RX Vega Owners Thread, Tests, Mods, BIOS & Tweaks !

What do you guys think of the Vega FE edition for 700? Its stupid but i would love to have one just for the 16GB of HBM2 lol

 

I'd like that also. 16GB HBM2
It's a steal, pick it and don't look back (8k Ready )

 

And since 17.11.2, it's seems that "gaming" drivers support it without mod needed (I've a Vega FE watercooled here, I had a very good deal on it, so, 16gb of vram here we go. I was short with my Fury X 4gb, which is a shame because the power is still there. So, with 16gb, I'm set for a while )

 

Finally got AB on, so i can see my clocks and that, and ran the Last Light bench, core was anywhere from 1570 odd to 1600+, and my temp was about 65, with a 52% fan speed, then played Prey for a bit, and my core was anywhere from 1660 to 1700+, and was doing over 100fps, lowest i saw was about 92, at 1440p, maxed out.

 

Nice.
I have TUL rx 64, so far laying on bed, but already took ref cooler and morpheus 2 is on it.
Done some benches thou, temps are lower by 30-40 degree...Had problem with hotspot temp but finally managed to have delta 15+ only. Most games I saw 100-250 FPS at 1080p...
But, as said, laying on bed, meant to be in my son's PC. My current with 550W power is rather too low...

 

Removed, for inconsitency...

 

Thats better

 

Still VEGA is the Best

1st. 499USD


2nd. 2999USD

 

Vega is XMR beast, and Volta is ETH beast. Times they are changing. Now we can say "Nvidia the way it's meant to be mined" Maybe we can get some cheap Vega cards for gamers next year

 

FP16 in Titan V Vs Radeon VEGA RAPID PACKED MATH

 

Vega AiO LC -> 3DMark FireStrike Ultra 4K benchmark Graphics score : 5841 for 180W

 

Did anyone get their hands on the Sapphire Nitro+ Vega64? Any chance of uploading the BIOS ?
What is the lowest UV anyone got on their 64? I am currently running -30% Power, 1337Mhz Core 835mV and 1100Mhz HBM2 with 835mV. No errors in CastXMR

 

The question is if any of this will ever translate to gaming performance.

 

It does pretty well in gaming even if some of the features that were revealed with the GPU aren't really doing much though the framerate can differ heavily depending on the game and DX11 as usual is trailing behind a bit compared to Nvidia.

It does seem capable of beating the AMD Fury with up to 40% faster performance when it really works but in other titles it's doing a fair bit worse though many of these benchmarks were with the launch day GPU's and drivers so this might have been improved since.
Definitively has potential to scale though unlike the Fury which is kinda bottlenecked now but we shall see I suppose, will also be interesting to see what AMD will release next and if that's some sort of 600 series Polaris follow up or a different Vega GPU or even that rumored Vega die-shrink and refresh but that might be less likely to happen, shortage isn't really going anywhere soon either and pricing is well it's terrible world-wide it would seem with these third party Vega models pushing it even further off balance.
(Still selling out though whenever products are available, maybe AMD can get the quantity up a bit for production but it's probably pretty complex.)

EDIT: I'm nowhere near a expert or anything on these subject matters though but the availability and pricing is evident as is how desirable these are for mining due to effectiveness for certain crypto currencies which are currently increasing in value so with the GPU able to pay for itself in a few months time even with the current price levels it's not too surprising it's selling out.

 

Being 40% faster than the Fury isn't a great feat, exactly, especially considering that the LC versions are 70% higher clocked.

Up to now actually it seems slower than Fiji per clock, hence my question about if all the tickboxes it has will ever translate to actual gaming performance.

 

That's the issue with GCN essentially, good shader performance but Vega still has a similar configuration with ROP's and how the shader engines are set up and as such it retains the similar issues (Since the 7000 series or so.) with scaling past a certain point so despite being up to 60% higher clocked on core (Some 1600Mhz compared to 1000 for Fury.) and almost double the HBM clock speed (1000Mhz compared to 500Mhz though it's 2048 bus width instead of 4096) then as has been evident in most benchmarks it doesn't translate into the same performance in-game since it's still somewhat held back for geometry performance although it can do some discarding of unused data like what the Polaris series can unlike the Fury.

