind of densities they will get even if none of us get to see that data.HStewart - Thursday, September 14, 2017 - link
Agreed - one example of this is AMD Threadripper vs Skylark Xhttp://digiworthy.com/2017/06/22/intel-skylake-x-v...
If you look at above link - the die size is pretty close - but it also appears that AMD has a lot of wasted space - not referring to Zen modules not in use. I don't know the exact transistor counts - but it looks like Intel has more transistors for each related size.
To AMD's merit, I would think it cost less to engine modules to fit together instead having more cores - it would be cheaper processed - to me it kind of like multi-cpu server box using single chip. So a dual cpu 32 core AMD Server chip is kind of like a 8 cpu 8 core MP system.
I believe Microsoft had to special changes to it operating systems so zen architecture does not count as multi-cpu setup.
MrSpadge - Thursday, September 14, 2017 - link
http://semiengineering.com/a-node-by-any-other-nam...So transistor gate length of 14 - 18 nm, depending on size and way of counting.
ImSpartacus - Friday, September 15, 2017 - link
The numbers mean nothing. It's all marketing.We live in a world where TSMC is upset because their "16" is higher than their competitor's "14", so they renamed it to "12" because "12" is less than "14".
HStewart - Friday, September 15, 2017 - link
Yes ... also change frequency to be higher than competitors thinking that more powerful and ignoring features on chip.Also lately it seams to be related to number of core - ignoring the speed of actual cores. Quad core in one manufacturer does not always equal quad core in competitor or even dual core - ie ARM vs x86 CPU.