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

Sapphire make some great AMD cards. I watched a teardown of the 5700XT Nitro+, the card is a work of art honestly.

 

Yes they do. I came so close to getting the Nitro+, but I just want a bit more. I'm glad I've held off so far. The performance of these next cards are looking Stella.

 

One of the weaker ones he made. He could not get over "AMD, does not know how to make heatsinks." So he repeated it like 25 times across video even after he realized that he said it way too many times in this particular attempt on making given video. As he says that he had to redo it already because before his rants were even worse.

And it is one where I really disliked his "power analysis" as he actually did not do it, just speculated when answers were right in front of him.
But one thing that saves him is that he is used to think about GPUs and not about entire systems. And that there are known changes in power behavior which are not yet present in AMD's dGPUs, so he has no 1st hand experience.
And it should be clear by now, that he learns best by 1st hand experience.

 

Better uder-hype than over-hype
Also he is (like me) thinking that ATI will use HBM2 in Big one.

 

Im on the same boat. Just waiting. THe Vega 64 is good for all I put to it, Even star citizen I was on it yesterday max setings at 1440p 60 fps.

I really hope Navi 21 is a HBM2E monster. 16G + hbcc we dont need more

 

Thing is that Card does not need 4x4GB HDM2e to reach memory bandwidth in between 3080 and 3090.
12x 1GB 18Gbps GDDR6 will do the trick too. It will not enable card to effectively compete with 3090 across the board even if GPU is quite powerful, but it will not limit GPUs capability to exceed 3080 performance.
And those extra 2GB VRAM is just bonus.

GPU that would be able to use all that bandwidth from 4x HBM2e would likely be quite expensive as that's double of what 3090 has available.
Would they use 3 stacks... Still plenty of VRAM and still excessive bandwidth even for 3090. So question remains about which is cheaper:
IMC for 12x GDDR6 + 12x 1GB GDDR6 @18Gbps + bit more complex PCB + extra VRMs for GDDR6 + bigger and stronger cooling
or
IMC for 3x HBM2e + 3x HBM2e + Interposer

And if AMD has GPU that can make use of 3x HBM2e bandwidth. Or if they can get good deal on slower HBM2e chips.

 

2x8GB 460GB/s will do the trick also (Less space & easier to implement on IF).
When OC it will cross 1TB/s magic line....

 

Yes this is what im expecting.
August 12, 2019, SK Hynix announced their HBM2E, featuring eight dies per stack, a transfer rate of 3.6 GT/s, providing a total of 16 GB and 460 GB/s per stack.[20][21] On 2 July 2020, SK Hynix announced that mass production has begun.[22]

Loads of combinations per stack possible so we can even get 2 stacks and 32G

Expensive

 

Everything has been pointing towards big navi being between a 3080 and 3070 and if thats the case im going to be very disappointed. I dont think amd is willing to go all out and use hbm2e for big navi. If they do ill be pleasantly surprised.

 

I'm not sure exactly where performance will end up as AMD has been very silent. But i have been reading rumors that only the prosumer version will get HBM memory, now that AMD has split the 2 divisions.

 

I wonder will this prosumer card rival a 3090 in gaming performance for 1000 bucks? I might just buy that for my gaming rig instead if thats the case.

I want AMD to target all of NVIDIAs skus and destory them. I want the glory days of radeon to return again.

 

Everybody likes competition
ATI was very good back in the days, now nV just leading the pack,
(But at last we have good Performance for Hi-end -> 3080 is a good GPU + some new features w/RT).

Note:
This time around You can't go wrong, either ATI or nV will suffice for at last 3-5 Years of gaming.
BigNavi or RTX3080 is a way to go this time (budget up to 700€).

 

Xbox Series S and Xbox Series X Price Revealed
Microsoft already confirmed the arrival of the Series X this November, but an exact release date wasn’t revealed yet. The same can be said about the next-gen console's pricing. According to reliable sources Windowscentral and Brad Sams from Thurott, however, we now know both the release date and the price of the console. On top of that, the design of the yet-to-be-announced Xbox Series S has leaked and we’ve got the price tag on this long-rumored console as well.
pic.twitter.com/VMkQY9271K

— WalkingCat (@_h0x0d_) September 8, 2020

Both consoles will be arriving on November 10, with Microsoft’s top entry, the Xbox Series X, selling for $499 at retail. The Xbox Series S, previously known as Project Lockhart, will be sold at a very attractive price point - $299.
According to Windowscentral, both consoles will be available through Microsoft’s Xbox All Access financing option, with the Xbox Series X being available for $35 per month and the Series S for $25 per month.

We know also have a glimpse at the Xbox Series S, which is expected to be officially announced shortly. Check out Brad Sam’s new video showing off the console down below:




From the looks of it, the Series S doesn’t come with a disc-drive.
Due to the angle of the shot, however, we can’t confirm this yet. The design of the S appears to be more in line with that of the current Xbox consoles, more specifically the Xbox One S.
The design of the black dimpled circular shape seems to have been borrowed from the design of the Series X, and might very well be used for ventilation.

 

Considering $500 price point, they should have exchanged those four 1GB chips for 2GB chips too and ask $550.

 

https://twitter.com/mooreslawisdead/status/1303136104019898368

 

Some food for thought on the speculation games;

The 5700XT, as well as the Vega 7, utilize TSMC’s 7nm 7N HP (High Performance node. 7N HP has a 66.7 million transistors per square millimeter (MTr/mm²) of transistor density. The TU102 Turing used TSMC's 12nm 12FFC process that has a density of 33.8 MTr/mm². To make up for the lack in process density, Nvidia shipped a 754² mm die size to meet their performance targets. Note: TSMCs 12FFC process is a power optimized variant of their 16nm process while AMD's 7N is a new node is a first run node process and took some time to get optimized. Ampere is using a power optimized Samsung 8nm UHD process with a density of 61.18 Mtr/mm², a die shrink of Samsung’s 10nm process. (Back of the napkin calculations estimate Ampere’s die size of > ~500mm²) RDNA2 is likely going to be utilizing TSMC 7P process, which is a power optimized 7N with no density optimization maintaining 66.7 MTr/mm². (Unless AMD changes to the density library then the density goes to 95.08 MTr/mm²) Note, both the PS5 and Xbox series X are using TSMC’s next generation N7+ EUV process with a density of 115.8 Mtr/mm².

For reference: Intel 14nn++ is 43.5 MTr/mm² Intel 10nm is estimated at 100.8 MTr/mm²

 

Tomorrow RDNA2 news?

https://twitter.com/AzorFrank/status/1303492877989879810?s=20

 

AMD Could Potentially Unveil Big Navi ‘RDNA 2’ GPU Powered Radeon RX 6000 Series today!

HBM2 -> or GDDR6?
Im 90% sure it will pack HBM.... 2 stacks x8GB

Rumour -> it can be XTX LC version also

Is the Sun -> an Arcturus

 

CDNA=HBM2, RDNA2=GDDR6. At least, that's my thoughts. More than happy to be proven wrong.