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

[AtenRa]

Vega Frontier Air will be available in Germany from 12th of July at 1199 Euro.

https://www.caseking.de/search?sSearch=amd+vega

 

[cytg111]

Hopefully they'll be able to work some more performance out with driver optimizations as well as helping tweek games for it..

For the life of me I cant find it now, didnt bookmark it back then cause BS meter, but there was some article with an interview with engineer? - that stated the key to unlocking vega performance was some coding trick, point being games has to use this trick to get full perf. Anyone know of the article I speak of, cause I sure as * cant find it again.. (maybe it was 100p BS).

 

[Veradun]

It is unknown how much Apple is paying, if you go by how they treat their other suppliers, it is pretty much bottom of the line, with very little if any profit for the supplier.
Unfortunately for us, we don't really know how Vega RX & APUs will perform, so, it is way too soon to tell if they can recoup anything, or if they will bleed red for the next few quarters.

If suppliers got almost nothing from Apple why bother? Also: why is IT up for sale if losing a customer like Apple is such a small deal? :>

 

[3DVagabond]

If suppliers got almost nothing from Apple why bother? Also: why is IT up for sale if losing a customer like Apple is such a small deal? :>

Don't you know that's the only reason AMD is in anything? Because they are the only tech company in the world who is willing to do business with Apple, the consoles, etc...

 

[Veradun]

Don't you know that's the only reason AMD is in anything? Because they are the only tech company in the world who is willing to do business with Apple, the consoles, etc...

 

[Cloudfire777]

If suppliers got almost nothing from Apple why bother? Also: why is IT up for sale if losing a customer like Apple is such a small deal? :>

The profit per product to Apple is small but Apple move millions of these

 

[Veradun]

The profit per product to Apple is small but Apple move millions of these

And that tells a lot about Vega volumes and yelds (because volumes need high yelds to be high in a capacity constrained environment like 14nm is) :>

Next is: how can they fit a power hog like Vega is rumored to be in a design like that?

Something doesn't click here, I'm still waiting for RX's reviews and I'm sticking to my respin theory for now.

 

[Tee9000]

how can they fit a power hog like Vega is rumored to be in a design like that (iMacPro)?

Ahhh... lower the frequencies of course, thus operating voltages, add the best binned chips and voila, a new Nano is born. 11 TFLOPS comes to 1340 Mhz for boost but I think base frequency will be around 1200 Mhz.

 

[tviceman]

Next is: how can they fit a power hog like Vega is rumored to be in a design like that?

Haven't you seen the specs? The Mac Vega version is rated for about 11 tflops, down a solid 15% from Vega FE. And keep in mind AMD quotes peak TF, not sustained.

 

[leoneazzurro]

Haven't you seen the specs? The Mac Vega version is rated for about 11 tflops, down a solid 15% from Vega FE. And keep in mind AMD quotes peak TF, not sustained.

As everyone else do

 

[tviceman]

As everyone else do

AMD reports TFLOPs based on their max boost clock. Nvidia reports TFLOPs based on their minimum boost clocks.

Big key difference.

 

[leoneazzurro]

Nvidia does not declare maximum boost frequency, AFAIK. Anyway, I thought you were referring to utilization.

 

[Jackie60]

If we really are looking at 1080 performance this late then I might buy two if they're $300-$350 depending on xfire scaling. Otherwise my next card is Volta. There's really no excuse for such poor performance this late in the game and Raja needs to go.

 

[n0x1ous]

If we really are looking at 1080 performance this late then I might buy two if they're $300-$350 depending on xfire scaling. Otherwise my next card is Volta. There's really no excuse for such poor performance this late in the game and Raja needs to go.

they need Eric Demers back from Qualcomm

 

[french toast]

If this Rx Vega launches with perf that is 15% above a gtx 1080 and consumes 300w it will still be ok, so long as it's priced right 449$ would make this a decent card, assuming it can get another 8-10℅ overclock.
As it a new uarch drivers will improve massively over the next year as AMD will look to implement it's fine wine technology™, playing into the perception that AMD architectures improve more than nvidia over time. (Even though tech nerds know that performance should have been there on day one)

If we get that scenario then it won't be too bad for amd, not great but not a disaster imo as they will be able to compete in a market that has been given to nvidia on a plate the last 18 months.

