AMD Vega Hardware Reviews

[Grall]

Reason being, ASUS shipped card with stock AMD Vega 64 BIOS instead of custom one.

*ahem*
Hopefully that doesn't mean they'll have to recall all stock already manufactured and shipped... Because that'd just be urrrggh, lol.

Interesting to note, the lower power consumption despite higher performance than reference vega64 using stock BIOS. No doubt the vastly more performant cooler is the cause of that.

Btw, just measured my existing card. It's about 27cm, or thereabouts. The strix vega is nearly 30, it's not gonna friggin fit in my computer case! BALH!

What to do what to do... *sigh*

*heads on over to Lian-Li's website*

 

[patienceofasaint]

AMD marketing feeling the "heat" and attempting to cast doubt on reviews?

I've been lurking here for many years. I can't recall you ever spreading rumors about nVidia. And as rumors go this one has the shakiest of foundations. Hilbert's testing methods were questioned and he traced the IP address of that forum poster "awfully close to AMD Markham HQ". Either the IP address is in AMD or it's not. It can't be close. And how does Hilbert know what IP addresses AMD are using? It's a nonsense deflection.

 

[Lightman]

*ahem*
Hopefully that doesn't mean they'll have to recall all stock already manufactured and shipped... Because that'd just be urrrggh, lol.

Interesting to note, the lower power consumption despite higher performance than reference vega64 using stock BIOS. No doubt the vastly more performant cooler is the cause of that.

Btw, just measured my existing card. It's about 27cm, or thereabouts. The strix vega is nearly 30, it's not gonna friggin fit in my computer case! BALH!

What to do what to do... *sigh*

*heads on over to Lian-Li's website*

Last time I had problem like this with my old case, I went to pick up my angle grinder and solved the issue with some sparks flying!

BTW my Vega 56 arrived and in my limited time I managed to find one test which is consistently slower on it than on old Fury Nitro!

Superposition Benchmark is returning following scores on my Ryzen 7 1700 4GHz and 3200 CL14-14-14 tight subtimings setup:
Fury Nitro Stock (1050 core) using custom Extreme settings at 2560x1440 (shaders Extreme, Textures High, DoF and Motion Blur ON) - 1914 points (min. 11.59, avg. 14.32, max. 16.40)
Vega RX 56 Stock (average 1400Mhz GPU) using custom Extreme settings at 2560x1440 (shaders Extreme, Textures High, DoF and Motion Blur ON) - 1844 points (min. 11.74, avg. 13.80, max. 15.80)

3DMarks and all 3 games I've tested were faster on Vega.
Interesting point to note was that in CS:GO I play with QHD res. and 144FPS limit to match my screen refresh, Vega is much better at saving power as it downcloks GPU to around 500-800MHz and HBM to 167-500MHz. Frequencies vary during a game, but power is really low compared to Fury! Fury would always stay at 1050MHz in this game with HBM fixed at 500MHz.

Finally my quick ETH mining test.
This is very first try at downcloking this card and things are not working as I would expect them with my card. I can't fix low GPU clocks nor can I change lower pState voltages, so the only thing I did was move Power Target slider.

Fury Nitro 1050/500 Power -41% vGPU 1125 (1250 stock) Hash - 29.2Mhs 277W
VegaRX 56 1100/800 (drops 700) Power -26% vGPU stock Hash - 29.5Mhs 273W


With some quick tweaking and increased HBM clocks to 900MHz (but they fluctuate if card is too close to power limit and GPU steals HBM clocks):
VegaRX 56 1230/900 (drops 500) Power -25% vGPU stock Hash - 35.2Mhs 334W

Power at the wall with my ASUS MG279Q monitor. Idle 113W Vega and 115W Fury.

 

[entity279]

*heads on over to Lian-Li's website*

Nevermind Vega then, you know you want that locomotive !

 

[Deleted member 13524]

With some quick tweaking and increased HBM clocks to 900MHz (but they fluctuate if card is too close to power limit and GPU steals HBM clocks):
VegaRX 56 1230/900 (drops 500) Power -25% vGPU stock Hash - 35.2Mhs 334W

As pointed out before, Vega 56 is power-limited. Also, for ETH the compute performance is way above the necessity for its memory bandwidth, so you can just keep downclocking and downvolting the core to 950MHz and sub-1000mV just to keep the memory clocks at higher values.

I've been lurking here for many years. I can't recall you ever spreading FUD about nVidia. And as rumors go this one has the shakiest of foundations. Hilbert's testing methods were questioned and he traced the IP address of that forum poster "awfully close to AMD Markham HQ". Either the IP address is in AMD or it's not. It can't be close. And how does Hilbert know what IP addresses AMD are using? It's a nonsense deflection.

FTFY.

Hilbert made a deflection to defend himself from what seems to me like a review with unexpectedly very odd results, further proved by the AIB who asked them to bring the review down because it was shipped with the wrong bios.
In the very next post that user explains he's from the Greater Toronto Area which has 6.5M habitants or 20% of Canada's whole population, meaning an IP with some of the most significant numbers being close to Markham means jack.

