The guy who wrote this article must be a PC gamer LOL, trying to explain his understanding about data center GPU ROFL

I'm so glad AMD removed those Indian crap out of Radeon.

What a crap...

QuoteAs a Vega derivative, Arcturus' IPC is a known quantity

Is it? Do we know yet how Vega "derived" it is - i.e. is it just Vega but bigger, or is has it been enhanced? I was under the impression that Vega couldn't scale past 64 CUs, so they must have changed *something*.

I'm not sure DOA is the right way to describe what sounds like it might yet be a competitive card in its market, at the right price. Not terribly exciting, for sure, but not inherently worthless.

I believe these guys don't know anything about GPU'S .

Instinct/VEGA7(2018), Zen2, Navi10, Navi14, Renoir, Navi21, Navi23, XsX, PS5, Dali, InstinctMi100(2020), Zen3(2021?).  Twelve AMD official products on TSMC 7nm(+).  Nvidia=Zero official 7nm designs as of now.

Clearly AMD has a massive head start on 7nm.  A total of 8 GPU designs between Instinct 2018 and Instinct 2020.  It would be foolish to assume that the leaked MI100(2020) would have the same technology just because they share the VEGA label.

sorry guy, you have no f**** idea of what you are talking about, please go to school and study maths, this card is faster than tesla v100s. You are comparing a lorry with a car.  Please go back to school and re-study maths. 

All I can say is WOW..How uninformed can you be? You clearly dont understand the progress being made..7nm...RDNA?..key members of the ZEN team are now working on Radeon..No Raja Koduri to screw things up.. Not ringing any bells?...This card will smash balls!! You are going to look stupid for writing this article when the actual benchmarks come out.You are nothing more than an agent of misinformation at this point..or worse..an Nvidia fan boy.

If this so true it makes little sense. Radeon VII has be available for a year. Based on your math it is just a dual R7. Why would they release it so late?  I think there is more to this.

Uhm, those does not sound too bad? Even on that base clock 1.09 GHz that would be 1.09*8192 = 8.92TFlops of fp64 compute power(most meaningful number for HPC)which is higher than Tesla V100S.

Comparing it to desktop GPUs is just pointless... maybe some Turing Tesla's but even then those does not offer any meaningful fp64 compute power compared to this so the use cases are not the same even on data centers.

I agree on Ampere though, 70% over any Tesla V100 is well over 10TFlops of fp64 and even with that boost clock of 1.333GHz does not really beat that(10.9TFlops FP64 vs most conservative 70% over V100 estimation 11.9TFlops FP64).

Vega improvements in 4000 igpus is definitely not just down to clock speeds they actually have a lower CU count (11 max to 8) :-*There are many articles explaining how they implemented changes that brought about around 50% increase in IPC and how they plan on implementing these improvements in RDNA 2.0.

The performance gains seen with Renoir iGPUs are down to higher sustained clockspeeds and a higher CU count. There are next to no architectural improvements.

"VEGA IPC already known" ???

Not necessarily known.  AMD has demonstrated the possibily of higher IPC for VEGA on the recently announced Renoir APU.  Wouldn't count Arcturus out just yet.  Lisa Su herself commented on higher IPC for the Vega IGP on Renoir.  Better performance/watt should be considered for Arcturus as well.