AMD Vega Hardware Reviews

[lanek]

If the clockspeed fluctuates during the test and downclocks a lot, what is the actual speed sent to Futuremark?

For what i have understand, FM only read the clock default ( as will read GPUz )... Basically if you set the turbo clock to 1630mhz, it will read 1630mhz ( even if when running the gpu was at 1200 or 1800mhz )

If you set your fan at 100% and the gpu dont throttle vs throttle by default, FM will read the same clock.

Theres something about the driver, 22.9.xxx.xxxx are closer for me to the OpenCL version of the driver ( not the driver package, and not even the D3D version )

On my system: OpencL version is 22.19.171.257 ... the driver reported by Futuremark is 22.19.653.0 The D3D block( driver ) on last official driver is 9.14.10.01261.

Driver Packaging Version
17.10.3211.1011-170612a-315159E-CrimsonReLive


the only who seems close to the 22.19.653.0 is the OpenCL, the problem is OpenCL driver can go through several drivers package .. they are not forcibly updated for each driver ( in reality, if this the stable last one, theres not much reason to update it ). and they are not related to graphics optimization anyway.

so i cant even conclude that drivers was the same. ( as different block ( D3D, openCL, audio, settings version ) can be different on a driver package. ( when mod manualy driver we just change the right "block" in the inf. or even copy text lines of code directly in it (

 

[Kaotik]

The only thing who seems to tell them that thoses scores are overclocked, is because they all use same driver ( well at least it show same driver version ( dont mean block inside are the same ), same written default frequencies but graphics score are higher.

Just aswell you could say the lower scores are underclocked, exactly same reasoning. 3DMark reports same clocks for each run so there's no reason whatsoever to think the card is OC'd in any of them.

 

[Geeforcer]

The issue is obvious: in the cases where scores are lower, the AMD tester is running the test in monitors's native resolution (4K).

 

[BacBeyond]

Just aswell you could say the lower scores are underclocked, exactly same reasoning. 3DMark reports same clocks for each run so there's no reason whatsoever to think the card is OC'd in any of them.

And this makes more sense, since if the max is 1630 having it throttle would still show it as 1630, it would just never actually run at that speed. If it was OC'd (>1630) it would be reporting those speeds that the driver was reporting it could max at.

Same reason we saw throttling Vega FE's report as 1600 in earlier tests even though they only ran @ 1400s.

 

[Deleted member 13524]

Just aswell you could say the lower scores are underclocked, exactly same reasoning. 3DMark reports same clocks for each run so there's no reason whatsoever to think the card is OC'd in any of them.

The guy from AdoredTV checked the reference numbers from each of those 3dmark11 results. The ref number basically means the order of arrival for each published benchmark. There's no relation between the time of arrival and the score, it's pretty random actually, so it's not like the tester was gradually increasing clocks throughout the benchmark runs.

It looks more like the tester was using different driver parameters than messing with the clocks.

 

[gamervivek]

The fastest and the slowest scores were done on 6700K and 1800X respectively. And 3dmark does get the overclocked speeds, for instance in this frontier benchmark,

http://www.3dmark.com/fs/13051960

The above graphics score is still 250 points shy of the LN2 1.4Ghz Fury with 1Ghz HBM on which I was basing Vega performance expectations.

http://wccftech.com/asus-radeon-r9-...y-unlocked-fury-features-1-tbs-bandwidth-ln2/

A youtuber who does great graphics cards comparison videos with different games has got the Vega FE, DOOM is up first,

 

[Pressure]

What's the results?

 

[yuri]

What's the results?

Vulkan, 4k, AA off - it reaches approximately GTX 1080 perf.

 

[AlNom]

:s

ROP limited?

 

[ieldra]

:s

ROP limited?

Similar behavior in AotS which is definitely not ROP limited

 

[Cat Merc]

Similar behavior in AotS which is definitely not ROP limited

Had the Fury X been included there, it would have been almost dead on with FE in the results here.

This is just strange.

 

[Alexko]

Probably just very immature drivers. Frankly, I think all this speculation around Vega FE and gaming is (almost entirely) pointless at this point.

 

[Malo]

Probably just very immature drivers. Frankly, I think all this speculation around Vega FE and gaming is (almost entirely) pointless at this point.

