New Upcoming ATI/AMD GPU's Thread: Leaks, Hopes & Aftermarket GPU's

Competition in the high end spurs companies on the bigger and better things. This is what we need.

 

Yeah it seems interesting though a bit odd too with the driver info and details here trying to think a bit on all of it.
(Which yeah that's not going to go well but still had to give that a try ha ha. )

Like there's a missing GPU model here with somewhere around 60 - 70 compute unit cores instead of AMD's 80 or bust where it's all down to 40 but then these are clocking really high to make up for it, wonder how a 72 or 64 unit or some such would perform though if scaling has been improved it could do really well.

Hmm wonder if AMD will fill out the 6700 and 6800's with either 50's or XT's later on I assume the 6900's already something like a 6900 XT and the B model is the same deal like Vega64 and Navi10 saw with the XTX type model.
Course that also leaves room for a non-XT 6900 that could be above the 40 CU's or whatever AMD's going to do.

6950 XT HBM2 too?
(Unlikely, maybe the CDNA's will be HBM though but that's for later.)


EDIT: GDDR6 is at speeds from to 14 up to 18 GB/s too I think now but then the X model I saw listed at 19 - 21 so that has further room too should NVIDIA go with faster for newer cards.

Not too sure what the higher speeds have for density or RAM capacity per module though, 2 - 4 for GDDR6 I think I saw somewhere and currently 1 for the X model but I would expect that to improve if GDDR6X is going to see further development and improve over time.

All those 40 GPU cores are going to be nice for ensuring availability though I would think, maybe not all of them will boost to acceptable as 6800's though but that leaves the 6700's with stock too.
(Bios CU unlocking as a possibility would be really neat for these but yeah that's probably not happening.)

 

IMHO - there won't be a 6950 XT with HBM2.... all "consumer" variants will be with GDDR6. On the other hand -- the CDNA2 branch is something different... to what "pure" RDNA2 for consumers offers.
But /// it is not necessarily bad that we won't have HBM2 variants.... remember the special sauce and Infinity cache?
But you haven't heard that from me

 

I'm dying to get my mitts on a 6800XT or 6900XT (or whatever they'd be called then), IF there's a 6900XTX, I may not go for such a powerful GPU as it'd be for my main monitor which does 3840x1080 144Hz. I almost went for the RTX3080, but due to lack of supply and the ridiculous price here in my neck of the woods, I missed out.....but I'm glad I'd missed out after hearing of the CTD issue.

I may not go for a Zen 3 CPU though, my 3900X is more than enough for what I need it to do....

 

I would not be at all too surprised, if AMD pulls a rabbit out of the hat at the last minute by announcing two very fast GPUs at the top of the Big Navi2 line-up.
Nobody expected the most extreme of the Ryzen 5000 series CPUs either and it was held right to the end of the introduction.
Same thing could be coming in the GPUs.
Maybe the XT6900 with GDDR6 and a further 6900XTX with HBM to administer a 'knock-out' punch to RTX3090 at maybe $1199.
It is very exciting to be in the market for a graphics card.
Lots of choices.

 

I'm thinking they might actually do it, 6900 series with HBM as a flagship halo product, very limited run but there just to have that performance crown

 

You guys realise that by doing that (releasing the consumer product with HBM2) they would basically double the development costs for the consumer drivers right?
I mean e.g. double the amount of tests, different paths in the driver depending on the memory type, different optimisations etc.
As much as I like the idea with HBM2 I will first believe it when I see it... unfortunately.

 

Update from Patrick Schur >
"
- Max. TGP of Navi 21 XT is 290 W.
- AIB partners are testing different TGP for their cards.
- The ~2.4 GHz was achieved with an AIB partner card.
So assume that the card was slightly overclocked.
"
https://twitter.com/patrickschur_/status/1317855621157392393?ref_src=twsrc^tfw

Update from rogame>
"
[Update Navi21]

Navi21 XL
> Base clock 1350MHz to 1400MHz
> Game clock 1800MHz to 1900MHz
> Boost clock 2100MHz (maybe 2200MHz)

Navi21 XT
> Base clock 1450MHz to 1500MHz
> Game clock 2000MHz to 2100MHz
> Boost clock 2200MHz to 2400MHz
"
https://twitter.com/_rogame/status/1317731737347248129?ref_src=twsrc^tfw

That leaves the XTX as an unknown and also XLE.
Navi 21 XTX (0x731F0)
Navi 21 XT (0x731F1)
Navi 21 XL (0x731F3)
Navi 21 XLE (0x731FF)

I assume that the XTX (the special edition) will have game clocks cranked up up to 2.3-2.4 GHz and boost maybe 2.5.
Remember ---- 2.4 was achieved by an AIB.... and AMD is keeping the highest SKU verrrrry close to the chest this time.... I even speculate that they are producing in on their own and AIBs are slapping only a sticker (similar as with my VEGA64 Liquid Cooled Edition).
It would be interesting if AMD teased the performance of the smallest big navi having 3 additional aces in their sleeves

 

Sounds like AMD might have a beast of a card.

