Poll: ** The AMD VEGA Thread **

[nashathedog]

While it should be over-rode by the nVidia drivers anyhow the best thing to do is turn off any V-Sync option ingame - you want the game to be able to run as fast as possible providing new frames as quickly as possible as the driver then selects the one that is closest to the V-Sync window to display reducing latency.

Somewhat counter-intuitively higher levels of max pre-rendered frames can actually produce better results here.

Thankyou.

 

[TheRealDeal]

They should be, It's selling faster than they can make replacements.

Have any reviewers done this type of review with a 1080ti and 980ti?

Wouldn't work as the 1080ti has a lot more Cuda cores. There was a comparison of the 1080 v 980ti at similar clocks but the 980ti has more cuda cores and was slightly faster as you would expect. You can bet the IPC is pretty much the same between Maxwell and Pascal.

Nvidia were able to boost the clocks up that much along with adding more cores to the top end that there was no need to change much as can be seen from the performance gains. I am not sure what AMD have added to the Fury X but there is 80% more transistors that appear to be doing nothing in a gaming sense at this moment.

 

[Nablaoperator]

They should be, It's selling faster than they can make replacements.

Have any reviewers done this type of review with a 1080ti and 980ti?

from whatever i have researched the 980ti and 1080 have similar IPC.. i had read somewhere that bumping 980ti to 1080 levels resulted in comparable performance. now its also known that 1080ti is much faster [double digits actually] when compared to 1080 at lower clocks so IPC of 1080ti will be much higher than a 980ti.

something like x > y, y = z... implies x > z line of reasoning

 

[WeSbO]

Thats bcoz the furyx doesnt have access to semi-pro drivers like the vega.. a possible explanation

Yep that's what I was hinting at , could non "semi-pro" drivers really make that much difference ? It would've been so interesting if we could have seen Fiji perfs with "semi-pro" drivers
We will know the gaming perf for sure soon. Considering they mentioned that the new NCU allowed for Higher clocks and Higher IPC when compared to CU

 

[harney]

This just isn't adding up. Fury X Release Date: June 24, 2015

More than 2 years later they are releasing a HIGH-END card that is barely faster than the Fury X? Something just isn't adding up here.

Maybe I'm just in denial.

Agree its not adding up...

Looking at the SPECviewperf results its killing the furyx clock for clock ..What i would like to know is this down to drivers or hardware or is it a mix of both ??

Can we not take these results gained in the pro stuff to give us an idea of what then gaming performance may be or is the card only good for compute stuff..

Surly some of them pro benchmarks like 3ds maya ect is still graphical textures meshes if running real time ect or is this just for rendering ...I am not expert of all this but surly it should be doing better in games stock for stock over the furyx

Maybe The vega FE is really still using fury x drivers for games and newer drivers for the prostuff ...I am confused tbh

 

[Nablaoperator]

Yep that's what I was hinting at , could non "semi-pro" drivers really make that much difference ? It would've been so interesting if we could have seen Fiji perfs with "semi-pro" drivers
We will know the gaming perf for sure soon. Considering they mentioned that the new NCU allowed for Higher clocks and Higher IPC when compared to CU

I am just waiting for AMD to announce the transistor count.. i just went through the 14LPP factsheet, talks abt 8.5m gates per mm2 atleast... thats like 4 billion gates atleast .. which should possibly translate to 12-16b transistors.. can someone from the semi-industry confirm this? am basically a chip size whore.. thats all

 

[stulid]

Have any reviewers done this type of review with a 1080ti and 980ti?

Good question and all i can find is -
https://www.youtube.com/watch?v=nDaekpMBYUA

which to sum up the people that cant sit through this is - the 1080 and 980ti perform the same when the tflops of each card is the same.

Edit. which is actually right? as if tflops was the same then the cards would both render at the same speed (frame rate)

 

[TheRealDeal]

Good question and all i can find is -
https://www.youtube.com/watch?v=nDaekpMBYUA

which to sum up the people that cant sit through this is - the 1080 and 980ti perform the same when the tflops of each card is the same.

Edit. which is actually right? as if tflops was the same then the cards would both render at the same speed (frame rate)

It stands to reason because there was no big major changes from Maxwell to Pascal in the architecture. Vega on the other hand is supposed to be almost brand new. It's the same Spec as Fury X in almost every way but with better this, new that and with this added in so you would think in a straight test at the same clocks all or at least one of these improvements would show up. It's not happening with this Frontier Edition so I guess we will find out when the gaming cards release.

 

[harney]

 

[AllBodies]

Doesn't this basically prove the new features aren't turned on for gaming?

The IPC is basically identical, especially accounting for the memory bandwidth difference.

