AMD Vega (FE and RX) Benchmarks [Updated Aug 10 - RX Vega 64 Unboxing]

[Qwertilot]

Well if the gains are that small from 210-295, then it really makes you wonder why on earth they're shipping one version consuming ~350w.....

 

[IntelUser2000]

Well if the gains are that small from 210-295, then it really makes you wonder why on earth they're shipping one version consuming ~350w.....

Remember how Fury X had minimal gain over the regular Fury?

Also there was someone saying Vega 56 is faster than GTX 1070 and rather close to GTX 1070. That may mean Vega 64 and higher versions don't offer much gains - just like with Fury.

You can't blame any one factor for it. They must have been scrambling. Like when you are panicking and everything is a blur and nothing works out?

They must have seriously miscalculated at some point. Project screw-ups plus probably assuming GTX 1080 would have been Nvidia's top chip. If it was, they'd be better off. Then they get a Ti out. Meaning Nvidia has room to cut 1080 prices if they want to. And even if AMD matches 1080 they won't be shoulder to shoulder with a halo product.

 

[Malogeek]

Well if the gains are that small from 210-295, then it really makes you wonder why on earth they're shipping one version consuming ~350w.....

1 test in 1 game. It requires a lot of thorough testing to determine overall power usage and performance characteristics. It's certainly intriguing though.

 

[AtenRa]

You really need a lot of talent to do this, 73% more transistors than GP104 (GTX1080) to get the same performance in gaming after more than a year and almost double the TDP.

Really its mind boggling how they managed to do this while having both 14nm FF + HBM2 at their disposal.

 

[EightySix Four]

You really need a lot of talent to do this, 73% more transistors than GP104 (GTX1080) to get the same performance in gaming after more than a year and almost double the TDP.

Really its mind boggling how they managed to do this while having both 14nm FF + HBM2 at their disposal.

It seems like AMD expected that their architecture would be the primary target for games after winning the console business, which is a pretty good strategic bet, but they ended up flat out wrong. Mantle/Vulkan and DX12 have not made nearly the dent they had hoped in the amount of time they hoped, leaving them with their pants down and nvidia doing a much better job of optimizing their architecture and software for games right now.

It's actually amazing how poorly their strategy played out and big props to nvidia for countering it.

 

[Malogeek]

It's actually amazing how poorly their strategy played out and big props to nvidia for countering it.

Yay for stagnating game development APIs! Aren't they fantastic?

 

[EightySix Four]

Yay for stagnating game development APIs! Aren't they fantastic?

While it is definitely important to move forward with new game dev APIs and such, the end goal is always for the gamers to get better looking video games, and I'd argue that games are continually improving right now even using the old APIs due to the additional performance afforded them with newer hardware.

We also need to remember that neither nvidia or AMD care about the games industry the way we do - they just want to maximize the size of the pie and their slice, which sometimes aligns with gamers and sometimes doesn't.

I should note with all of this that I actually lean towards the red side but can't help but give props to the way nvidia has played the last few years.

 

[Malogeek]

The basis of newer APIs is to do more with less. I personally have an issue with the concept of simply throwing more transistors and power at the problem rather than looking at the bottlenecks and moving to a more efficient approach.

Alas getting off topic here and probably rousing the rabble.

 

[Mercennarius]

I want to see more 1070 vs Vega 56 benchmarks....if Vega 56 ends up a better performer than a 1070 then that's the real story here. 1070+ performance is enough GPU power for 99% of gamers.

 

[Tweak155]

I want to see more 1070 vs Vega 56 benchmarks....if Vega 56 ends up a better performer than a 1070 then that's the real story here. 1070+ performance is enough GPU power for 99% of gamers.

This is what I'm interested in... I've been holding off upgrading my release GTX 970... I also want to replace my monitor at some point and want *sync. So it will likely be either Vega or Volta probably. If I don't like Vega results enough I may as well buy a GSync monitor now!!

 

[Cookie Monster]

Maybe it might make nVIDIA push the price down the 1070s and make room for a 1070Ti (spec'd like the mobile version + clocks). Vega 56 is looking like a solid card for the price, power and performance.

 

[Glo.]

Why not you all wait for reviews of the GPUs to judge the architecture, and draw conclusions?

 

[Mopetar]

You really need a lot of talent to do this, 73% more transistors than GP104 (GTX1080) to get the same performance in gaming after more than a year and almost double the TDP.

Really its mind boggling how they managed to do this while having both 14nm FF + HBM2 at their disposal.

It's not that mind-boggling if you think about it. First you'd want to compare it to GP102 since Vega itself is more similar to the Titan in that is has compute and prosumer features that aren't necessary for gaming.

The second is that AMD themselves had indicated that a lot of the transistor budget was to make GCN clock higher. It's never been as good at high clock speeds, and may have some low-level design decisions that make doing this difficult. It seems like in lieu of AMD trying to build a new architecture (or perhaps in lieu of it not being ready yet) they did what they could in order to get clock speeds up, at the expense of using a lot of transistors that probably could have been better spent elsewhere. AMD may have well be better off focusing less on clock speed gains and instead putting the additional transistors to more shaders or other resources.

Hopefully for AMD's sake, Navi is a new architecture rather than just a shrink. GCN has had a good run, but it needs to be retired. It's pushing 6 years old at this point, and that doesn't count time spent developing it. They could get more life out of it if they take a similar approach as with Ryzen where they build multiple smaller units and connect them together, as this would make it more economical for AMD to build chips with more CUs running at lower clock speeds without having the usual problems of poor yields with larger dies.

I want to see more 1070 vs Vega 56 benchmarks....if Vega 56 ends up a better performer than a 1070 then that's the real story here. 1070+ performance is enough GPU power for 99% of gamers.

