• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

The RX Vega 64 Owners Thread

Last edited:
old gamer said:
theres the alphacool eiswolf vega which i was quite interested in but someone in the 56 thread was not too impressed and a couple of people have modded the nzxt g10 which i think im going to give a go if i can find a second hand one

https://forums.overclockers.co.uk/threads/rx-vega-with-kraken-g10-aio-mod.18792093/
Click to expand...

The Eiswolf is good with some caveats - you should cut out some areas for the fan cable and UEFI switches. Seating of the back-plate is critical.

Additionally the connectors need to be handled with care/hidden/positioned well.

For the price and the ability to pop it in and out of the loop, it's good value.

Oh and it's also a total PITA putting on the thermal pads, they really should be pre-cut.
 
Davedree said:
Yeah looking back as long as people buying the reference cooler are aware that you've still had to tweak the hbm2 clock, power limit and voltages for Vega then I don't think there's a problem.
Click to expand...

Totally agree :) .

Davedree said:
On the other side though the 1080 was at a similar advantage, (like you say) from the water cooling helping it boost higher, just without memory oc. In essence both gcards have more to untap, but the Vega I don't doubt has to more to tweak and more to unleash at the cost of heat/power draw but heyho, even from your current 1610-/+ clock it'd be nice to plot frequency gain vs fps scaling say if you're lucky to get 1750-1800 mhz. Then we could put the difference in ref vs custom cooling in perspective of value.
Hwbot shows a greater range of 1080's at firestrike 1080p scaling from 23k- 25k+. But regardless I have just stumbled onto your vega bios thread and that's going to keep me reading for a while :)
Click to expand...

I agree I did not push the GTX 1080, so it did have some left in the tank. Currently I didn't believe I am pushing the VEGA fully either. I reckon I have done mild tweaks so far, biggest change is probably HBM clock/PowerLimit.

I will have a nose on HWBot. HWBot allows tessellation tweak, which improves AMD cards results a lot, no idea about nVidia as I have very low experience/knowledge of them. The other think I once stumbled on was on Fiji if I pushed HBM to a unstable level it would remove say the smoke in Graphics test 2, this would vastly bump FPS/score. On OCN an avid HWBot user with Fiji also noted this, some of the subs "we" both agreed were at the level they were because of this. So I would take some results with a pinch of salt.

Davedree said:
Great info there thanks, I've followed the stilt for years in his examination of Kaveri and then Godaveri/ Carrizo etc. How different or advanced is Vega over Polaris in the registers and powerplay tables?
For example Polaris looks like this but with Avfs/ temperature, current load etc the voltage and frequency tables obviously ride as close to the p-state.
Here's two examples of my rx470, 0% power limit and 15% power limit at auto voltage and auto clocks,
38tU5ue.jpg
hfqlEES.jpg
Click to expand...

Registers dumps is so small that it reveals little :( . Even though Polaris had AVFS the PowerPlay table revision I believe was older version than VEGA (will have to check). It had pointers to create manually set VID, VEGA has none, the VDDC_Lookuptable differs.

On another note a run my tweaked VEGA setup in TimeSpy last night, link. I believe AMD have shifted to more focus on DX12 driver improvement, combined test does not lag behind as in 3DM FS. Over my period of Fiji ownership 3DM FS so very minimal gains driver to driver (pretty much static), TimeSpy did see improvement, I'll have to fish out results.

However my tweaked setup failed playing SWBF, crashed with <5min or so :( . It had last well for several runs of SP 4K/3DM tests and several hours of f@h. I had to bump DPM 7 from 1100 to 1112mV. Then no issues in ~1hrs usage.

This profile is showing signs that due to the increase of VID for DPM 7 GPU is increasing GPU clock max even though I did not change DPM6/7 MHz.

Tweak 2 (the one that failed SWBF).

DPM6: 1557MHz 1000mV
DPM7: 1652MHz 1100mV
HBM: 1100MHz 1050mV
PowerLimit: 65%

ruZVrq8.jpg

So max 1100mV / 1708MHz, average 1070mV/1657MHz.

Tweak 3

DPM6: 1557MHz 1000mV
DPM7: 1652MHz 1112mV
HBM: 1100MHz 1000mV
PowerLimit: 65%

I lowered HBM mV to see if I would gain lowering of VDDC on lower end of "scale", it had no effect.

vOF8IGE.jpg

So max now 1113mV / 1716MHz, average 1087mV / 1668MHz.
 
AMDMatt said:
HBM Overclocking up to 1200Mhz is now possible. Tested and i see a performance boost up to 1175Mhz. 1200Mhz crashed instantly.

Safe to say Vega overclocks pretty damn well and you can increase performance nicely with tuning.

2qv.png
Click to expand...