One thing I thought was telling was the fact the FuryX's lead increased with resolution. One of the big Vega features is the enhanced discarding of overlapping triangles, to the extent Vega has 2.6x the triangle throughput in best-case scenarios. And the amount of triangles obviously increases with resolution, therefore Vega's lead should increase with resolution not the FuryX's. The improvement in triangle-culling should outweigh the small memory bandwidth deficit (especially since the FuryX couldn't use the full 512 GB/s anyway, so their effective bandwidth is probably near-identical).

 

[stulid]

It stands to reason because there was no big major changes from Maxwell to Pascal in the architecture. Vega on the other hand is supposed to be almost brand new. It's the same Spec as Fury X in almost every way but with better this, new that and with this added in so you would think in a straight test at the same clocks all or at least one of these improvements would show up. It's not happening with this Frontier Edition so I guess we will find out when the gaming cards release.

Using a car as an analogy,

Say you are testing a car down a road which we will call "the game"

And had two engines "the gfx cards"

Both engines had the same power/torque per rpm, but one was a little 1.0 3pot with Turbo and the other was a naturally aspirated 4pot 1.6

You would expect the car (the game) to accelerate and hit a max mph down the road identically.


That video I posted is junk? lol

 

[Doobedoo]

Has anyone tested the Vega FE with Ryzen yet? Was wondering if the early stages of Infinity Fabric had any positive effects yet.

 

[Rroff]

Doesn't this basically prove the new features aren't turned on for gaming?

Increasingly it seems that many of the new features need games to actually program support for them at some degree - some just with a "simple" patch others needing wholesale adjustments and/or alternative render paths within their core architecture.

If this is the case then I can't see it being anything other than a big fat fail as even if we assume adoption by the time its commonly used in games Vega will be obsolete.

 

[TheRealDeal]

Using a car as an analogy,

Say you are testing a car down a road which we will call "the game"

And had two engines "the gfx cards"

Both engines had the same power/torque per rpm, but one was a little 1.0 3pot with Turbo and the other was a naturally aspirated 4pot 1.6

You would expect the car (the game) to accelerate and hit a max mph down the road identically.


That video I posted is junk? lol

I agree with the video. Maxwell and Pascal are basically very similar so if you get similar specced cards like the 1080 and 980ti and run them at similar clocks then they should perform pretty similarly. The gtx980ti has slightly more Cuda's so is slightly faster at the same clocks.

 

[LoadsaMoney]

I remember ages ago, when those with Furys were on about upgrading to Vega, i said be a waste of time, as you'll just be sidegrading for the extra memory, as it'll just be a Fury with 8GB instead of 4GB, god if that joke comes true

 

[cainer]

so inbetween a 1080 and 1080ti, wonder how well it'l over clock

 

[Zeebedy]

so inbetween a 1080 and 1080ti, wonder how well it'l over clock

Sounds good a. Im just praying for a good, friendly price. one can hope...

 

[r7slayer]

Ohh wow, so if that "leaked benchmark" is real. It's pretty much same speed as a 1080. So best is between a 1080 and noticeably behind a 1080ti. Problem is most 1080 will OC some what so will probably leave the RX vega behind. Also the 1080Ti OC some what too. This looks terrible. WTF have AMD done? Where are all these improvements with the time they have spent. There is literally zero IPC improvements it's looking like its literally just a die shrunk FuryX with 8Gb of HBM2. Disappointing if true. So glad i got a 1080ti now.

Price is the only thing that can help AMD now. "Affordable 4k" is what Raja said.

 

[Zeebedy]

How will pre-orders work on this? if they are launching end of the month/beginning of august, i wanna get my foot in before these ******* miners take all the stock!!

 

[TheRealDeal]

Ohh wow, so if that "leaked benchmark" is real. It's pretty much same speed as a 1080. So best is between a 1080 and noticeably behind a 1080ti. Problem is most 1080 will OC some what so will probably leave the RX vega behind. Also the 1080Ti OC some what too. This looks terrible. WTF have AMD done? Where are all these improvements with the time they have spent. There is literally zero IPC improvements it's looking like its literally just a die shrunk FuryX with 8Gb of HBM2. Disappointing if true. So glad i got a 1080ti now.

Price is the only thing that can help AMD now. "Affordable 4k" is what Raja said.

The leak shows Vega in between 1080 stock and 1080 ti stock.

3dmark 11 Graphics scores.

1080 = 30030

Rx Vega = 31873

1080ti = 33961

They do have OC scores for the 1080/ti but we don't know if Vega can oc further as well. I wouldn't read much in to this either tbh as it's a pretty old bench and you can see that the gap between 1080 and ti is only around 13%