Question 6800 vs 6800XT vs N32

[biostud]

I'm planning to upgrade my system feb/march when the 7800X3D is released so I can decide which 8-core CPU makes most sense to me. I have an Odyssey G9, but I mostly play older games, and world of tanks and no interest in raytracing (at current pricing and performance). The 6800 series seems to offer bester price/performance atm, an N32 arrival is unknown. So what would you do? Buy now before stock is gone, wait and maybe miss out, or something else? I really dont want to spend much more than $800 in Danish prices for a video card.


(Prices for comparison)
3080 10GB ($999)
7900XT ($1142)
4080 ($1570)

 

[Chicken76]

I mostly play World of Tanks, so that is why I chose that benchmark. It is only the shadows that are ray traced (they are not implemented in the game yet). I was able to go from medium settings to highest settings and double my framerates in game (70-100) . It should be between 2.5-3.0 times faster at same settings in most games, but I will also use it to improve visual quality, so I think I can get 2-2.5 better fps and better visuals. And then in some games I will test FSR 2.x to see how that will affect performance/visuals.

That's good to hear. Considering how big it is and being the middle of winter, you shouldn't be hearing it. How are the temps?

 

[biostud]

That's good to hear. Considering how big it is and being the middle of winter, you shouldn't be hearing it. How are the temps?

I've only played with it 1,5 hours and am not likely to use it until next week as my dad is staying over a taking care of one of my daughters, and for the first time in six years me and the wife is going away for the weekend without the children.

 

[igor_kavinski]

I've only played with it 1,5 hours and am not likely to use it until next week as my dad is staying over a taking care of one of my daughters, and for the first time in six years me and the wife is going away for the weekend without the children.

Congrats in advance for another round of triplets

 

[biostud]

Congrats in advance for another round of triplets

Yeah, but no

 

[biostud]

So I started to do some tweaking and here are some results.

First of all the adrenalin software seems to ignore your voltage settings completely. When I tried running the frequency at 2600 the voltage simply stayed at 1150mV no matter what. I set it to 1050mV and when I changed the freqency to 2400, the frequency stayed around 2300Mhz and 1015mV, and when I set the frequency to 2300Mhz it dropped to 2200mhz and 950mV. So a bit weird but it seems stable. I have set memory frequency to 2100Mhz and fast timings.

If these scores are similiar to other games and stable then I'm pretty satisfied

 

[igor_kavinski]

If these scores are similiar to other games and stable then I'm pretty satisfied

Excellent! 100W power reduction for not that much performance loss. Pathetic that the stock settings waste so much unnecessary power.

 

[In2Photos]

So I started to do some tweaking and here are some results.

First of all the adrenalin software seems to ignore your voltage settings completely. When I tried running the frequency at 2600 the voltage simply stayed at 1150mV no matter what. I set it to 1050mV and when I changed the freqency to 2400, the frequency stayed around 2300Mhz and 1015mV, and when I set the frequency to 2300Mhz it dropped to 2200mhz and 950mV. So a bit weird but it seems stable. I have set memory frequency to 2100Mhz and fast timings.

If these scores are similiar to other games and stable then I'm pretty satisfied

View attachment 75192

Hmmm, I was able to get undervolting working on my daughter's 6800XT using Adrenaline. Can you post a screenshot of your adrenaline settings? My GPU came in today and I'm working on finishing my build, but it might be a few days before I can test it on mine. Maybe I can check on my daughter's for you if you want?

 

[biostud]

Hmmm, I was able to get undervolting working on my daughter's 6800XT using Adrenaline. Can you post a screenshot of your adrenaline settings? My GPU came in today and I'm working on finishing my build, but it might be a few days before I can test it on mine. Maybe I can check on my daughter's for you if you want?

 

[biostud]

Running the radeon overlay on the WoT benchmark will deduct 1000-1200 from the score.

Also in BL3 it is around 200-210W which is also pretty substansial.

