Vega RX tested against the GTX 1080ti

[the54thvoid]

You haven't read everything that I said or you didn't pay enough attention. Vulkan uses low level functionalities taken from Mantle that are no longer hardware and platform dependent otherwise if it does contain hardware design specific functionalities like you claim it wouldn't work on hardware different from AMD's. An open source , cross platform low-level API , cannot be biased from a functional point of view towards a set of hardware , it either supports it or not. ( and there is definitely hardware that doesn't support it from both AMD and Nvidia ) . That is like saying assembly code is biased towards Intel because you can use AVX instructions that their chips are built to better support it. While we all know that is dumb , AMD is free to make a design that better fits those instructions as well without having to fight with an inherent bias for Intel . All these arguments against Vulkan as a viable medium for comparing performance can be literally had for every other API.

You misread my post. You initially said Mantle formed Vulkan, I was suggesting it wasn't that way as it would undermine your comment.
Vulkan has little (though not nothing) to do with Mantle. And my other point was AMD worked closely with Khronos to get the best out of it for their hardware. I read a lot about Vulkan and how AMD worked hard to make it work well for them.

 

[FordGT90Concept]

Mantle was co-developed between AMD and DICE, and yes, Mantle is the foundation of Vulkan which Khronos Group expanded:
https://www.extremetech.com/extreme/200286-not-dead-yet-amds-mantle-powers-new-vulkan-api-vr-efforts

Mantle was designed for GCN which is why GCN has generally better support for it. Pascal supports Vulkan as well as GCN does.

Edit: There are Khronos Group documents that directly speak to Mantle.

 

[Vya Domus]

And my other point was AMD worked closely with Khronos to get the best out of it for their hardware. I read a lot about Vulkan and how AMD worked hard to make it work well for them.

And I though you were suggesting it is inherently biased towards AMD no matter what , which seems to be the attitude most people have that is simply wrong. Nvidia can shell out at any time hardware that makes even better use of it than AMD's. As a matter of fact they technically sort of did , a 1080 with the same ballpark compute performance to a Fury X beats it by fair amount in Doom. In no way games using it should be dismissed in comparisons , it is dumb.

 

[JalleR]

They wish , they can't afford to do it. They were the ones that pioneered the exclusion of shady tactics fighting Intel off.



Nope , no one gets paid. Not directly. If you review a product from a company and they don't like your review in the future they wont send you a sample , if they don't send you a sample you as reviewer whom mostly can't afford to buy all the stuff your testing are going to be in a tight spot. Nowadays companies can't really pay you for "fake" reviews , that's illegal( at least not the ones as big as AMD ) but they can hurt your paycheck indirectly.

Aaa so that is Why TPU is so AMD friendly these days...... CHECK...... And we all know TPU make some of the best reviews so I guess the Knife in the Heart on the RX480 review was not in AMD's interest lol........ AMD try selling your S..... on TV-Shop that is where this Whole Circus belongs.....

 

[XiGMAKiD]

AMD's Mantle is used as the foundation of Vulkan API made by Khronos Group which the President is also working for Nvidia

 

[Vya Domus]

AMD's Mantle is used as the foundation of Vulkan API made by Khronos Group which the President is also working for Nvidia

It's all just a conspiracy.

 

[trog100]

i just actually watched the video.. 6 people could not tell the difference.. one saw a night and day difference.. 3 i think saw a minor difference .. 3 preferred the amd card and one the nvidia card..

conclusions.. frame rates are not the be all and end all it takes a fair variations to be noticeable.. nvidia loses out because of the high cost of g-sync panels..

in the end it comes down to cost with amd the clear winner even though i am sure nvidia will have the higher frame rates..

i am pretty sure the same would happen if similar blind tests were done between ryzen and intel gaming systems with identical graphics cards.. at normal gaming resolutions any half decent cpu will do and the gpu is king amd or nvidia.. he he

trog

 

[Vya Domus]

Considering they brought up the 300$ difference again even though these systems are clearly more expensive one can conclude this time they used a different more expensive card, the only thing that I take away from this is that AMD might in fact have a card to go against the 1080ti one way or another.

 

[the54thvoid]

And I though you were suggesting it is inherently biased towards AMD no matter what

No, people think Vulkan is biased but it's (IMO) down to how DX9 up to DX11 proved difficult for AMD to adequately code for, making innefficient use of it's (some might argue) superior hardware. Vulkan allows more effective use of the AMD hardware without any penalty to Nvidia. Nvidia on the other hand coded very well during DX9-11 so the uplift in performance wasn't there until Pascal added some magic sauce to make better use of Nvidia's warp scheduling. If you compare hardware, AMD have on paper at least some very powerful looking cards which lose out to Nvidia's streamlined (after Fermi) approach. I think AMD's problem is linked to it's hardware though in that more hardware = more power = less desirable design for mobile and OEM's (that would prefer lower power envelopes).
If AMD stripped back on compute (like Nvidia did) they'd probably have a higher efficiency and allow faster clocks (that power play of core versus clocks is out of my league to discuss though). Look at Volta, GV100 as example of Nvidia's approach (which it can afford to do). GV100 is a different card to what we consumers will likely see - we have no need in gaming for Tensor (AI) cores. AMD design the same card for HPC and gaming. One is ideal, one is over engineered.