If Vega 56 were a big winner here, I think AMD would be doing more to market this. They were pretty mum in terms of information so I don't expect it to be that much better than a 1070.

What I'm most interested in right now is the Nano, which we don't know a lot about other than it's size (similar to the Fury Nano) and that it has an even lower TDP than Vega 56. I'd like to know if it's just a specially binned Vega 56, or if it's an even more cut version of the chip (48 CU) that might come in closer to $300, especially if it's only got 4 GB of HBM2.

 

[crisium]

Maybe it might make nVIDIA push the price down the 1070s and make room for a 1070Ti (spec'd like the mobile version + clocks). Vega 56 is looking like a solid card for the price, power and performance.

nV haven't had any control over 1070 prices for a few months now. But I agree that 1070 vs 56 looks to be an intriguing battle. And good point, there is room for another card between 1070 and 1080. I don't think Nvidia is interested in more tiers right now though. The trend has been fewer tiers.

 

[leoneazzurro]

There are some rumors hinting that the bad mining performance per watt of RX Vega could be in reality not quite true (unfortunately for who wants it for gaming)

https://www.overclock3d.net/news/gp...vels_of_mining_performance_on_amd_s_rx_vega/1

 

[Krteq]

Oh God, please no!

... but it can be expected due to implementation of some hash / cryptocurrency mining instructions to NCU :/

 

[crashtech]

Here's to hoping the bottom falls out of the cryptocurrency market soon.

 

[stockolicious]

Remember how Fury X had minimal gain over the regular Fury?

Also there was someone saying Vega 56 is faster than GTX 1070 and rather close to GTX 1070. That may mean Vega 64 and higher versions don't offer much gains - just like with Fury.

You can't blame any one factor for it. They must have been scrambling. Like when you are panicking and everything is a blur and nothing works out?

They must have seriously miscalculated at some point. Project screw-ups plus probably assuming GTX 1080 would have been Nvidia's top chip. If it was, they'd be better off. Then they get a Ti out. Meaning Nvidia has room to cut 1080 prices if they want to. And even if AMD matches 1080 they won't be shoulder to shoulder with a halo product.

I think AMD did what they did with VEGA on purpose. They built the card for Gaming "And" Pro space probably because they had to save money. So what we will get is a card that games well and does some great things in the pro space. It will not unseat NVDA gaming GPU's, Vega is a multipurpose card. AMD is "Smartly" putting alot of emphasis in the workstation and server world with Vega. Its smart because AMD has never made a dime on gaming GPU's in the past. Now with AI and the data center they see an opportunity to "Make money" and that is what its all about. Vega will succeed due to attach rate, when AMD sells a CPU over 50% of the time an AMD GPU is in the build -- the only issue was in the past nobody was buying AMD CPU's.

 

[IEC]

Here's to hoping the bottom falls out of the cryptocurrency market soon.

It already did. Anyone buying new GPUs today to mine will probably never see a ROI.

 

[Glo.]

I think AMD did what they did with VEGA on purpose. They built the card for Gaming "And" Pro space probably because they had to save money. So what we will get is a card that games well and does some great things in the pro space. It will not unseat NVDA gaming GPU's, Vega is a multipurpose card. AMD is "Smartly" putting alot of emphasis in the workstation and server world with Vega. Its smart because AMD has never made a dime on gaming GPU's in the past. Now with AI and the data center they see an opportunity to "Make money" and that is what its all about. Vega will succeed due to attach rate, when AMD sells a CPU over 50% of the time an AMD GPU is in the build -- the only issue was in the past nobody was buying AMD CPU's.

Vega is the architecture that has to compete with: GTX 1080, GTX 1080 Ti, and GV100 in different markets.

Its obvious that software is not ready for this architecture, yet.

 

[PhonakV30]

http://www.nag.co.za/2017/07/10/amd...of-shortages-thanks-to-cryptocurrency-miners/

damn XMR! It's high score while my RX480 can only do 600 h/s

Well, with a couple of tweaks to a Monero miner and a constant underclock of 1.3GHz, a Vega Frontier Edition card hits 1.16kH/s. That is 34% faster than a GeForce GTX 1080 Ti, and 43% faster than a single Radeon RX 580 8GB. If the gaming-focused version of Vega, Vega RX, hits those same numbers at stock frequencies and is priced to compete with the GeForce GTX 1080 (which only hits 0.48kH/s), then we are likely going to experience another shortage of high-end GPUs. Mining with it is almost twice as profitable as a GTX 1080 on Monero, and the card would pay for itself in about half the time as well if priced to sell at $499.

 

[nathanddrews]

RX Vega 56 benchmarks!
http://www.tweaktown.com/news/58635...-leaked-benchmarks-gtx-1070-killer/index.html

 

[Mopetar]

RX Vega 56 benchmarks!
http://www.tweaktown.com/news/58635...-leaked-benchmarks-gtx-1070-killer/index.html

This was posted in another thread (the Vega / Navi rumors one) and the conclusion over there was that either this is outright a fabrication or that it's using minimum / 99th percentile frame rates as the 1070 numbers didn't match up with what's been reported in numerous benchmarks.

 

[PeterScott]

I don't think it is fabrication, but those are pretty much the most favorable AMD cherry picked benchmarks. These are the ones AMD chooses for marketing, because GCN cards fly on them. So it is almost certainly an AMD "leak" for marketing purposes.

 

[Elixer]

I don't think it is fabrication, but those are pretty much the most favorable AMD cherry picked benchmarks. These are the ones AMD chooses for marketing, because GCN cards fly on them. So it is almost certainly an AMD "leak" for marketing purposes.

Can't be... AMD marketing only does minimum framerates these days.

Most likely a AIB/OEM board partner if anything.