Still trying to read up on and understand more on how this works, it's incredibly technical but in short GCN has limits to how it's configured and how well it scales and driving it further ramps up voltage requirements for diminishing real performance returns though AMD still pushed Vega pretty far even on it's stock configuration and then it's throttled through thermal and power limitations.

It gets even more complicated trying to understand these additional features, how they work and how they were meant to work but in effect HBCC is possible but primitive shaders and the other form of discarding unused geometry data is not doing too well though Adrenline and updates on AMD's open source utilities seem to point to them working on this although it's mostly rumors and speculation and whatever gains from this might be underwhelming compared to what initial hype suggested for these features.

The Vega architecture.
https://www.anandtech.com/show/11002/the-amd-vega-gpu-architecture-teaser/2

Discussions around Vega compared to Polaris.
https://www.reddit.com/r/Amd/comments/7l8ee5/is_vega_a_more_efficient_architecture_than_polaris/

The ongoing discussions around the various features Vega has but which currently aren't all there yet.
https://www.reddit.com/r/Amd/comments/7jfrta/request_official_statement_about_dsbr_primitive/

Data from the AMD white paper.
https://www.reddit.com/r/Amd/comments/6twjc7/some_insights_from_the_vega_whitepaper/

Comparisons from someone who upgraded from a Fury Nitro.
https://www.reddit.com/r/Amd/comments/6x6u1d/upgrading_an_r9_fury_nitro_to_an_rx_vega_56/


It's going to be interesting to see what AMD can do in newer drivers and perhaps we'll see these features used to some extent in AMD supported games if nothing else but yeah it exceeds thanks to it's high clock speeds but hits limits on how far it can scale while having similar bottleneck areas like earlier GCN cards but it's not in as poor of a position as the Fiji cards find themselves in currently.

Reading up on the architecture and what's being posted plus the various benchmarks and reviews of the GPU is interesting but I guess only AMD can give the real answer and not all details on Vega are available.

This could be of interest too from the vertex issue with the current 17.12.1 and 17.12.2 drivers.
https://www.reddit.com/r/Amd/comments/7li6ph/vertex_processing_broken_ngg_fastpath_primitve/

If it does end up being related to the primite shaders and the DSBR features but again not too many details available on these features for now from what I can find at least, AMD has been fairly silent since Vega's launch but it would be nice if this ended up working out and gave the cards a bit of a boost to set it apart from the other AMD cards in more than just clock speed.

Time will tell I guess, sooner is better but it could take time although it'd make for a good start for the 18.1.1 drivers next year.


EDIT: Not the best explanation probably, even if the cards end up gaining some good performance from this that also doesn't resolve the supply issues and way above recommended price levels for these cards and might end up pushing demand even higher. Perhaps that too could improve next year though, but it's a complicated GPU so who knows how many chips they can get per batch and turn into retail sold Vega's plus other sources also wanting these chips such as Apple.

 

It is faster than fiji per clock on avg, but there is a problem somewhere on the chip, at the same clocks as fiji its as faster or faster than fiji, but with increased clock speeds scaling degrades, this can be eliviated by overclocking the memory(which is tied to the SOC/fabric speed i believe), but its not enough , this would suggest some bandwidth saving/ improving feature is not fully functional,

If i were to guess what the problem is is that the hbcc or memory compression tech doesn't work as expected,

Could also be that the infinity fabric doesn't clock as high as expected or the internal timings are too loose.

either way its choking on something.

 

If Fiji was build to use up to 350GB/s of bandwidth (Huddy's words 2 years ago) therefore Vega can easily find bottleneck in 350+40%= of wait.. 490GB/s Who would have thought.

 

Yeah Fiji has something like a theoretical 512 GB/s bandwidth from it's hardware but in testing it's ~350 GB/s due to I think it's the clock speeds? Whereas the Vega delivers around 490 GB/s
https://www.reddit.com/r/Amd/comments/7b7gfb/disappointed_in_the_lack_of_fiji_optimizations/
http://techreport.com/review/28513/amd-radeon-r9-fury-x-graphics-card-reviewed/4