TR Forums

This has been blowing up reddit/r/amd and WCCFTech; Fellow acquired a Vega FE and ran it in 3D Mark, getting a score on par with a GTX 1080:

https://disqus.com/by/klaudiuszkaczmarzyk/

Vega FE acheived a graphics score of 22,916. My GTX 1080 at 2012/5500 gets 23,000+ for reference:

http://www.3dmark.com/compare/fs/12895098/fs/12987144

There's of course going to be the usual brouhaha over drivers not being "optimized" so the test isn't "fair".

On the flipside, while it's probably true that the drivers are not perfectly optimized, you'd think that a card with Vega's level of raw hardware power and Vega's level of lateness (meaning AMD has already had time to at least get competent drivers together) would at least show a decent boost over last year's 1080.

We'll see how those numbers translate to other game benchmarks.

Indeed. I'm not expecting much though since we're only a month out from release. A 5% improvement over what is achievable now is what I'd expect.

Darn. I don't know what I was expecting but I was hoping it'd be a little faster than a 1080. If it somehow consumes less power and comes in at a lower price I'd be a little more satisfied.

This is the one and only time in my life that I'd like to buy a flagship-class GPU and I want some competition to make things interesting!

That still leaves room for the top RX Vega to compete with the 1080 Ti as all indications from AMD have been that RX Vega will be better in gaming than the FE. But I'm not expecting Vega to beat Nvidia's products at release at this point.

Hmm... The guy in the link says it wasn't holding a constant 1600 MHz during the 3DMark run.

DrDominodog51 wrote:

Hmm... The guy in the link says it wasn't holding a constant 1600 MHz during the 3DMark run.

That could be due to a lot of factors including throttling (bad) or the fact that some sections of the benchmark are just less demanding (neutral).

I would not count on Vega (top o the line) using less power than 1080. That'd be pretty crazy engineering by AMD, but given the Polaris stuff, and GF process tech, including the chips we know that have been produced on it, there is very little chance they'll use less power than Nvidia at the same performance level.

AMD keeping even with NVIDIA on power/performance is really a win for us. If both companies are at parity they'll have to compete on price in the short term, and find a way to increase generational price/performance in the longer term.

Vhalidictes wrote:

AMD keeping even with NVIDIA on power/performance is really a win for us. If both companies are at parity they'll have to compete on price in the short term, and find a way to increase generational price/performance in the longer term.

I would agree with you if AMD were even close to Nvidia on efficiency but they aren't. Like I said, unless they did some amazing engineering on Vega, Pascal will still be quite a bit ahead on perf/W.

Not to gloat, but you can't say I didn't warn you: http://techreport.com/blog/31924/spitba ... phics-card

Virtually every prediction I made in that post was proven correct yesterday: http://techreport.com/news/32163/update ... 999-and-up

Oh wow. I'm not sure the Newegg page is right, but it suggests the TDP is 300W for the air cooled. And it says 375W for the liquid version!!! That's gonna be warm.

https://www.newegg.com/Product/Product. ... 00524&SID=

https://www.newegg.com/Product/Product. ... 00524&SID=

edit: fixing the broken link.

Jeff Kampman wrote:

Not to gloat, but you can't say I didn't warn you: http://techreport.com/blog/31924/spitba ... phics-card

Virtually every prediction I made in that post was proven correct yesterday: http://techreport.com/news/32163/update ... 999-and-up

If it helps, Jeff, I thought you were perfectly correct when the post first came out. The numbers just added up.

The only real variable (as also mentioned in your post) is whether the Game version of Vega has a different power curve or design tweaks.

The closer the RX Vega gets to launch, the more I feel like it's going to be a flop. Kinda makes me feel like AMDs engineers are diverted to Ryzen and the GPU team was left barebones to hobble along for a couple years.

This seems to be in line with TR's prediction of Vega's performance (that it would be around GTX 1080 levels).

Having said that, 3DMark Firestrike scores in general seem to be all over the place. I never see a 1080 get the same score in 2 different reviews, and I've seen scores vary wildly from 19k to 23k on the GTX 1080. I don't think 3DMark/Firestrike is a particularly reliable or repeatable indicator of real-world gaming performance, and especially not single-user scores in isolation.

It's also worth mentioning that the Vega FE is a 'prosumer' card, and may not have the same clock, thermal and throttling behavior that consumer/gaming versions will have.

