News AMD Announces Radeon VII

[Hitman928]

More info to come. . .

 

[n0x1ous]

Quake champions 3440x1440 ultra uses 14gb of vram on Radeon 7 lol. Probably just reserving it and doesn’t really need it but funny to see that number

 

[zinfamous]

Quake champions 3440x1440 ultra uses 14gb of vram on Radeon 7 lol. Probably just reserving it and doesn’t really need it but funny to see that number

Is there any software that gives an accurate reading of VRAM that is actually being used by a game or other program? All I've ever seen when running a game is that it will "use" whatever memory you allocate to it, and report it that way. Even if I set HBCC Memory segment to 12 GB on my Vega 64 (to grab any system DRAM it thinks it needs, right?) Radeon and in-game displays will report 12GB memory being used. It's kinda pointless.

 

[n0x1ous]

Is there any software that gives an accurate reading of VRAM that is actually being used by a game or other program? All I've ever seen when running a game is that it will "use" whatever memory you allocate to it, and report it that way. Even if I set HBCC Memory segment to 12 GB on my Vega 64 (to grab any system DRAM it thinks it needs, right?) Radeon and in-game displays will report 12GB memory being used. It's kinda pointless.

Ya its a bit frustrating. I don't know of any program that accurately reports it.

 

[Feld]

Ya its a bit frustrating. I don't know of any program that accurately reports it.

I imagine it's a difficult thing to quantify. It's easy enough to monitor how much memory is being consumed, but how would a utility know what fraction of that amount is actually "needed"?

 

[guskline]

Just got an Asrock Radeon VII gpu this afternoon delivered from Newegg.
Here are some Firestrike scores run on my 2700x with a GTX 1080 and now the Radeon VII and also the scores for my 5960x OC'd to 4.4 with a slightly Ocd GTX 1080TI.
I'll run more tests when I have time. The Radeon VII is a solid card, well built.

Scores for GTX 1080, Radeon VII and GTX 1080TI
Fire Strike 1.1

Cpu 2700x
GPU GTX 1080

Overall 17310
Graphics 20882
Physics 21383
Combined 6739

Cpu 2700x
GPU Radeon VII

Overall 20883
Graphics 27826
Physics 21409
Combined 7181

Cpu 5960x
GPU GTX 1080TI

Overall 23049
Graphics 29385
Physics 20277
Combined 9556

Fire Strike Extreme 1.1

Cpu 2700x
GPU GTX 1080

Overall 9682
Graphics 10077
Physics 21453
Combined 4575

Cpu 2700x
GPU Radeon VII

Overall 12209
Graphics 13496
Physics 21378
Combined 5177

Cpu 5960x
GPU GTX 1080TI

Overall 13462
Graphics 14321
Physics 20269
Combined 6893

Fire Strike Ultra 1.1

Cpu 2700x
GPU GTX 1080

Overall 5169
Graphics 5044
Physics 21420
Combined 2649

Cpu 2700x
GPU Radeon VII

Overall 6718
Graphics 6874
Physics 21308
Combined 3059

Cpu 5960x
GPU GTX 1080TI

Overall 7274
Graphics 7209
Physics 20401
Combined 3835

 

[Igo69]

I have to say a very good video card:

 

[DrMrLordX]

@guskline you are welcome to join us in the Radeon VII owner's thread in the AMD subforum if you'd like to share your experiences with the card! More data is always welcome.

Also, this is what happens when you start to overclock radeon VII:

https://www.3dmark.com/3dm/33836147

GPU settings: 1940 MHz GPU, 1.09vGPU, 1200 MHz RAM. Highest TJunct is 95C for that run.

 

[guskline]

DrMrLordX, Thank you. Will join.

 

[AtenRa]

[HardOCP] AMD Radeon VII Chilling & Undervolting

https://www.hardocp.com/article/2019/02/24/amd_radeon_vii_chilling_undervolting/1

 

[DrMrLordX]

@AtenRa no surprises there. Radeon VII undervolting is highly rewarding. Also check out what it does to minfps. Awesome.

 

[Stuka87]

The article was updated to show why temps were higher while undervolting. Turns out the fan speeds are tied to power draw. So when you drop the power usage by 20-40 watts, the fan also slows. With the card in this config, its power usage matches that of the 2080.

 

[maddie]

Just curious. Anyone ever tried undervolting + chill? Can that even work?

 

[DrMrLordX]

The article was updated to show why temps were higher while undervolting. Turns out the fan speeds are tied to power draw. So when you drop the power usage by 20-40 watts, the fan also slows. With the card in this config, its power usage matches that of the 2080.

That's weird. Undervolting lowered temps on mine using the standard fan configuration. By a lot.

Just curious. Anyone ever tried undervolting + chill? Can that even work?

Not yet, but that does sound like an interesting experiment.

