Vega 56 & 64 Shader Comparison at Same Clocks

[SimianRob]

After reading the HardOCP comparison between Vega and Fury X at the same clocks, thought this one was interesting too. The conclusion seems to be that there is literally no difference in performance between Vega 56 and 64 when they are at the same clocks, regardless of the fact that the 56 is a cut down chip. It makes even more sense now why AMD chose to handicap the 56 with lower memory clocks. This is the first time where it almost seems like selling the chips based on shader counts makes less sense than just binning them and selling them based on clockspeed if the shader counts are having no effect on performance. The other thing I'm seeing posted a lot is the idea that Vega is so bandwidth starved that you would only see the 64 pull away at extremely high memory clocks. Thoughts?

 

[razor1]

After reading the HardOCP comparison between Vega and Fury X at the same clocks, thought this one was interesting too. The conclusion seems to be that there is literally no difference in performance between Vega 56 and 64 when they are at the same clocks, regardless of the fact that the 56 is a cut down chip. It makes even more sense now why AMD chose to handicap the 56 with lower memory clocks. This is the first time where it almost seems like selling the chips based on shader counts makes less sense than just binning them and selling them based on clockspeed if the shader counts are having no effect on performance. The other thing I'm seeing posted a lot is the idea that Vega is so bandwidth starved that you would only see the 64 pull away at extremely high memory clocks. Thoughts?

They matched up memory clocks too, ends up being memory bound, then both cards will end up at the same performance.

They should have done tests that matched up core clock vs matched up memory clocks, then we could see the difference.

Vega 56 isn't that much away from the 64 anyways 15% or less than 15% difference? Increasing the memory clock of the 56 could cover most of that gap then the added frequency to the core of the 56 can make up the rest.

And yeah Vega with high memory cocks does do better, but only up to the first 10%-15% of memory overclock, after that its more reliant on core clocks w/memory clocks.

 

[SimianRob]

Yeah, it just seems odd that the differences mainly seem to be clockspeed. If you clocked an R9 290 to a 290X, there would still be a 5% fps difference, nothing huge but it was there. There was a similar gap between the Fury and Fury X. But here we literally see zero difference. Really makes you question the reasoning for purchasing the 64 at all, unfortunately. I guess you get better performance out of the box but that's about it.

 

[N4CR]

Vega with high memory cocks does do better

It's been a long night, I'll show myself out now..

 

[razor1]

opps

 

[razor1]

Yeah, it just seems odd that the differences mainly seem to be clockspeed. If you clocked an R9 290 to a 290X, there would still be a 5% fps difference, nothing huge but it was there. There was a similar gap between the Fury and Fury X. But here we literally see zero difference. Really makes you question the reasoning for purchasing the 64 at all, unfortunately. I guess you get better performance out of the box but that's about it.

Well Vega 64 is yeah kinda a waste after seeing what Vega 56 can do.

 

[noko]

Well Vega 64 is yeah kinda a waste after seeing what Vega 56 can do.

Limited testing to games. Would like to see some hashing rates and other GPU compute stuff as well. Looks like Vega 56 is the best buy of the bunch if you can get it at the suggested pricing.

 

[ccityinstaller]

I am extremely happy with my 56 @ 1100Mhz HBM and a solid ~1725Mhz on the core now that I modified the power limit so the damn card isn't starved. Really makes me happy I didn't spring for that 64 I saw in the case @ MicroCenter!

 

[Gideon]

Like I said the 56 is the card to get for the money, the 64 is kind of meh but if you need the best AMD has then so be it.