But if Vega punches above its weight class to actually challenge the 1080 Ti when it is fully reviewed, I'd be surprised.

EDIT: Ninja'ed by Jeff. Damn you, Jeff! /shakes fist.

DPete27 wrote:

The closer the RX Vega gets to launch, the more I feel like it's going to be a flop. Kinda makes me feel like AMDs engineers are diverted to Ryzen and the GPU team was left barebones to hobble along for a couple years.

They may work together on some things, but RTG and the CPU teams are mostly separate AFAIK. I'm not even sure they work in the same place.

DPete27 wrote:

The closer the RX Vega gets to launch, the more I feel like it's going to be a flop. Kinda makes me feel like AMDs engineers are diverted to Ryzen and the GPU team was left barebones to hobble along for a couple years.

I don't know that this is fair to the AMD engineers. They are coming from a huge power efficiency deficit over many previous generations, and the entire time they've been working on that they've also needed to maintain a least a general performance parity with NVIDIA.

Of course as consumers we don't have to care about that, we just need to buy the better card for the needed performance level.

chuckula wrote:

DrDominodog51 wrote:

Hmm... The guy in the link says it wasn't holding a constant 1600 MHz during the 3DMark run.

That could be due to a lot of factors including throttling (bad) or the fact that some sections of the benchmark are just less demanding (neutral).

I'm inclined to believe the former because the less demanding part the benchmark (Graphics Test 1) still uses nearly 100% of the GPU. It is possible the link's OP is confused by the downclocking between tests.

This is from my Fury running Firestrike at stock:

(Sorry for the poor writing with my mouse) The numbers/letters state the test which corresponds with the spike its on.

The Fury goes down 7 Mhz at most during the least demanding segment of the test.

In anycase, ~23000 score is pretty bad. Either Vega needs to overclock well or get a big boost from disabling tessellation because a Fury can nearly reach 20000 overclocked.

DancinJack wrote:

Vhalidictes wrote:

AMD keeping even with NVIDIA on power/performance is really a win for us. If both companies are at parity they'll have to compete on price in the short term, and find a way to increase generational price/performance in the longer term.

I would agree with you if AMD were even close to Nvidia on efficiency but they aren't. Like I said, unless they did some amazing engineering on Vega, Pascal will still be quite a bit ahead on perf/W.

Yeah. The 1080 is a 180W card. Vega FE (air cooled) is... 300W. @_@

DancinJack wrote:

I would not count on Vega (top o the line) using less power than 1080. That'd be pretty crazy engineering by AMD, but given the Polaris stuff, and GF process tech, including the chips we know that have been produced on it, there is very little chance they'll use less power than Nvidia at the same performance level.

Isn't Vega fabbed by TSMC on 16nm FinFET? We also don't have any proof that GF process are inherently less efficient than TSMC one

stefem wrote:

DancinJack wrote:

I would not count on Vega (top o the line) using less power than 1080. That'd be pretty crazy engineering by AMD, but given the Polaris stuff, and GF process tech, including the chips we know that have been produced on it, there is very little chance they'll use less power than Nvidia at the same performance level.

Isn't Vega fabbed by TSMC on 16nm FinFET? We also don't have any proof that GF process are inherently less efficient than TSMC one

Not sure where you got that idea? Any link? As far as I know, AMD is using GF for all their products. Sure, we don't have any apples to apples proof, but we can draw some conclusions from products that are fabbed on those processes and how they clock/perform.

DancinJack wrote:

stefem wrote:

DancinJack wrote:

I would not count on Vega (top o the line) using less power than 1080. That'd be pretty crazy engineering by AMD, but given the Polaris stuff, and GF process tech, including the chips we know that have been produced on it, there is very little chance they'll use less power than Nvidia at the same performance level.

Isn't Vega fabbed by TSMC on 16nm FinFET? We also don't have any proof that GF process are inherently less efficient than TSMC one

Not sure where you got that idea? Any link? As far as I know, AMD is using GF for all their products. Sure, we don't have any apples to apples proof, but we can draw some conclusions from products that are fabbed on those processes and how they clock/perform.

Well, I'm not sure, it was an old rumor, that why I asked. But what I know is that AMD paid a fee to GF to derogate from their contract and be able to go to other foundries.