 

[darkswordsman17]

While the minimums is impressive (and makes me shake my head at AMD still doing so little to tweak things as its making them look worse for no reason), The rest isn't too crazy (although I might be not understanding the overall perf/W improvement that's bringing, plus I don't know what settings they were using so they might be at higher voltages than others were getting or might be getting better than others too), although its still nice and is exactly what people have been seeing from AMD GPUs. I wonder if the APUs are similarly affected (well the big console ones seem to be, hence Microsoft putting effort towards that with their "Hovis Method"), and I think it'd be extra beneficial there (but I also guess the CPU guys do a better job of managing voltage, plus I believe they run at lower clocks as well that are probably more in the efficient range of these GPUs).

The article was updated to show why temps were higher while undervolting. Turns out the fan speeds are tied to power draw. So when you drop the power usage by 20-40 watts, the fan also slows. With the card in this config, its power usage matches that of the 2080.

How much watts do the fans use? And I assume you can do fan profiles like you can on other cards?

Just curious. Anyone ever tried undervolting + chill? Can that even work?

It should work, but I personally don't ever use Chill as it causes stuttering/hitching too much when playing as it tries to aggressively change power states, and so doesn't end up being worth it. It'd benefit people that spend a lot of time Alt+Tab out of games where the game is minimized or not the focus though, so it'd drop FPS for games that don't have a setting for that (Diablo 3 does, and I'd guess other Blizzard games have that as an option).

That's weird. Undervolting lowered temps on mine using the standard fan configuration. By a lot.



Not yet, but that does sound like an interesting experiment.

Yeah I'm not sure why temps wouldn't be lower, and wouldn't the standard fan profile be fairly aggressive so you'd think it'd be more aggressive as its expecting to deal with higher temps from the higher stock voltage and TBP, but perhaps there's some weird issue. Does H test on open bench or something where its causing a delay in the fan response to temps, or that might be causing some sensors to report cooler than stock (like around VRMs or something) and so its not kicking the fans up like it does for stock.

 

[Stuka87]

How much watts do the fans use? And I assume you can do fan profiles like you can on other cards?

Its not an issue related to how much power the fans draw. The fan speed is set by how much power the card is drawing, meaning more power equals more heat (typically).

And yes, you can easily set a fan profile, which they mention.

 

[Stuka87]

That's weird. Undervolting lowered temps on mine using the standard fan configuration. By a lot.

Interesting. I wonder if maybe there was a driver change between when they tested and what you have?

 

[DrMrLordX]

Interesting. I wonder if maybe there was a driver change between when they tested and what you have?

If they tested with the day-one driver then yes.

 

[guskline]

I just installed a Bykski waterblock on the Radeon VII and the power usage max dropped from 285W to 206W.

 

[amenx]

I just installed a Bykski waterblock on the Radeon VII and the power usage max dropped from 285W to 206W.

Awesome! How do you like it? Any quirks, observations vs your 1080ti?

 

[DrMrLordX]

I just installed a Bykski waterblock on the Radeon VII and the power usage max dropped from 285W to 206W.

That's really impressive. Radeon VII likes to be kept cool.

I might have to break down and install a loop when I get my Ryzen 3.

 

[amenx]

Its interesting that the fans can draw so much power at load. 80 watts is plenty incentive to LC the card.

 

[coercitiv]

Its interesting that the fans can draw so much power at load. 80 watts is plenty incentive to LC the card.

The fans are definitely not drawing that much power.

 

[guskline]

It's the fans and the LCD lighting. BUT those fans do crank up at full load so I think they are drawing some wattage.

BTW, I used the Wattman auto undervolt function and here are the readings in brackets to the right of the normal volt reading. No overclocking.
GPU clock 1803 Mhz (1808)
Mem Clock 1015 Mhz(1016)
UVD clock 33 Mhz
VCE clock 33Mhz
GPU Temp 42C (40)
GPU Temp(hotspot)56C(52)
Mem Temp 45C(43)
GPU VRM Temp 44C(41)
SOC VRM Temp 40C(39)
Mem1 VRM temp 41c(40)
Mem2 VRM temp 42C(41
GPU load 100%
Mem Controller load 25%(26)
Mem Used Dedicated 634MB(1346)
Mem Used Dynamic 131MB(150)
GPU only Power Draw 206W(188)
GPU Voltage 1.1V(1.037)
Mem Voltage .850V

This was with a 15Minute+ running of Render test in GPU-Z.

 

[guskline]

Awesome! How do you like it? Any quirks, observations vs your 1080ti?

No quirks so far other than running a lot cooler. I ran Firestrike with the core at 2050 and voltage maxed without a problem.

It will take some time to see how much this enables OCing but so far it looks like custom water cooling really helps.

 

[Stuka87]

Its interesting that the fans can draw so much power at load. 80 watts is plenty incentive to LC the card.

They aren't. Power draw increases as heat increases. This has been shown in several tests by various sites.