 

[EarthDog]

@Vya Domus

I dont think this test should be thrown away.. hardly. The circumstances of the test need to be taken into consideration is all. Its absolute best light and subjective.

You said yourself GE doesnt have many games... this is one where amd gpus respond very well. Its running vulkan...which nvidia doesnt scale as well in... they are limiting frame rates and using free/gsync based panels.

I guess what i am saying is, first... cool, well done amd. You beat it out, subjectively, in the most favorable circumstances possible. Can we see what that is like without using vulkan (since so few games support it)? Without limiting fps or using free/gsync panels? Id think this is how most people game...

 

[FordGT90Concept]

No, people think Vulkan is biased but it's (IMO) down to how DX9 up to DX11 proved difficult for AMD to adequately code for, making innefficient use of it's (some might argue) superior hardware. Vulkan allows more effective use of the AMD hardware without any penalty to Nvidia. Nvidia on the other hand coded very well during DX9-11 so the uplift in performance wasn't there until Pascal added some magic sauce to make better use of Nvidia's warp scheduling. If you compare hardware, AMD have on paper at least some very powerful looking cards which lose out to Nvidia's streamlined (after Fermi) approach. I think AMD's problem is linked to it's hardware though in that more hardware = more power = less desirable design for mobile and OEM's (that would prefer lower power envelopes).
If AMD stripped back on compute (like Nvidia did) they'd probably have a higher efficiency and allow faster clocks (that power play of core versus clocks is out of my league to discuss though). Look at Volta, GV100 as example of Nvidia's approach (which it can afford to do). GV100 is a different card to what we consumers will likely see - we have no need in gaming for Tensor (AI) cores. AMD design the same card for HPC and gaming. One is ideal, one is over engineered.

Vulkan moves most graphics optimizations to software developers. D3D11 and down as well as OGL put the graphics optimizations in hardware developer hands. NVIDIA was better at driver optimizations than AMD was. Or rather, I think NVIDIA is more willing to cut corners for higher frames than AMD is.

Price versus TFLOPs, AMD usually comes out on top. In gaming, AMD GPUs have more idle hardware. Even gimped at gaming as Vega is, it has higher theoretical TFLOPs than GTX 1080 Ti and Titan Xp.

As I said, NVIDIA cuts corners. Really crappy FP64 and FP16 performance is evidence of this. NVIDIA makes more GPU SKUs to force higher budget customers to higher budget hardware. AMD likes to give everyone access to everything so that, for example, games change significantly over to FP16, Vega will see a doubling of performance while Pascal will be better off using FP32 not benefiting at all.

Edit: if memory serves, D3D12_1 full compliance requires FP10 or FP16 support. Pascal is screwed on that one because they don't want AI customers buying up GeForce cards. Edit: Yup.

 

[Vya Domus]

@Vya Domus
I guess what i am saying is, first... cool, well done amd. You beat it out, subjectively, in the most favorable circumstances possible. Can we see what that is like without using vulkan (since so few games support it)? Without limiting fps or using free/gsync panels? Id think this is how most people game...

I don't even consider who beats who with Vulkan. All I am saying is that this is not subjective at all. If you want to see what is like without it you are probably talking about DX 11 , but we already know that side of the story , AMD due to their drivers wont perform great. And DX12 is too much of a clusterfuck to take anything from it. With something like Vulkan you are seeing AMD and Nvidia cards on equal grounds with no software handicap probably for the first time since DX 9. I consider Doom to be the most relevant game to compare cards at their maximum potential , it scales perfectly with core count , clocks , memory bandwidth , compute performance , with no CPU bottlenecks. It is unfortunate that it is the only game where you get that.

Yes this test is useless with no fps/frametime meter , I already agreed with that.

 

[EarthDog]

It couldnt be more subjective bud. There arent any measureable metrics to form an objective conclusion. Just what 'looks' and 'feels' smoother. This testing is based off opinions and in a best case scenario environment. Id simply like to see more testing in games not using vulkan, without limiting fps and without using free/gsync panels.

Edit: Please refrain from putting words in my mouth. . I did not mention, nor infer anything about having a fps counter. Same test in different games without limiting fps or using free/gsync panels. I dont care about the fps count. Lets see if that same subjective result holds up without the crutches. Simple.

 

[Vya Domus]

It couldnt be more subjective bud. There arent any measureable metrics to form an objective conclusion. Just what 'looks' and 'feels' smoother. This testing is based off opinions and in a best case scenario environment. Id simply like to see more testing in games not using vulkan, without limiting fps and without using free/gsync panels.