 

Another day, another update, this time in regard of RAM per SKU ---- from _rogame this time (one of the verified sources imho) >
https://twitter.com/_rogame/status/1318117797562646528?ref_src=twsrc^tfw

"
Add Navi22 10GB VRAM to the list
>Navi21 16GB VRAM
>Navi22 10GB & 12GB VRAM
"

 

12gb-2.5ghz game clock navi22 for me. If priced at something like 400$ it will turn the market upside dawn.

 

Just as a hint -- remember what they said in regard of RDNA2 :

AMD's Radeon Product Management Lead, Mithun Chandrasekhar:
“Similar to Ryzen,” he says, “all of us need a thriving Radeon GPU ecosystem. So, are we going after 4K, and going to similarly disrupt 4K? Absolutely, you can count on that. But that’s all I can say right now.”

 

Been watching this thread for a while. Looks very promising.

I have a 2070S but could see an upgrade next year depending on how things shake out.

My last Radeon was the 7900 series.

Haven't had an all AMD build since Phenom days. Nice to see the great competition for pc hardware once again!

 

The fact 3070's are 8gb and was thinking of a 4k screen now has made this gpu def more interesting! I mean people are saying you prob need a 3080 for 4k anyways but dont like the fact nvidia are almost forcing me to get a 3080 because of the extra 2gb vram.

 

Well.... 12 GB should be just enough for 2-3 years of 4k gaming....
But even with memory compression the 10GB of an "old" 3080 (cause we already know "new" 20GB is coming for like 150-200 $ more) might be somewhat tight.... I mean I already saw one or two game examples, where in 4k + ultra textures we have topped at slightly under 10GB.... Of course you can't tell if the game engine is only reserving the memory or actively using it... with one exception I think.... where NV compared 3080 to 2080 (non Ti with 8 GB of RAM) in one of the Doom games (eternal ? not sure). Where you saw the framedrop due to exhaustion of RAM on the 2080.

Plus don't forget that AMD also has memory compression... Not that good as NV but still within 10-20% margin.

It will be a veeeery interesting fight - that I can tell!

 

The thing and am sure its been suggested already in discussions why cut the 3070's mem and also the cores does seem like an underhand trick to put people off getting the 3070 for 4k use or maybe my tin foil hat has fallen off.

Maybe was wishfull thinking the 3070 was gonna cut it anyways.

 

Yeah... my spider sense was also tingling that is why I have eventually cancelled the preorder for the 10GB 3080 EAGLE OC that I have been able to "buy" for the MSRP price (699 EUR + shipping). I fear that I would face the RAM bottleneck sooner rather than later on 3080 + my C49RG90.....
And on top of that - the paper launch.... must be fate that I will buy something from AMD again, if the Perf is ok (3080-ish +5% level would make me happy as an "acceptable" one) .

 

Actually, AMD's memory compression is better now. Shaders can process textures in compressed state for some time. And that applies for small chunks of compressed data too.
Each has different advantages. And if maximum compression is not needed due to sufficient amount of VRAM (which is true for any 8GB VRAM card), bandwidth matters.
And that means Infinity Cache (as long as real) will tackle bandwidth side of things a lot and again puts limits of memory subsystem to rest.

In general, I say that cards are more often than not limited by memory bandwidth than memory capacity. And it looks like RDNA2 focused on both. And that it improves on many different things. It may happen that jump from GCN to RDNA1 will prove to be smaller than jump from RDNA1 to RDNA2.

 

Interesting.
https://www.reddit.com/r/Amd/comments/jechk0/xt_is_full_die_for_navi_21_so_what_is_xtx/

Even more so for some of the responses on the 6900 XT and 6900 XTX details.

 

Im still holding on to my own Leaks:

BigNavi XTX will have HBM2 (2x8GB) -> 749€ (maby less cuz' of 3080 performance)*
6900/6800/6700 will rock GDDR6 18GB/s -> those SKUs will shake the market with very agro pricing.

6900XT 499-549€ (min. 15% faster than 2080Ti)
6800X 399€
6700XL 249-299€

* One guy from ATI (Canada) hint me that LC version is possible (4 months ago )
So we will have Halo product like FuryX & Vega XTX LC