It's only pointless by virtue of not having RX to compare to.

 

[CarstenS]

It should not be.

Anyway, since I cannot remember it for sure from the top of my hat: Did AMD explicitly say, that RX Vega, i.e. the gamer variant would use exactly the same chip?

 

[eastmen]

The fastest and the slowest scores were done on 6700K and 1800X respectively. And 3dmark does get the overclocked speeds, for instance in this frontier benchmark,

http://www.3dmark.com/fs/13051960

The above graphics score is still 250 points shy of the LN2 1.4Ghz Fury with 1Ghz HBM on which I was basing Vega performance expectations.

http://wccftech.com/asus-radeon-r9-...y-unlocked-fury-features-1-tbs-bandwidth-ln2/

A youtuber who does great graphics cards comparison videos with different games has got the Vega FE, DOOM is up first,

in doom it looks like it edges out the 1080 esp when it can stay above 1500mhz. IF amd can release the gaming edition at more stable clocks and get more out of the driver it could solidly land between the 1080 and 1080 ti and if priced closer to a 1080 it can be a big seller. Of course this is just doom it would have to do that well in other titles also

 

[seahawk]

in doom it looks like it edges out the 1080 esp when it can stay above 1500mhz. IF amd can release the gaming edition at more stable clocks and get more out of the driver it could solidly land between the 1080 and 1080 ti and if priced closer to a 1080 it can be a big seller. Of course this is just doom it would have to do that well in other titles also

However this would mean an AiO cooler most likely, if they want to get the clocks up without pushing power consumption further. Add the nearly 500mm² big chip, HBM2 and interposer to compete with the small GP104 is not a good looking option. I think the driver has so many features deactivated that this must be intentional, which means RX Vega will probably beat the Titan XP by a good margin and the price, power consumption and chip size would fit well to the competition.

 

[gamervivek]

What's the results?

<10% better than a 1.8Ghz 1080 and 1080Ti has a big lead. DOOM is of course very AMD favoring, he played Crysis 3 at 1440p first and the results were likely less than a 1080, close to 1070 in places.

Taking a look at the most recent 1080Ti review on TPU, the Fury X is ~30% faster than the 390 or around 20% faster than the 390X whereas the difference in TFLOPs is >40%. I was hoping that Vega changes this with all those improvements talked about but it isn't looking like that's going to happen by the end of this month, it'd be pretty good if they can get it to perform linearly with the clockspeed compared to Fury X.

 

[eastmen]

However this would mean an AiO cooler most likely, if they want to get the clocks up without pushing power consumption further. Add the nearly 500mm² big chip, HBM2 and interposer to compete with the small GP104 is not a good looking option. I think the driver has so many features deactivated that this must be intentional, which means RX Vega will probably beat the Titan XP by a good margin and the price, power consumption and chip size would fit well to the competition.

Well i doubt your last part . however with the first part we don't know how old these chips are , amd has been showing vega off for almost a year , tthe frontier editions could be older chips that are more power hungry while the gamer editions will get less power hungry chips. Less poewr should equal less heat output. We also don'tk now if any parts of the chip that aren't needed for gaming will be disabled further reducing power usage. And then again I am sure the 4 and 8 gig varients will use less power . So i will wait and see , its not much longer till the announcement

 

[eastmen]

<10% better than a 1.8Ghz 1080 and 1080Ti has a big lead. DOOM is of course very AMD favoring, he played Crysis 3 at 1440p first and the results were likely less than a 1080, close to 1070 in places.

Taking a look at the most recent 1080Ti review on TPU, the Fury X is ~30% faster than the 390 or around 20% faster than the 390X whereas the difference in TFLOPs is >40%. I was hoping that Vega changes this with all those improvements talked about but it isn't looking like that's going to happen by the end of this month, it'd be pretty good if they can get it to perform linearly with the clockspeed compared to Fury X.

shit you can see it drop down to under 800mhz at some points in that video . Something is def screwy with the frontier edition

 

[Rootax]

Better cooling can lead to less power consumption too. A AiO like on the Fury X will do better than the blower on the FE edition, that's for sure (and if RX Vega is good enough to upgrade from my Fury X, I'll will wait for a block from EK or XSPC anyway...)