 

[pharma]

I've been lurking here for many years. I can't recall you ever spreading rumors about nVidia. And as rumors go this one has the shakiest of foundations. Hilbert's testing methods were questioned and he traced the IP address of that forum poster "awfully close to AMD Markham HQ". Either the IP address is in AMD or it's not. It can't be close. And how does Hilbert know what IP addresses AMD are using? It's a nonsense deflection.

Maybe that's because no GPU review thread ever suggested a "negative" post from a first-time poster was "awfully close to AMD Markham HQ". Hilbert is known for tracing IP addresses of banned people who repost so think he has some experience. If your intent is trolling to create an AMD vs nVidia scenario I suggest you keep lurking ....

 

[Grall]

Nevermind Vega then, you know you want that locomotive !

Didn't find anything tasty, tbh. And their site is just ludicrously slow for me, you'd think the pipe out of their server was a bunch of carrier pigeons hauling IP packets...

Not sure what's for when it comes to gaming uATX cases these days. I've had my current case (Silverstone Fortress FT03) for so long, and it's a cool case and all, but the layout is so damn wonky. It's hard fitting premium components in it, you bang your head on limits everywhere. PSU can't be too long or you can't fit the right side panel, GPU can't be too long or it won't fit at all, or too tall either or it interferes with the cardslot fan holder. And if you mount the fan on the outside of the holder for more clearance, you can't fit the left side panel either.

Also, cable management is a bitch in this thing. *sigh* It's compact, but too small, and kind of large all at once. One advantage is that front panel is detachable, so after removing the front fan holder and side panels, the mobo is easily accessible.

The Fractal Design Define Mini C TG (damn, doesn't exactly roll off the tongue does it? lol) is looking pretty fine. Again there's the issue of whether it'll fit my existing and fairly long power supply tho, which is fucking dumb, because damn, look at all that space in the PSU compartment. *shrug* I'll check some reviews to see what they have to say.

Still, this is one of the better uATX chassis I've seen. It doesn't have a million useless 3.5" drive bays - or worse, 5.25" bays. I hate that. Still, it's a bit big for a uATX case. 5 card slots? What uATX mobo has 5 slots? *boggle* Maybe the design team was a bit heavy on the Absolut when they made this case.

 

[entity279]

Well you mentioned Lian Li and I had a crave to go check what's new on their site too. And gave up due to performance. I do like how their designs look, but they do change quite slow (e.g i like how some manufacturers increased the size of the supported fans beyond 140).

All was not lost, since still looking at cases I've discovered the existance of this : The Tower 900 Snow from Thermaltake . So pretty.. Or just pretty huge

We have quite different requirements for case; i'm searching for stuff that can accomodate E-ATX cases instead - i'll eventually be going towards one the new many cores platforms and will attempt to install some liquid cooling as well. And i'm still thinking about wether to go for a Liquid Vega (vs non-reference Vega 64). Or two, to still do some mining on the main rig.

But we seem to agree on the uselessness of drive bays. 2 - 3 2.5" drives can be installed behind the motherboard, no reason they shouldnt be nuked away in some cases. Personnally I hate the extra cables and will also attempt to go M2 - only

 

[silent_guy]

Guru3D took Asus STRIX article off after ASUS request. Reason being, ASUS shipped card with stock AMD Vega 64 BIOS instead of custom one. Funny how manufacturers can get basics so wrong.

There's a curse on AMD GPUs that makes it impossible to do a launch without drama, it seems.

 

[Grall]

There's a curse on AMD GPUs that makes it impossible to do a launch without drama, it seems.

Probably has something to do with the anticipation and expectations of no enthusiast GPU for two years as well, and then add manufacturing delays and complications due to HBM2, and so on... People are eager for news, and when there is little or perhaps none, they start making up their own.

 

[CarstenS]

With some quick tweaking and increased HBM clocks to 900MHz (but they fluctuate if card is too close to power limit and GPU steals HBM clocks):
VegaRX 56 1230/900 (drops 500) Power -25% vGPU stock Hash - 35.2Mhs 334W

Try 1000 MHz for core clock (should settle a little below 1000, i.e. a 980-990ish), -20% Power Target only and give your HBM2 the stick for 960 MHz. That should net you above or at least around 36 MH/s, while the Voltage according to newly updated GPU-z will be in the (very) low 8xx mVolt range.

 

[patienceofasaint]

If your intent is (some inflammatory stuff)...

My intention was to point out the curious patterns in your postings, and hopefully alert the other readers to what *your* intentions are.

 

[Malo]

My intention was to point out the curious patterns in your postings, and hopefully alert the other readers to what *your* intentions are.

pharma tends to collect information from around the traps and post them here as informative. Though some of the sources I don't agree with, I don't believe he/she has ever intended to be inflammatory or biased.

 

[Lightman]

Try 1000 MHz for core clock (should settle a little below 1000, i.e. a 980-990ish), -20% Power Target only and give your HBM2 the stick for 960 MHz. That should net you above or at least around 36 MH/s, while the Voltage according to newly updated GPU-z will be in the (very) low 8xx mVolt range.

