By using this site, you agree to our Privacy Policy and our Terms of Use. Close
Intrinsic said:
  1. This will be either a 72CU@1500Mhz GPU or a 80CU@1385Mh. Both configs I believe are possible if lookin at the same kinda progression seen from 28Nm to 16nm and from that to 7nm. Also expected if AMD has broken that 64CU limitation

Or it could be 64CU's @ 1700mhz.

Intrinsic said:
  1. those clocks are high though, which makes me lean towards the 80CU setup as its usually better to have more CUs running at a lower clock than less at a higher clock. It would also suggest that Navi is very efficient, we have nvidia cards that clock as high as 2Ghz and there are rumblings that desktop Navi cards should be expected to clock that high..... so a downclocked Navi at ~1300Mhz down from 1800-2000Mhz doesn't sound outta reason.

Not always better to have more CU's running at a lower clock than less CU's at a higher clock.

Vega 7 has less CU's than Vega 64 and yet offers a substantial uptick in performance.

The thing to keep in mind is that the entire chip runs at the core clockrate or a tied multiplier of it... Throwing more CU's at the problem doesn't increase things like ROP performance which is super important... But clockspeed does.

Plus Graphics Core Next tends to be ROP and Geometry starved... So Clockrates help with that.

It also helps keep the chips small... So if 7nm has good clocking characteristics, but poor yields with larger chip sizes, then having a smaller CU count but higher clockrates will make the chips more economical. (Sometimes the reverse holds true as well though, depends on the process.)

Navi isn't likely to change how poorly inefficient Graphics Core Next is though... AMD had big promises with Vega... Then Polaris before that... But at the end of the day, Graphics Core Next is still Graphics Core Next with all it's Pro's and Con's.

Navi to me isn't next-generation graphics technology.

Intrinsic said:
  1. I expect the 4GB DDR4 OS RAM. I don't know about the 24GB GDDR6 though..... thats either 672GB/s (14Gbs chips), 768GB/s (16Gbs chips) or 864GB/s (1 Gbs chips)  

I think Microsoft will opt for a single DRAM pool to keep complexity down... That way when higher density DRAM chips come out they can take advantage of that, besides their Operating Systems tend to have the edge over the other console manufacturers in terms of memory consumption anyway.

Sony though may see what they did well with the Playstation 4.. And go down a similar route and include an additional DRAM pool+ARM CPU cluster.

Should be interesting either way.

Nate4Drake said:

Believe me, I was very close to quote you on this :D   I've been thinking about your post and specs prediction.

---------------------------------------------------------------------------------------

   OK, who of you has enough technical understanding to explain to me the content of a recent PATENT from Mark Cerny, in few basic words ?!   

 From what I have understood, It concerns " the heavy use of caches for GPU operations", in order to greatly speed-up graphics application; give me a break if I'm wrong.

""According to aspects of the present disclosure, a computer graphics processing method may include writing vertex indices output from a vertex shader to a cache; accessing the vertex indices written to the cache with a pixel shader; and accessing vertex parameter values associated with the vertex indices from a memory unit with the pixel shader.""

  https://patents.justia.com/patent/20190035050

---------------------------------------------------------------------------------------

  Plenty of leaks will come out, as expected; some are in line with the expectations of some of us, and the same sound "too much" for others; what I find interesting is discussing about new tech which can be related to Next Gen, I love this.

Could be primitive shader related... It's pretty ambiguous either way and I wouldn't read to much into it right now.

Nate4Drake said:

  There's also been a recent development which refers to GDDR6: 24GB on a 384bit bus can reach 864 GB/s.

 ""Samsung Officially Begins Mass Production of GDDR6 Memory – Reaching Speeds of Up To 18 Gbps and 16Gb Density To Produce High-End Graphics Cards With Up To 24 GB VRAM, 864 GB/s Bandwidth""

https://wccftech.com/samsung-gddr6-16gb-18gbps-mass-production-official/

 I can't wait for an Official announcement.   It cannot come soon enough.

Still very much doubt we will get anything more than 16Gbps chips on a 256bit bus... Cost is still an important factor with consoles... I wouldn't be surprised if they leveraged a crossbar either to that end.



--::{PC Gaming Master Race}::--