cancel
Showing results for 
Search instead for 
Did you mean: 

General Discussions

more amd leaks

It seems that leaks are a dime a dozen as more and more comments out of asian hardware sources

ESFlN-7U8AAV-DR.jpg

0 Likes
4 Replies

Hmm, a picture of a screen with an Excel sheet which is not reported on any of the reputable tech sites, such as TomsHardware, Guru3D, WCCFTech, Techspot, TechPowerUp, or Anandtech...Yea, that's trustworthy.

0 Likes

at least this one is more believable showing GDDR6 etc

0 Likes

I'll tell you why I think this is BS, and why this thread should be deleted as it presents totally unsubstantiated figures you clearly just made up today:

1) The inclusion of a "6970XT" goes against AMD's stupid new naming convention of only using model numbers ending in "00".

2) It lacks any space for any meaningful non XT cards. We pretty much know Ampere is going to be extremely powerful, and that GDDR6 supply will be extremely tight, so AMD will not shave a few TMUs or ROPs to create a card which will not sell and waste memory supply. Also, being a new manufacturing node, supplies will again be tight, so AMD cannot afford to heavily segment and strangle supply.

3) It does not include the entire list of cards since we know AMD will release a top to bottom range based on RDNA2 per Lisa Su.

4) The last verified performance "leak" was a VR benchmark showing the fully enabled "Big Navi" GPU would have around 2080Ti levels of performance, or about 40% faster than the 5700XT, which means 55-60 CU IF RNDA2 were to have the same per-CU performance as RNDA, which we know is not the case, it will be higher, lending evidence to the "leaked" 51CU fully enabled RDNA2 chip.

5) Power. RDNA2 will be 7nm+, which has EITHER a 15% power efficiency improvement OR 10% performance improvement compared to 7nm, per TSMC. The 5700XT already pulls 230-250w depending on which version of the card you get. Assuming they're able to somehow squeeze out a 20% efficiency improvement, they will only be able to add about 50% more CUs (60 total) to hit 300w, which again is in the expected range of 50-60CUs for a fully enabled RDNA2 GPU. Also it must be said that 300w would be a cataclysmic fault only to compete against the 2 year old RTX 2080Ti, which itself can pull 300w in the higher end boards, considering nVidia will be moving to a much more efficient manufacturing process (12nm-> 10nm OR 7nm OR 7nm+, depends on who makes them, Samsung or TSMC, or both), and with all expectations pointing to a card of 2080Ti power being placed at the RTX 2070 slot, it would be yet another hot, loud, power hungry top end AMD card against a quiet, efficient mid range nVidia card.

0 Likes

I read a lot of asian forums and sometimes there are some interesting tidbits on upcoming hardware.

some  speculation of a new nVidia V100-SXM2-24GB workstation card too: 7552 CUDA cores (118 CU), clock 1.11 GHz, 24 GB memory. Tesla cards for workstation are expensive.

0 Likes