AMD RX Vega Leaked Benchmark Shows It Ahead Of GTX 1080 – Specs Confirmed, 1630MHz Clock Speed, 8GB HBM2 & 484GB/s Of Bandwidth

[Taf the Ghost]

3 minutes ago, Master Disaster said:

But from sub 1070 to 1080+ performance just from drivers? Also if it is driver optimisation then its safe to assume the FE will get the same boost, right?

It should slot in right between the 1080 and 1080 Ti. What public information we've had has pointed to that for a while. Also, most of the time the Vega FE is sitting around 1430 Mhz, where RX Vega is going to run over 1600 Mhz. 12% more clocks speed, some actually activated new tech (which doesn't seem to be on Vega FE) and you can find 25-30% uplift. 

 

Rise of the Tomb Raider found 20% improvement on Ryzen by a patch. Using the compute power available is important. (People forget Hyperthreading/SMT is actually just an on-die optimization approach for scheduling, but you can see 40% uplift in certain scenarios.) 

[tom_w141]

23 minutes ago, Master Disaster said:

Unless someone can explain how halfing the RAM and boosting the core clock by 30Mhz can push performance so far ahead of Vega FE I'm sorry but I just don't see these numbers as being real.

Its actually fairly plausible. The RX has a beta Vega driver. Whereas the Frontier edition is launched so they needed it to ship with a driver known to be stable, which may have knee capped many features in the new NCU architecture but was still able to work because it is GCN based.

 

How else would you explain Fiji level performance on a card that's 400Mhz faster and on a newer architecture? It doesn't take 3 years to develop that.

 

IMO it has to be that drivers are holding back performance.

[cj09beira]

13 minutes ago, Master Disaster said:

But from sub 1070 to 1080+ performance just from drivers? Also if it is driver optimisation then its safe to assume the FE will get the same boost, right?

Yes vega Fe will probably see the same boosts but probably later so that any unfound bug has time to show up. 

11 minutes ago, Taf the Ghost said:

Current Vega FE drivers are Fiji derivatives...

Agreed.

With the new rasterizer working i would like to see what could be done with fp16 (how much perf gain) and if there would be any visual fedelity loss

[tom_w141]

14 minutes ago, Master Disaster said:

Also if it is driver optimisation then its safe to assume the FE will get the same boost, right?

Yes when the driver update is released (in theory). It could also be the jack of all trades driver switching nature that hurts the FE edition.

[Taf the Ghost]

2 minutes ago, tom_w141 said:

Yes when the driver update is released (in theory). It could also be the jack of all trades driver switching nature that hurts the FE edition.

I've been curious about that. The ability to switch drivers isn't some trivial activity; it has to be built into the deeper parts of the hardware driver. It's not just code paths and hooks for programs. It should come at an optimization cost in general. (I believe the Quadro cards require removing & installing separate drivers, they can't just freely switch?)

 

Though the funny bit is that the Gaming Mode is currently a worse gaming performer than the Pro Mode.  Not a joke.

[cj09beira]

14 minutes ago, Tedny said:

It won't be, it AMD GPU and have fastest memory on market (it won't be top 1, because of power consumption) 

It was tested for ethereum and its not much better than a fully moded 580 asits reaching 32 MH/s so you dont need to worry much as the other coins have a much lower market

[zMeul]

VideoCardz has something to say: https://videocardz.com/70777/amd-radeon-rx-vega-3dmark11-performance

Quote

Well, I didn’t really want to post this because I think it’s still too early, but since the highest score started to float around the web I think it’s worth to clear some misunderstanding. The highest score the 687F:C1 has achieved is an overclocked chip. 3DMark11 does not recognize unreleased overclocked graphics cards very well. The good news is that this puts RX Vega above overclocked GTX 1070, bad news, it might still be slower than overclocked GTX 1080. I guess time will verify those results. 

 

[tom_w141]

1 minute ago, cj09beira said:

It was tested for ethereum and its not much better than a fully moded 580 asits reaching 32 MH/s so you dont need to worry much as the other coins have a much lower market

It will still be purchased though. Many alt coins are ASIC resistant and require VRAM but some are also core clock dependant (which Ethereum isn't really at all) So these cards will definitely be good at some alt coins.

[tom_w141]

2 minutes ago, zMeul said:

VideoCardz has something to say: https://videocardz.com/70777/amd-radeon-rx-vega-3dmark11-performance

 

To be honest I don't think so but if that is true then R.I.P Vega.

 

EDIT: Though if true that shows a larger overclock headroom than the 1080.

[dorin]

Why Raja Koduri is still working at AMD? Just why?

The man is responsable for Fiji failure, Polaris seni-mediocracy and another failure with Vega most probably.