Thanks for tips! I had to reinstall drivers as Radeon Wattman would not force lower clocks before. It's confusing with Vega as only last two power states can be adjusted on my card. I have to give it high enough power target to reach them and lower them to desired frequency, otherwise card will apply higher (more efficient) clock and never go to frequency I set it to.

My HBM will not do much more than 900MHz, instant black screen! BTW GPU-Z reads my HBM temperature at 85C even when GPU is only 67C! Is this normal or do I have badly installed radiator?

 

[pharma]

One positive from the Asus Rx Vega 64 review is we should have benchmark data for both old and new bios.

 

[itsmydamnation]

Maybe that's because no GPU review thread ever suggested a "negative" post from a first-time poster was "awfully close to AMD Markham HQ". Hilbert is known for tracing IP addresses of banned people who repost so think he has some experience. If your intent is trolling to create an AMD vs nVidia scenario I suggest you keep lurking ....

As someone who designs large networks all day everyday I find this quite laughable. Does he have access to all carriers LNS's , LAC's, OLT's or Radius servers? I suppose we should also value the input here of people who are experienced in reading tea leaves?

The negative connotations that came with the original post backed with such poor substance is pretty insightful and likely to provoke a response don't you think? I would have done it if someone else didn't.

 

[CarstenS]

Thanks for tips! I had to reinstall drivers as Radeon Wattman would not force lower clocks before. It's confusing with Vega as only last two power states can be adjusted on my card. I have to give it high enough power target to reach them and lower them to desired frequency, otherwise card will apply higher (more efficient) clock and never go to frequency I set it to.

My HBM will not do much more than 900MHz, instant black screen! BTW GPU-Z reads my HBM temperature at 85C even when GPU is only 67C! Is this normal or do I have badly installed radiator?

That's the same for me (not 100% identical values, but general trend). Note though that Ethereum mining stresses memory more than regular gaming and causes higher temps as well. For mining, you might want to manually increase fan speed a bit, so your HBM2-temp stays below 85 °C at which point internal throttling begins, which you can verify by looking at diminishing hashrates while mining.

What I did for my V56 was to set clocks on highest DPM-state to 1132, apply, lower power target to -20, apply, maximize HBM2 clocks (for me 960), apply. That nets 1008 MHz GPU clock, 960 MHz HBM2 clock and lower temps on the HBM2 since the chip is not so hot anymore. According to GPU-z 2.40, GPU temp is at 70 °C, Hot Spot at 73 °C, HBM2 at 81 °C, GPU-only power draw at 96 watts, core voltage at 0,8125v and fan at 1.700 rpm. All that at 36,3 MH/s.

 

[Lightman]

That's the same for me (not 100% identical values, but general trend). Note though that Ethereum mining stresses memory more than regular gaming and causes higher temps as well. For mining, you might want to manually increase fan speed a bit, so your HBM2-temp stays below 85 °C at which point internal throttling begins, which you can verify by looking at diminishing hashrates while mining.

What I did for my V56 was to set clocks on highest DPM-state to 1132, apply, lower power target to -20, apply, maximize HBM2 clocks (for me 960), apply. That nets 1008 MHz GPU clock, 960 MHz HBM2 clock and lower temps on the HBM2 since the chip is not so hot anymore. According to GPU-z 2.40, GPU temp is at 70 °C, Hot Spot at 73 °C, HBM2 at 81 °C, GPU-only power draw at 96 watts, core voltage at 0,8125v and fan at 1.700 rpm. All that at 36,3 MH/s.

My card will not accept vGPU lower than 0.9250V. I've tried values from 0.8 to 1.1V and everything works as expected till I get to 0.95V (reported 0.925V). Any setting below it will Apply in Wattman but GPU-Z reported voltage and power consumption at the wall will not change.

I let my PC to mine at 1029MHz GPU and 900MHz HBM for 3h (33.3MHs) and temps are stable at 64C GPU, 69C HotSpot and 73C HBM. Increasing fan speed for mining was the first thing I did after spotting 85C HBM temps. Now it's set to 2030RPM and it is quiet enough not to bother me. Oh, and GPU only power during mining is reported at 125W with settings mentioned above and 0.925V (minimum my card is willing to go down to).

So many new intriguing quirks to learn about Vega I'm very pleased! There are things like much improved power savings, especially when gaming with frame rate cap, or how much better Vega is at simultaneous gaming + mining tasks compared to previous AMD GPU's.

 

[CarstenS]

Since undervolting does not work for me, I leave the slider alone and only lower the core clock which automatically reduces voltage.

 

[entity279]

My card will not accept vGPU lower than 0.9250V. I've tried values from 0.8 to 1.1V and everything works as expected till I get to 0.95V (reported 0.925V). Any setting below it will Apply in Wattman but GPU-Z reported voltage and power consumption at the wall will not change.

For what is worth, I'm experiencing the same with rx 580. For me it's not even a limit - if I set any voltage in watman it applies it for 1 second or less, than back to default. So it may not be related to Vega. I'm definetly running into a Watman bug .