Gtx1080ti will still be the performance and efficiency king, but
Honestly if you had a 450-500$ budget then Rx Vega is the way to go until poor Volta arrives to deliver a smack down Q2 2018.
Rx Vega would likely vastly outpace gtx 1080 over the next 6 months even if it consumes way more power.

 

[Glo.]

If this Rx Vega launches with perf that is 15% above a gtx 1080 and consumes 300w it will still be ok, so long as it's priced right 449$ would make this a decent card, assuming it can get another 8-10℅ overclock.
As it a new uarch drivers will improve massively over the next year as AMD will look to implement it's fine wine technology™, playing into the perception that AMD architectures improve more than nvidia over time. (Even though tech nerds know that performance should have been there on day one)

If we get that scenario then it won't be too bad for amd, not great but not a disaster imo as they will be able to compete in a market that has been given to nvidia on a plate the last 18 months.

Gtx1080ti will still be the performance and efficiency king, but
Honestly if you had a 450-500$ budget then Rx Vega is the way to go until poor Volta arrives to deliver a smack down Q2 2018.
Rx Vega would likely vastly outpace gtx 1080 over the next 6 months even if it consumes way more power.

Nope. The mindshare perspective will only be somewhat redeemed if Vega will cost 399$, for 300W part, that is barely faster than 175W GTX 1080.

If AMD will come up with Nano Vega that has 150W TDP, 1.2 GHz, and price it lower than GTX 1070 - then yes, this is a killer GPU offering, that can drive AMD's brand perception.

Imagine however one thing. Vega 11, in current state, as we look at Vega 10, will be slower or on par with RX 580.

What this will tell about AMD?

 

[PeterScott]

Sigh, these myths of saturated markets.

There wasn't a 1080 for 1070 price, or a 1080 Ti for 1080 price. And even if there were, there is constantly a stream of new people upgrading. Most users don't upgrade every year, yet despite the massive success of 970, the 1070 sold by the bucketloads.
Why?
Because there are other people looking to upgrade from their Kepler, Fermi, or even older era cards.

Sure, there will be a trickle, but being a year late, means the big wave of buying has expired. Also even when AMD is on time with similar price/performance, they don't do very well. GTX 1060 and RX 480 were similar price/performance at about the same time, and NVidia outsold them 5:1.

Vega will likely have volume issues, which is also a profitability issues.

As for die size, both 14nm and 16nm are very mature, and yields shouldn't be a problem. A decent profit would still be made even at 1070 pricing.

With the already mentioned volume issue, A >Titan sized die, more expensive HBM memory, selling at 1070 pricing will be a disaster.

 

[Tup3x]

Nope. The mindshare perspective will only be somewhat redeemed if Vega will cost 399$, for 300W part, that is barely faster than 175W GTX 1080.

If AMD will come up with Nano Vega that has 150W TDP, 1.2 GHz, and price it lower than GTX 1070 - then yes, this is a killer GPU offering, that can drive AMD's brand perception.

Imagine however one thing. Vega 11, in current state, as we look at Vega 10, will be slower or on par with RX 580.

What this will tell about AMD?

They are probably just refactoring their GPU into something... That approach may or may not pay off later but currently it looks like Vega is just a testbed for infinity fabric. Probably the whole point of Vega is to make GPU arch for their APUs. Or so it seems...

 

[french toast]

Nope. The mindshare perspective will only be somewhat redeemed if Vega will cost 399$, for 300W part, that is barely faster than 175W GTX 1080.

If AMD will come up with Nano Vega that has 150W TDP, 1.2 GHz, and price it lower than GTX 1070 - then yes, this is a killer GPU offering, that can drive AMD's brand perception.

Imagine however one thing. Vega 11, in current state, as we look at Vega 10, will be slower or on par with RX 580.

What this will tell about AMD?

I'm not saying it will be massively successful, but it would at least be competitive, perhaps favourably so in its price point, how much do OC gtx 1080s consume? 230-250w I presume? If Vega is faster than that for another 20℅ extra tdp PLUS over clocking and the certainty of much better drivers and speed over time I think it would look a decent deal at 450$.

This makes Rx Vega essentially a 1080 after market/OC competitor which it competes quite well.
Not sure about the profit margins on a die that big with HBM2 but that's another discussion entirely.

 

[Glo.]