Older driver limited SOC clock to 1107MHz, SOC clock acts as a limit for HBM clock. I on OCN posted a SOC Clock mod, link. I had spoken to Martin Malik via PM to expose SOC Clocks so when we "played" with it to see what would happen, he posted an update ASAP, link. As I had no card at the time, others tried it and gained higher HBM on drivers prior to v17.10.2, link.

After driver v17.10.2 if HBM was pushed past 1107MHz then SOC automatically went to ~1200MHz, so the mod wasn't needed.

As W1zzard follows that thread on OCN, he private messaged me, future release of GPU-Z will also show SOC Clock.
 
Nemesis.ie said:
Sweet, thanks for pushing this and I hope/expect the AMD driver team are also monitoring this and tweaking stuff in the driver all the time too. :)
Click to expand...

If I was honest my only part is "sharing my meddling" and at the time more of a theory than having tried it. It was Nuke33 and others that tested/shared results. Which were invaluable for "our" understanding.

AMD driver team/developers on VEGA must have noticed this aspect within their labs or be aware anyhow. A boon for us that they improved driver, hopefully we see more soon :) .

When we view the VEGA whitepaper we see SOC is an integral part of VEGA, link. So lifting the clock probably has other effects as well, which we may not yet be aware or know how to test.

Also there is what looks like a GPU clock hidden/sandwiched between tables in PowerPlay, link. This I must at some point test, people may already be testing it IDK.

The Linux driver being open source has helped vastly with understanding of VBIOS. When in 2015 I and others on Guru3D looked at Hawaii we really spent hours/days marking tables manually, trying to suss things. Near the end we cottoned on that Linux was our friend ;) . Due to this fact Tonga/Fiji/Polaris fell into place a lot easier and even VEGA for PP mods.
 
TrixX said:
Not too bad, under 3K RPM, kinda loud air noise above that. Not a lot of fan noise, mostly air which is much better than screaming tech. Still it's a stock blower and a Morpheus II with dual 140's on it is practically silent and much more efficient.
Click to expand...

I think undervolting is also good for noise?
 
TrixX said:
Not too bad, under 3K RPM, kinda loud air noise above that. Not a lot of fan noise, mostly air which is much better than screaming tech. Still it's a stock blower and a Morpheus II with dual 140's on it is practically silent and much more efficient.
Click to expand...

Thanks you both for your replies

I just seen a review on the Morpheus II which show good results.

Only problem I will lose the warranty on the card, may look at the reference water cooling options as well if I can find a good deal. Seen.

Any problems with the Vega reference water cooling cards?

JediFragger said:
I think undervolting is also good for noise?
Click to expand...

That is a good tip JediFragger, I could live with around 2K RPM fan speeds.
 
Last edited:
Yesterday I went to HBM 1150MHz. What I noted was I got repeatable small gains in SuperPosition 4K preset, 3DM FS and large memory copy gains in AIDA64 GPGPU, but I lost FPS/points in 3DM TS :o .

Example of SP 4K left 1150MHz, right 1100MHz.

Example of 3DM FS left 1150MHz, right 1100MHz.

Example of AIDA64 GPGPU left 1150MHz, right 1100MHz.

Example of 3DM TS 1150MHz vs 1100MHz.

I noted no artifacting, etc. I had also ran f@h on GPU for ~3hrs on HBM 1150MHz without errors.

I also tried SOC Clock mod of 1200MHz even though not need on v17.10.3 drivers and still TimeSpy with HBM 1150MHz tanks. I can only assume the higher speed HBM is having errors and those errors create performance loss in 3DM TS test.

Has anyone also noted this issue or had it?
 
gupsterg said:
Yesterday I went to HBM 1150MHz. What I noted was I got repeatable small gains in SuperPosition 4K preset, 3DM FS and large memory copy gains in AIDA64 GPGPU, but I lost FPS/points in 3DM TS :o .

Example of SP 4K left 1150MHz, right 1100MHz.

Example of 3DM FS left 1150MHz, right 1100MHz.

Example of AIDA64 GPGPU left 1150MHz, right 1100MHz.

Example of 3DM TS 1150MHz vs 1100MHz.

I noted no artifacting, etc. I had also ran f@h on GPU for ~3hrs on HBM 1150MHz without errors.

I also tried SOC Clock mod of 1200MHz even though not need on v17.10.3 drivers and still TimeSpy with HBM 1150MHz tanks. I can only assume the higher speed HBM is having errors and those errors create performance loss in 3DM TS test.

Has anyone also noted this issue or had it?
Click to expand...
That looks similar to the error correction on my Rx470 if I push the memory over 1850 on the stock strap. I can also see error fails in the memory section of hwinfo. No stutters or corruption but I actually lose fps, although I am also governed by a strict power regulation, unless I fool it by using current scale in wattool.
 
You must log in or register to reply here.
Back
Top Bottom