Sorry, when I said sweeping architectural improvements I meant the Vega performance over DOOM Vulkan earlier this year which was only 10% faster than the 1080. I doubt it's performance can actually improve to be competitive against a 1080ti or Titan Xp, as evident from that leaked Time Spy benchmark.
Timespy isn't any accurate way to determine what the final product will be like.
Plus benchmarks like timespy/firestrike tend to be very much hands off for nvidia/AMD, futuremark has to do the optimization, and especially with DX12 on a per architecture basis.
I really doubt they have optimized Timespy at all for VEGA, implying anyone at futuremark has even seen a VEGA card.
If VEGA 10s engineering sample clocked between 1000 to 1200mhz as evident by all the leaks is 10% faster than a GTX 1080, when it launches at 1500mhz it will be considerably faster than even 1080ti which is only 32% faster than 1080s.
Always remember than the earliest working samples of new silicon will always run at slower speeds clockspeeds than the finished product, especially something on a new architecture.
And even more importantly the drivers for something that new will be less than stellar.
As long as the price is right I shall be buying it to replace my 390x.
Some leaksies of possible low end vega card. The high Device ID is the indicator for that.
With Vega being a new architecture its performance will need drivers in good shape to fly. My R9 Fury Nitro is doing so well at 1440p gaming that I would have no trouble waiting a month or so in case of teething problems, or late third party cards.
I don't really care about what it will directly compete with, as long as it's a good upgrade from what I have. Preferably at a slightly lower power draw as well.
Comment from WhyCry of videocardz.com:
"All over the place" to put it mildly. It might very well be that the C1 is the early engineering sample and that C3 in that Sisoft sandra bench is a later sample.
It's in the comment thread to:
This is all pure speculation but..
If there is a "1070 equivalent performance" variant then I am thinking there is going to be price drops along the 500 range cards as otherwise its hard to fit into the price stack
1070 is roughly £350 territory, and I am seeing 8gb 580's for £220 anywhere up to £280.
Now given amd's usual tactic of "similar performance for less money" its going to put it too close to 580 shop price territory I think, for sake of argument "1070 equivalent performance vega" to be £300 to £320.
It occured to me it is a Xbone Scorpio development card for a console development PC. As console software and games are made on PC's and Scorpio is a modified RX480.
Like tech tubers I can make up news too 
"The Radeon RX Vega card itself will be coming in a rather slick black, red and silver box, complete with LED lighting on the box itself. Underneath the Vega logo is the simple tagline" - “It’s Got a Brain, It’s Got a Soul, It’s Almost Human.”
PC users: we need Vega fast and it should perform better than 1080 too with fancy new features and low tdp
Amd: LED lighting on the box
PC users: ...
ayymd
that was from the april fools day leak lmao that is fake
1 Like
Yes, we don't really know. The Vega numbers looks like a engineering sample leaking some results. Probably not at all representative for final product, there have been numbers around from at least two different samples too. AMD is keeping it close to the west, as they should IMHO.
Computex is less than a month away now. Real Soon Now...
Yesterday, May 1, was also AMD:s the day they reported on their Q1 financials. They mentioned Vega:
AMD’s “Vega” GPU architecture is on track to launch in Q2, and has been designed from scratch to address the most data- and visually-intensive next-generation workloads with key architecture advancements including: a differentiated memory subsystem, next-generation geometry pipeline, new compute engine, and a new pixel engine.
It's in the press release, so the rumors of it being delayed are false.
Here's another interesting bit from the press release
AMD unveiled that its "Vega"-architecture based GPUs have been selected to power LiquidSky's cloud gaming platform, enabling gamers to enjoy the power of "Vega" from virtually anywhere, and affordably through LiquidSky's low-cost and free subscription models.
Yes, it was talked about at the Capsaicin event. Also reported back in March that the Vega powered servers were online.
AMD’s CEO Dr. Lisa Su confirmed during the company’s Q1 2017 earnings call yesterday evening that Vega is on track for its Q2 2017 launch. The company’s highly anticipated next generation graphics architecture is set to debut in the gaming, enterprise and professional workstation markets before the end of June.
"AMD’s “Vega” GPU architecture is on track to launch in Q2, and has been designed from scratch to address the most data- and visually-intensive next-generation workloads with key architecture advancements including: a differentiated memory subsystem, next-generation geometry pipeline, new compute engine, and a new pixel engine."
I think we've all learned not to over hype the clock speeds of AMD's upcoming GPUs from the original Polaris launch though. We all saw the rumored screenshots before the release that it was gonna 1250mhz, lo and behold, it was 1250mhz at launch. If it's out in 2 months, that 1200mhz could likely be right on the ballpark of where it's gonna be at for launch.
It doesn't make sense for AMD to release only 1 Vega part, there would be too much of a gap between the 580 and Vega. That's like NV offering a 1060 and a 1080ti and nothing in between. They will have a lower binned Vega part that performs slightly better then FuryX to fill the void.
@Giulianno_D it doesn't make sense for them to release vega at 1200mhz when Fiji can already hit those speeds on 28nm..
Ignoring the fact that Polaris can hit 1500mhz on the improved 14nm process, a more optimized architecture on the same node should be able to do 1500mhz stock with ease.
I also doubt that AMD would lie about their specifications to the server/professional market about their MI25 machine learning vega part being able to hit 25 tflops of FP16, 4096 shaders need to be above 1500mhz to hit that.
@The_Riddick We're likely going to see something akin to the normal Fury with Vega's cut down versions, aiming for above 1080 performance just like Fury was above the 980.
1 Like
Not necessarily. A chip twice the size of Polaris will have a lot more trouble reaching high clockspeeds. Now AMD has talked about that Vega is more geared to clockspeed, still I don't think it will have an easy time reaching higher speeds. 150-200 MHz higher than Fiji wouldn't be bad actually. I honestly don't expect much more. I was very surprised that Nvidia managed to get such high clockspeeds on Pascal, it was an impressive feat. Sure, efficiency falls off a cliff when you overclock to 2GHz, still it is a freaking high clock for a complex GPU like the GP104. Hopefully AMD will be able to pull of something similar, but I don't bet on it as they have very little resources compared to NV.
Sooon.... sooooon...