#FireRaja

[tom_w141]

11 minutes ago, dorin said:

Why Raja Koduri is still working at AMD? Just why?

The man is responsable for Fiji failure, Polaris seni-mediocracy and another failure with Vega most probably.

#FireRaja

That is pretty naïve. Raja is a very capable engineer and an intelligent man. They would be far worse off without him.

 

EDIT: Look at it like this if you want. Radeon Technology Group, RTG, are producing a 1080 competitor with a fraction of the budget NVidia has. That's pretty damn good even if it is late

Edited July 5, 2017 by tom_w141

[Taf the Ghost]

Someone already got their Vega FE to almost 1700 Mhz and about 11% OC on the Memory. So even the Vega FE has some headroom. 

[samcool55]

7 minutes ago, dorin said:

Why Raja Koduri is still working at AMD? Just why?

The man is responsable for Fiji failure, Polaris seni-mediocracy and another failure with Vega most probably.

#FireRaja

Well, there are a few things to be said about the Fury line-up.

It was the first with HBM and the Fury nano was the fastest GPU in its class.

 

And the AMD team got something you can't really put a value on, experience and knowledge.

Their team knows how HBM works and how to optimize for it. Vega is a really ambitious GPU for AMD but it can still end up being a good thing.

[PCGuy_5960]

I don't think 3dmark11 is a very good benchmark

 

FireStrike and TimeSpy are much better

[VerticalDiscussions]

Aw, it cant beat the 1080 ti .

 

 

[TheRandomness]

2 minutes ago, samcool55 said:

Well, there are a few things to be said about the Fury line-up.

It was the first with HBM and the Fury nano was the fastest GPU in its class.

 

And the AMD team got something you can't really put a value on, experience and knowledge.

Their team knows how HBM works and how to optimize for it. Vega is a really ambitious GPU for AMD but it can still end up being a good thing.

Considering the fact that they co-developed HBM, starting development in 2010, they really should've had it a little better. Though, it does save so much on power consumption, the two stacks on the FE should only consume anywhere between 12 and 24W under load according to its VRM, which buildzoid went over in a gamersnexus video.

[RedSpade]

1 hour ago, Droidbot said:

AMD finally got their driver-related shit together, it seems. 

 

Adored's video on the shit that AMD is going through is really quite good: 

 

That video is very good.

 

From what Adored said and going off these benchmarks, it would fit in well with what one of their execs (Jim Anderson) said, "compared to the nearest priced competitor you'll see anywhere from 30% to 70% performance advantage"

 

If so expect ~1080 performance for 1070 price.

[PCGuy_5960]

20 minutes ago, dorin said:

Why Raja Koduri is still working at AMD? Just why?

The man is responsable for Fiji failure, Polaris seni-mediocracy and another failure with Vega most probably.

#FireRaja

Without Raja AMD would have rebranded the 290X for the 4th time claiming that it's a "new GPU"

[DLex]

2 minutes ago, RedSpade said:

If so expect ~1080 performance for 1070 price.

MSRP or actual retailer price?

[Taf the Ghost]

2 minutes ago, RedSpade said:

That video is very good.

 

From what Adored said and going off these benchmarks, it would fit in well with what one of their execs (Jim Anderson) said, "compared to the nearest priced competitor you'll see anywhere from 30% to 70% performance advantage"

 

If so expect ~1080 performance for 1070 price.

If it was, say, 15% faster than the 1080 (with the 1080 Ti being about 35% faster than the 1080), but priced at 399 USD... that'd go pretty far. 

[cj09beira]

23 minutes ago, dorin said:

Why Raja Koduri is still working at AMD? Just why?

The man is responsable for Fiji failure, Polaris seni-mediocracy and another failure with Vega most probably.

#FireRaja

You...

Raja is what? 2 years into the company. When he got there fury was done and vega was deep into development

[cj09beira]

7 minutes ago, RedSpade said:

That video is very good.

 

From what Adored said and going off these benchmarks, it would fit in well with what one of their execs (Jim Anderson) said, "compared to the nearest priced competitor you'll see anywhere from 30% to 70% performance advantage"

 

If so expect ~1080 performance for 1070 price.

No way amd will be selling vega (the full chip) that cheap

[PCGuy_5960]

2 minutes ago, cj09beira said:

You...

Raja is what? 2 years into the company. When he got there fury was done and vega was deep into development

The first architecture designed (exclusively) by Raja Koduri is Navi

[Taf the Ghost]

The RX Vega will be $499 USD, I think.

 

Also, DEAR LORD WHAT HAPPENED TO PRICES?

[PCGuy_5960]

Just now, Taf the Ghost said:

The RX Vega will be $499 USD, I think.

It will have to cost less than $500, unless it beats the 1080 by more than 5-10%