If you are referring to mobile SoC (which was Apple to Apple in any sense ) I don't think that would be a worthy comparison since desktop grade GPUs are far different beast, and If I recall correctly the difference was inconsistent, some test did better on chip from one fab and some to the other and NVIDIA's PG107 which is produced by Samsung didn't show any tangible efficiency difference.

stefem wrote:

DancinJack wrote:

stefem wrote:

Isn't Vega fabbed by TSMC on 16nm FinFET? We also don't have any proof that GF process are inherently less efficient than TSMC one

Not sure where you got that idea? Any link? As far as I know, AMD is using GF for all their products. Sure, we don't have any apples to apples proof, but we can draw some conclusions from products that are fabbed on those processes and how they clock/perform.

Well, I'm not sure, it was an old rumor, that why I asked. But what I know is that AMD paid a fee to GF to derogate from their contract and be able to go to other foundries.

I believe that was for CPUs and APUs.

Until Polaris (which is fabbed at GF), AMD had been manufacturing GPUs at TSMC.

AMD also has some sort of fab deal with Samsung, although I don't know if they've built anything more significant than NAND there.

Voldenuit wrote:

stefem wrote:

DancinJack wrote:

Not sure where you got that idea? Any link? As far as I know, AMD is using GF for all their products. Sure, we don't have any apples to apples proof, but we can draw some conclusions from products that are fabbed on those processes and how they clock/perform.

Well, I'm not sure, it was an old rumor, that why I asked. But what I know is that AMD paid a fee to GF to derogate from their contract and be able to go to other foundries.

I believe that was for CPUs and APUs.

Until Polaris (which is fabbed at GF), AMD had been manufacturing GPUs at TSMC.

AMD also has some sort of fab deal with Samsung, although I don't know if they've built anything more significant than NAND there.

I don't know for what it was, I think that the contract to produce console chip is between Microsoft/Sony and TSMC so AMD didn't had a reason to pay GE for that.
Samsung is already fabbing GP107 for NVIDIA but I don't think they are making anything for AMD right now, some times ago they released a statement saying that their primary manufacturing partners were GF and TSMC with Samsung being an option.

DrDominodog51 wrote:

Hmm... The guy in the link says it wasn't holding a constant 1600 MHz during the 3DMark run.

Could that have something to do with the tester using a barely adequate PSU (550W)?

cynan wrote:

DrDominodog51 wrote:

Hmm... The guy in the link says it wasn't holding a constant 1600 MHz during the 3DMark run.

Could that have something to do with the tester using a barely adequate PSU (550W)?

550W PSU with 504W available on the 12V line. Not inadequate (on paper).

Voldenuit wrote:

cynan wrote:

DrDominodog51 wrote:

Hmm... The guy in the link says it wasn't holding a constant 1600 MHz during the 3DMark run.

Could that have something to do with the tester using a barely adequate PSU (550W)?

550W PSU with 504W available on the 12V line. Not inadequate (on paper).

To put that in perspective, people were screaming about heavily overclocked i9 7900X systems under torture-test loads that were pulling close to 400 watts from the wall (not from the PSU itself, which would be somewhat lower due to conversion inefficiency).

The tested platform here is a 4790K that's nowhere near as power hungry as a heavily overclocked 10 core CPU, so hopefully even with a power-hungry GPU the FireStrike benchmark (which is heavy but not a power-bug GPU test) wouldn't overtax the PSU.

To put this in perspective for most people. My r9fury and heavily overclocked 4930k pulls about 500 at the wall. I'm talking about cpu burn and kombuster at the same time.

I am a little disappointed in the score it is getting. But hopefully like people have been saying RX Vega should be a bit faster. I was expecting it to match 1080 Ti... Something seems wrong because it's 52% higher clocked than Fury X yet only around 40% faster...

That said , If it can match AIB 1080 for £450~ I will be happy, as they still sell for around £480+ here in the UK. The cut-down version should be able to match beat 1070 for the same or less price. That will still be nice and I will likely get one to replace my RX 480 as it struggles a bit with my 1440p monitor.

cynan wrote:

DrDominodog51 wrote:

Hmm... The guy in the link says it wasn't holding a constant 1600 MHz during the 3DMark run.

Could that have something to do with the tester using a barely adequate PSU (550W)?

The klaudius guy said he was able to get sustained 1600 MHz with the right Wattman settings, so it's not the PSU.