They are probably just refactoring their GPU into something... That approach may or may not pay off later but currently it looks like Vega is just a testbed for infinity fabric. Probably the whole point of Vega is to make GPU arch for their APUs. Or so it seems...

I have no idea what has happened with Vega.

On paper it should not perform like it is performing. It has higher throughput than Fiji, and Polaris, both on compute and graphics, yet in graphics it performs like it has lower throughput.

I don't know.

Is AMD's marketing this rubbish to launch unfinished product to lower the hype about it, and surprise with performance at the release of gaming cards?

I would not count on this. Its hard for me to also think this GPU is a failure, because on paper this is very brilliant, and modern architecture, with very high throughput.

 

[Head1985]

another vega vs 1080 and 1080ti

 

[tential]

Nope. The mindshare perspective will only be somewhat redeemed if Vega will cost 399$, for 300W part, that is barely faster than 175W GTX 1080.

If AMD will come up with Nano Vega that has 150W TDP, 1.2 GHz, and price it lower than GTX 1070 - then yes, this is a killer GPU offering, that can drive AMD's brand perception.

Imagine however one thing. Vega 11, in current state, as we look at Vega 10, will be slower or on par with RX 580.

What this will tell about AMD?

When I brought up that Vega 11 stuff no one wanted to talk a out it. It's too far away I was told and I was accused with all types of doll and gloom Now a few weeks later it's OK? Yes I've been saying this as well about Vega 11.

However, recall Polaris. The 11 chip in Polaris brought the efficiency improvements. So Vega 11 really depends on how much of an efficiency improvement it brings.
. But yes, Vega 11 is in a by far worse state as it will come out so close to Volta that I'm wondering if it will come out at all. I wonder if there would be any timetable gains from skipping production of vega 11 and moving to the next move.

 

[Snarf Snarf]

This is entirely an outlandish idea that I was entertaining today, what if Vega 10 is an HPC compute card and Vega 11 is the gaming variant without the extra wasted die space that will be leaner and more efficient?

Like @Glo. mentions the talk of Vega 11 has completely disappeared. Is the reason the RX Vega will be much better than the FE because the silicon is indeed actually different? I'm not really sure any of whats in the FE can be cut out to improve efficiency because it doesn't have DP performance correct?

 

[PeterScott]

I have no idea what has happened with Vega.

On paper it should not perform like it is performing. It has higher throughput than Fiji, and Polaris, both on compute and graphics, yet in graphics it performs like it has lower throughput.

I don't know.

Is AMD's marketing this rubbish to launch unfinished product to lower the hype about it, and surprise with performance at the release of gaming cards?

I would not count on this. Its hard for me to also think this GPU is a failure, because on paper this is very brilliant, and modern architecture, with very high throughput.

From a graphics perspective, It has the exact same Shader/Render/Texture unit count as Fiji, and has slightly less memory bandwidth than Fiji, and that is exactly what it performs like.

For compute loads, FP16 was improved to double performance, but that has ZERO effect on graphics.

The Missing part for graphics, is that there is supposed to be some kind of Tile Based Rendering capability added to improve efficiency and performance, that is clearly doing nothing.

Given the exact same unit count as Fiji, it is almost like AMD placed all their bets on performance increase coming form the Tile Based rendering that is still MIA.

Without that, it is just a clock speed boosted Fiji. The delay for Vega RX is all about getting TBR working right.

 

[crisium]

I'm not saying it will be massively successful, but it would at least be competitive, perhaps favourably so in its price point, how much do OC gtx 1080s consume? 230-250w I presume? If Vega is faster than that for another 20℅ extra tdp PLUS over clocking and the certainty of much better drivers and speed over time I think it would look a decent deal at 450$.

This makes Rx Vega essentially a 1080 after market/OC competitor which it competes quite well.
Not sure about the profit margins on a die that big with HBM2 but that's another discussion entirely.

TPU Peak Gaming Power consumption (typical gaming is lower) for aftermarket GTX 1080s:
203w
203w
213w
217w
224w
234w (11Gbps)
243w
245w

Quite the variance, but they all beat the stock blower (184w) in performance by around 5-10% at 1440p, albeit with worse performance-per-watt as Pascal scales poorly above 1800mhz.

Vega wattage will be noticeably above any of these, especially aftermarket OC variants.