john_This is disappointing. I wonder if they had to lower the IPC -compared to Fiji - to get higher frequencies. In any case I don't expect anything from drivers. Either RX Vega will be some kind of different in hardware, or it will have to be able to sustain a frequency close to 1700MHz to be highly competitive.
For now Vega looks like the GPU big companies where expecting, not like the GPU gamers where expecting.

RejZoRLaunch was crappy, but lets be honest, this is a production card, not a gaming one. It's not meant to be released with all the colorful and noisy PR that we expect with gamer cards. Anyone remembers a lot of fanfare around every Quadro card release? I sure as hell don't. When was the last time anyone ran Fallout 4, Dirt, Heaven and 3DMark on them (be it FirePro or Quadro)? I honestly can't remember a single time. It doesn't matter what they say, basing whole RX Vega state on FE card is like evaluating gaming performance of GeForce cards using Quadro. Yes, they can run games, but they usually do it like shit as well.

So, people, don't be stupid. With all this excitement of finally having a first actual Vega core, people forgot they just threw all the logic out the window by testing it as a gamer card. If gaming RX Vega will be rubbish, piss on it by all means, but doing that now just makes everyone look like absolute cretins. Yes, I watched PCPer's live stream, but I still believe the same to be true.

RejZoRThe chip is Fury X on steroids (imagine Fury X at 1,6 GHz) in its worst case scenarios and yet it sometimes performs even worse than old Fury X. How it doesn't click in anyone's heads that this just doesn't compute on any level in any way, makes no sense and is entirely illogical?

Old Fury X alone could be ultra competitive if they could run it at 1.6 GHz. So, seeing brand spanking new core with a lot of things further optimized and also learned from mistakes on Fury X from design perspective just doesn't make any sense for it to be still slower.

xkm1948I am going to mark this post and come back to it after July30th. Let's see how people are going to spin after the RX Vega drops. It will be glorious.

And for people calling FineWine, better driver maturation. Here is a little comparison of how much performance Fiji has improved over its current life.

First benchmark right after I got my card, no GPU overclocking. Using launch driver from AMD
www.3dmark.com/fs/5488050


2yrs of driver optimization, no GPU overclocking. 17.6.1 driver
www.3dmark.com/fs/12815443

Not the perfect analysis, but basically over the span of 2yrs I got additional 5% performance increase. That might give you some glimpse of how much FineWine you may get for Vega.

Power to you if you want to buy Vega.

FrickMeh, $/performance is the only true metric. Not everyone's an overclocker.

RejZoRThe chip is Fury X on steroids (imagine Fury X at 1,6 GHz) in its worst case scenarios and yet it sometimes performs even worse than old Fury X. How it doesn't click in anyone's heads that this just doesn't compute on any level in any way, makes no sense and is entirely illogical?

Old Fury X alone could be ultra competitive if they could run it at 1.6 GHz. So, seeing brand spanking new core with a lot of things further optimized and also learned from mistakes on Fury X from design perspective just doesn't make any sense for it to be still slower.

ratirtit's totally different than fiji and It's not for gamers but it gives some information. if in fallout 4 this Vega shows the performance of 1080TI maybe Vega gaming edition will be as fast as1080 TI or even better. Since it's not gaming card and pulls of that much FPS in one game that can be an indication of the speed of RX Vega but across all the games not just one. It would seem that Vega is capable of reaching 1080 Ti's performance.

xkm1948AMD should position RX Vega no more than $399 to be competitive. $349 would be sweet.

They surely love their new slogan "Disruptive" so I am waiting for some disruptive pricing. I agree, if the final gaming variant comes out with a good surprise in pricing it may still sell some.

FrickSo they have to do a Ryzen or burn? I don't think it's that bad.

xkm1948How much do you think AMD should price this? Anything over current 1080 price line-up will be bad.

john_If I am not mistaken, usually when someone is trying to create/change an architecture to work at higher frequencies, it ends up lowering the IPC. Pentium 4 was slower than Pentium III but could go at much higher frequencies. AMD FX was slower than Phenom II but could work at higher frequencies. Yes, these are CPUs not GPUs, but it could be the same case here.
It wasn't making sense to me either when AMD was introducing the 8150. My reaction was the same as yours. We have seen the failure of Pentium 4. How can they come 5 years latter and create a Pentium 4? Why not enhance Thuban cores, shrink them at 32nm and come out with a real 8+ core Thuban monster at 4GHz?


That Fallout 4 score is a big hope for RX Vega, but I am not holding my breath.

cdawallDepends how it works in rx form. If it competes with the 1080/1080ti it can be priced as such just like the fury

dwadeAMD has unofficially lost the GPU war. Volta is coming and that GTX 2060 will plow the fastest Vega.

xkm1948I find it amusing people still trying to defend a failed product launch and a bad flagship produxt.

dwadeAMD has unofficially lost the GPU war. Volta is coming and that GTX 2060 will plow the fastest Vega.

cdawallVolta isn't coming they are refreshing pascal

RejZoRThat's nonsense. You don't understand the principles or basics of chip designs. No one "lowers IPC" to gain clock. IPC and clock are entirely separate things. Clock only depends on the chip design. NVIDIA is currently using narrow but faster pipeline. AMD is using wide but slower pipeline. Because, if you want to achieve high clocks, you're required to have a lot of stages in the pipeline, making it long. And that isn't always a best thing (Pentium 4 was a huge failure because of this) because whith longer pipeline, you're gaining latency and discarding things half way through costs more, causing even higher performance penalties.

IPC on the other hand means Instructions Per Cycle (IPC). It doesn't matter whether core has 500MHz or 3GHz. One cycle is 1Hz essentially. So, IPC tells you how much work a chip can perform in 1 cycle. That's it.