EVGA Releases Optional BIOS Update for RTX 3080 FTW3 ULTRA - XOC 450 W

[Zach_01]

Exactly. The thing is that the raw performance uplift comes with a TDP uplift of ~equal measure, so I can't really see any perf/W increase at all.

When they announced the prices and everybody started panic selling their 2080 Ti's, I had a feeling that there was a reason behind it. Nvidia has never been famous for bringing good value propositions in the higher market segments - don't get me wrong, people, they're good cards, just not as good as Jensen claimed in the announcement. It also puzzles me how they managed to achieve a 30% performance uplift with ~double the cuda cores.

Not every cuda core is created equal. Its like trying to compare different cores from different CPUs.

-------------------------------------------------

Its easy if you know the specifics. According to Jensen a process unit to be called cuda core must be able to execute FP32 instructions.

Pre-Turing architectures every cuda core was capable of 1 Int or 1 FP instruction, and cannot execute both.

This has changed in Turing. In Turing every cuda core was performing exclusively FP instructions and the number was 4352. For Int instructions Turing had another 4352 units that was not counted for cuda cores (see Jensen)

Ampere 3080 now has (4352x2) 8704 cuda cores and the 4352 of them are executing exclusively FP instruction and the other 4352 are executing Int or FP instructions, but not both.

See?

 

[AusWolf]

Not every cuda core is created equal. Its like trying to compare different cores from different CPUs.

-------------------------------------------------

Its easy if you know the specifics. According to Jensen a process unit to be called cuda core must be able to execute FP32 instructions.

Pre-Turing architectures every cuda core was capable of 1 Int or 1 FP instruction, and cannot execute both.

This has changed in Turing. In Turing every cuda core was performing exclusively FP instructions and the number was 4352. For Int instructions Turing had another 4352 units that was not counted for cuda cores (see Jensen)

Ampere 3080 now has (4352x2) 8704 cuda cores and the 4352 of them are executing exclusively FP instruction and the other 4352 are executing Int or FP instructions, but not both.

See?

So basically Int cores count as a cuda core now, while in Turing, they were just an extra unnamed component next to the FP cores? Interesting.

 

[Zach_01]

So basically Int cores count as a cuda core now, while in Turing, they were just an extra unnamed component next to the FP cores? Interesting.

Process units are called cuda cores, always was, as long as they can execute FP instructions, despite the Integer capability.

In Ampere all units can process FP but half of them can process Int. Turing had cuda cores only capable of FP, and other units (same amount) only capable of Int that are called... Int processing units, but not cuda cores.
Its the FP capability (do or dont) that defines a unit to be (called) cuda core or not.
Its Jensen on the work!

 

[AusWolf]

Process units are called cuda cores, always was, as long as they can execute FP instructions, despite the Integer capability.

In Ampere all units can process FP but half of them can process Int. Turing had cuda cores only capable of FP, and other units (same amount) only capable of Int that are called... Int processing units, but not cuda cores.
Its the FP capability (do or dont) that defines a unit to be (called) cuda core or not.
Its Jensen on the work!

Ah, I see. The cynic in me starts to think that the term 'cuda core' is just another marketing BS (8000 is better than 4000, right?) However, it doesn't change the fact that Ampere offers minimal (if any) efficiency gains over Turing, which is sad.

 

[Zach_01]

Ah, I see. The cynic in me starts to think that the term 'cuda core' is just another marketing BS (8000 is better than 4000, right?) However, it doesn't change the fact that Ampere offers minimal (if any) efficiency gains over Turing, which is sad.

Precisely! Numbers of one aspect alone mean absolutely and utterly nothing. Core counts, speeds, IPCs, cuda cores, stream processors, even bus width and almost everything. Its the combination of all that counts and still you cant tell the actual performance even if you know everything.

Only real life apps/games.
Marketing is marketing and their job is to throw out some numbers to impress. And its working, because the majority of people (maybe 90%) do not spend their time to be informed about the details or the actual performance.