Graphics AMD 6xxx Series (RDNA 2)

#RX6000 whispers...(2/2) New tech introductions -> NV 1. Lossless anti-aliasing 2. Nano-subsurface effect 3. A NCSN type tech (Noise Conditional Score Network) introducing training functions [unverified + salt]

https://mobile.twitter.com/Avery78/status/1307765955888476162

3 could be their answer to DLSS apparently.

 

If it is I doubt it'll work anywhere near as well as DLSS. Nvidia is head of a lot of the big research groups when is comes to CNN/DL, MD never even gets a mention, so I would be beyond surprised if they magically pull out something from nowhere with no other CNN/DL news...

 

https://www.overclock3d.net/news/so...ty_cache_-_a_radeon_gpu_memory_breakthrough/1

Infinity Cache confirmed.

 

Today on Wild Speculation, anything is possible and the sky is not the limit

 

Not even close.

There is no evidence that said trademark has anything to do with the soon to launch GPUs.

 

There's also unconfirmed reports of a layer of Infinity Cheese.

Everything's better with cheese.

 

Does it come with an infinity pool?

 

Rumors about huge cache in Big Navi circled around for some time. On itself it isn't that interesting, but it could possibly mean that AMD is preparing architecture for MCM. Maybe we won't see it in Big Navi yet, but if anyone is good at chiplet designs, AMD sure is.

 

https://www.chiphell.com/thread-2264233-1-1.html



The obligatory mountain of salt.

 

If what was shown in the AMD PR is a 6900XT, then 340W TDP from two 8pins, I would doubt this info very much from the link above which is someone's estimates...

I doubt AMD thought Nvidia was aiming over 300W TDP, so I will be surprised if AMD came out with a GPU over 300W TDP at the end of the month with Big Navi announcements.

 

Latest rumour mill thinking:

Navi 21 XT 16GB GDDR6 255W TGP ~2.4 GHz (Game Clock)




https://videocardz.com/newz/amd-nav...hz-game-clock-250w-tgp-and-16-gb-gddr6-memory

Obviously remains to be seen, but it certainly smells like they have something competetive with 3070/3080.

 

NVIDIA is so far ahead of AMD in the deep learning arena it's not even funny; they're actually putting out some pretty good research in their own right. That said, superscaling is pretty straightforwards tech these days, and in my mind the only question is how do you do it with a limited compute budget. I suspect DLSS itself isn't too hard to emulate (although I should probably look at exactly how they do it) but AMD may be hindered by their lower tensor op thoroughput.

 

That's a YouTubers speculation on that rumour @Spraduke.
I'm very skeptical on the clock speeds and TDP. As they have taken info from a PS5 chip that runs at 2.23GHz.

 

TDP seems higher than I expected, not long left to wait now though.

I would be surprised if not shocked if it’s above a 2080ti in performance.

 

According to the 3 game teaser from AMD "unspecified Big Navi model" is slightly slower than a 3080, so even if we subtract the usual marketing BS margin it should still stomp the 2080TI.

 

Seems AMD are doing their best to prevent scalping bots etc. for the launch of both RNDA 2 and Zen 3.

 

According to Wallossek, availability for AMD's Navi 21 graphics cards will start in mid-November, only a few weeks after AMD's RDNA 2 launch on October 28th

https://www.overclock3d.net/news/so...ies_power_consumption_and_memory_specs_leak/1

Sounds like another paper launch is coming then.

 

It has always been the plan to release mid-November. Somewhere between the 15th and 20th iirc.

I guess they could not say anything until then but this'll give them two weeks or so of press flooding.

 

Oh dear I'm now confused, wasn't there already a series of AMD 6XXX gpus a decade ago? Or were they still Ati back then?

Could lead to some brilliant Ebay fails though.