Let the discussion begin!
80 CUs
[yt]xUjaZEG4vt0[/yt]
80 CUs

[yt]xUjaZEG4vt0[/yt]
Update February 26th, 7:53 - The SK Hynix team has contacted us indeed verifying our expectations, as well the technical configuration did not make much sense. The photo is fake.:
This is **** at SK hynix Global PR Team.
We have a great concern over your recent article about the AMD specs leaked: https://www.guru3d.com/news-story/am...4-gb-hbme.html The picture or the screen capture that CyberPunkCat posted, on which your article is based, is fake. First of all, the actual specifications of our HBM2e are wrong in the document.
You can check our specs in our previous press release: https://news.skhynix.com/sk-hynix-de...-memory-hbm2e/ The size of the chip and the bus doesn’t add up, if you do the math. Also, the Korean text in the document says “With Local Communities… 2020 Happiness,” which has nothing to do with HBM2E development.
Moreover, this part actually appears on the main page of our Korean newsroom (https://news.skhynix.co.kr/) as a title for one of our content. And as you can see, you can copy and paste the FY2019 Earnings part from our English newsroom: https://news.skhynix.com/sk-hynix-in...arter-results/
I’d also have to say this is not a document format we use at SK hynix. So, overall, this whole document looks very fake, which certainly undermines the original Twitter user’s claim, and eventually the credibility of your article. We are greatly concerned that this fake news might have some negative influence in any form.
The 7nm flagship in question, which could likely be called the AMD RX 5950 XT (or anything else really, AMD is free to shake things up you know), will feature a huge 505 mm² die that based on simple maths will offer at least 2x the performance of the 'smol Navi' RX 5700 XT GPU.
The RX 5700XT contains 40 CUs based on RDNA1 architecture and the upcoming RX 5950XT (or whatever AMD decides to call it) based on Navi 21 GPU could easily contain 80 CUs based on these numbers. This would result in a grand total of 5120 stream processors. Not counting any efficiency improvements going from RDNA1 to RDNA2 (which there certainly will be), this is a performance increase of at least 2x.
Comment