So now that we have benchmarks in. What did they change to make it slower per clock?
Something is clearly a strong bottleneck in the design at this point what is it?
Wait for Rx vega
Somehow the card is slower than a Fury X OCed to 1400mhz with 1000mhz ram clock
How is this possible? Same core count yet the Vega FE is 1600mhz on the clock.
11,511 is Fury X score on extreme
10,585 is what the Vega FE gets on xtreme
Nearly 10% lower performance vs the 2+ year old card
>>61132316
Yea no you're talking out of your ass.
Even if Vega FE had 1080 tier of performance Fury X is nowhere near 1080 or Vega. It's not even on the same level as 1070.
Vega FE timespy results are 6785 which is much lower than some of the OCed 1080s that run nearly 9,000 on the benchmark
>>61132387
This is a LN2 OCed Fury X look at the clocks vs default...
Point is to show that even if they changed nothing besides the fab size and got it to clock to 1600mhz it would peform better than what the Vega is performing now. Someone they reduce performance per clock by a large amount
There is no reason at all that crossfire 570s should perform close to the Vega FE
Its less core count over and much lower clocks on a narrower bus with normal GDDR ram and with multi card scaling not being 100%
Something is not working right on Vega this is very clear
4096 cores @ 1600mhz should outperform 4096 cores @ 1128mhz(two 570s each having 2048 cores)
>>61132482
>Something is not working right on Vega this is very clear
It's the drivers, stupid.
>>61132279
No point. Just waitâ„¢ for Navi. No more mining craze, no more shit arch, no more shit GloFo process. Polaris and Vega are the most shameful GPU archs AMD ever had.
I don't understand what's so bad with Vega being on 1080 level to begin with. Those speculations were right since that DOOM bench back in December. I don't understand where this whole "1080 TI KILLER" meme came from.
>>61132681
Price difference, is a big thing
but considering this is not even cheaper RX version, Who knows.
>>61132681
Its a year late and almost certainly wont be cheap
>>61132758
Can't possibly be more expensive than a 1080.
>>61132279
I don't quite think the software is there for one. But even if it was and theoretically, being generous, making it perform 10-15% better, the main thing I have against Vega is how underwhelming the HBM2 implementation is on the Frontier Edition. They cut the bus width in half and gave it less memory transfer bandwidth compared with the Fury. Why? This is obviously gimping this card a lot.
Are they running into thermal limits or something? Something went horribly wrong during the design of this.