I was referring to using Vulkan as an objective test with metrics in general. Yes , this "blind test" is subjective and useless for the most part. The only thing I take from it is that AMD might have something to compete with 1080ti since I do no believe this is the same card they used previously for the 1080 blind test.

 

[EarthDog]

Not sure how you can take away the rx vega competes with it from these testing parameters... but ok.

Good conversation... restores a sliver of faith in tpu conversations can be had where the parties disagree and not degrade to insults.

 

[Vya Domus]

Not sure how you can take away the rx vega competes with it from these testing parameters... but ok.

Not from the performance , but in the video the 300$ price difference was mentioned again. Since the system with the 1080ti obviously costs more than the one with the 1080 but the price difference remains the same one can assume logically that the AMD card in this test also costs more and it is a different one most likely faster than the one they put against the 1080 previously. It's just an educated guess , I don't know how it would work otherwise.

Good conversation... restores a sliver of faith in tpu conversations can be had where the parties disagree and not degrade to insults.

Well , this isn't wccftech after all , so no cancer.

 

[TheoneandonlyMrK]

Vulkan moves most graphics optimizations to software developers. D3D11 and down as well as OGL put the graphics optimizations in hardware developer hands. NVIDIA was better at driver optimizations than AMD was. Or rather, I think NVIDIA is more willing to cut corners for higher frames than AMD is.

Price versus TFLOPs, AMD usually comes out on top. In gaming, AMD GPUs have more idle hardware. Even gimped at gaming as Vega is, it has higher theoretical TFLOPs than GTX 1080 Ti and Titan Xp.

As I said, NVIDIA cuts corners. Really crappy FP64 and FP16 performance is evidence of this. NVIDIA makes more GPU SKUs to force higher budget customers to higher budget hardware. AMD likes to give everyone access to everything so that, for example, games change significantly over to FP16, Vega will see a doubling of performance while Pascal will be better off using FP32 not benefiting at all.

Edit: if memory serves, D3D12_1 full compliance requires FP10 or FP16 support. Pascal is screwed on that one because they don't want AI customers buying up GeForce cards. Edit: Yup.

You have it as i see it except for the principal of cutting corners , imho nvidia didn't cut any corners , they juat made their last three generations of consumer cards pretty much pure 32bit power houses using an old but efficient shader design then optimising the crap out of it and node swapping it, it wasn't lazy or corner cutting it was sound technical design.
They can do this because they soley made gpus and had massive rnd budgets , and also because they had the resources to make three to four different skus per generation with a top dog compute card for double precision work and some outlier socs to break new markets.
Amds agenda and focus are both different and in scope larger so they don't have the resources to make four gpu, skus to compete and the two they make have to address the full market Nvidia also sells too.
Ie compromises abound for both but less so for nvidia atm.

 

[P4-630]

Asus Vega XT64

14.099 (air-cooled) en 15.999 (water-cooled) South-African Rand, about 923 en 1047 EUROS.

https://nl.hardware.info/nieuws/52912/prijzen-en-persfotos-van-amd-radeon-rx-vega-64-duiken-op

 

[r9]

They say blind test I hear we failed.
I could understand blind test only if Vega was suppose to be half the price of 1080ti.
Look its half the price its slower but you can't tell the difference.
And if they compare it to overclocked 1070 provable people could not tell the difference either.
Consumer advice if you want Vega performance for half the price buy 1070.

 

[Vya Domus]

Asus Vega XT64

14.099 (air-cooled) en 15.999 (water-cooled) South-African Rand, about 923 en 1047 EUROS.

View attachment 90587

https://nl.hardware.info/nieuws/52912/prijzen-en-persfotos-van-amd-radeon-rx-vega-64-duiken-op

Titan X was listed at 2000 euros when it first made it's way in my country. I wouldn't hold my breath that these prices are the official MSRP.

 

[JalleR]

my guess is it is priced a Little under the 1080ti that is why this blind test S.... is using a 1080ti, look it is as grate as.......

 

[trog100]

if most people cant tell the difference it could be argued that the difference dosnt matter.. only maybe to a bunch of anally retentive argumentative brand name loyal geeks.. he he

trog

 

[BiggieShady]

if most people cant tell the difference it could be argued that the difference dosnt matter.. only maybe to a bunch of anally retentive argumentative brand name loyal geeks.. he he

trog

How shall I put it ... I know, I'll use car analogy ... this test is an equivalent of testing acceleration of cars by throwing them off the plane

 

[Vayra86]

Blind testing is now bad...... Mmmmkay.

Or perhaps its your own blind eyes.

Its bad in this context yes, because it feels like a super weak excuse to sell a horribly overexpensive GPU with horrible performance per dollar, on the basis of 'it feels smooth'.

Come on.

@BiggieShady best car analogy 2017

 

[TheoneandonlyMrK]

Its bad in this context yes, because it feels like a super weak excuse to sell a horribly overexpensive GPU with horrible performance per dollar, on the basis of 'it feels smooth'.

Come on.

@BiggieShady best car analogy 2017

Nah hed have to put a passenger in each then gauge their experience