 

[biostud]

So did a little testing on BL3
@200W avg. FPS 66.52
@280W avg. FPS 71.25

So 7% faster with 40% more power consumption, or 6.6% slower with 28.5% less power used.

Or 3W/fps vs 4W/fps

 

[In2Photos]

View attachment 75199

I seem to recall that min frequency is supposed to be closer to the max frequency. It seems counter intuitive I know. Most of what I read/watched said to set it 100 MHz below the max. Other than that your settings don't seem out of whack.

 

[PJVol]

First of all the adrenalin software seems to ignore your voltage settings completely. When I tried running the frequency at 2600 the voltage simply stayed at 1150mV no matter what.

In Navi 2x the "voltage setting" aka "slider" just doesn't do what you expect it to.
AMD driver sources referred to it as the offset applied on target voltage calculated.
The picture below shows how the actual voltage supplied by VRM (can be seen in HWInfo64) and GFX frequency are affected by VddGfxOffset. The workload is Shadow of tomb raider settings menu.
The most favorable scenario when Power Limit always hit prior to the Frequency one.

 

[biostud]

In Navi 2x the "voltage setting" aka "slider" just doesn't do what you expect it to.
AMD driver sources referred to it as the offset applied on target voltage calculated.
The picture below shows how the actual voltage supplied by VRM (can be seen in HWInfo64) and GFX frequency are affected by VddGfxOffset. The workload is Shadow of tomb raider settings menu.
The most favorable scenario when Power Limit always hit prior to the Frequency one.

Well, that explains it, but then why even put a slider there? Even if I lowered the power limit, I couldn't get close to 200w.

 

[Insert_Nickname]

Excellent! 100W power reduction for not that much performance loss. Pathetic that the stock settings waste so much unnecessary power.

You can use Radeon Chill for pretty much the same effect. If you have a 60Hz monitor, just cap the framerate at 60.

Power savings can be pretty dramatic even without undervolting.

 

[biostud]

You can use Radeon Chill for pretty much the same effect. If you have a 60Hz monitor, just cap the framerate at 60.

Power savings can be pretty dramatic even without undervolting.

I run a 240hz monitor, not that I will ever be close to 240 fps @ 5120x1440p

 

[PJVol]

Even if I lowered the power limit, I couldn't get close to 200w.

Idk what workload you ran but assuming the GPU was fully loaded ~ 99% (i.e. NO frame limiters applied), you just hit the Frequency limit there.
Anyway, having HWInfo64 sensor panel running in the background often helps to understand WHY.
Just post a screen of the sensors during workload when something is not clear.

PS: regarding the efficiency, tuning the card allows it to perform @200W without ANY loss of performance compared to stock settings (@255W TGP/PPT)

 

[biostud]

Idk what workload you ran but assuming the GPU was fully loaded ~ 99% (i.e. NO frame limiters applied), you just hit the Frequency limit there.
Anyway, having HWInfo64 sensor panel running in the background often helps to understand WHY.
Just post a screen of the sensors during workload when something is not clear.

Yes, it was fully loaded.

I guess if I can't make it run at a higher frequency at a lower voltage, then it is quite limited how much time I should invest in tweaking it, as the gains would be minimal.

or have I misunderstood something?

 

[In2Photos]

Idk what workload you ran but assuming the GPU was fully loaded ~ 99% (i.e. NO frame limiters applied), you just hit the Frequency limit there.
Anyway, having HWInfo64 sensor panel running in the background often helps to understand WHY.
Just post a screen of the sensors during workload when something is not clear.

PS: regarding the efficiency, tuning the card allows it to perform @200W without ANY loss of performance compared to stock settings (@255W TGP/PPT)

Maybe I did better than I thought on my daughter's card. I was able to get a little more performance than any of the stock or "one click" settings and the card peaked at roughly 225W with similar temps from stock settings.

Yes, it was fully loaded.

I guess if I can't make it run at a higher frequency at a lower voltage, then it is quite limited how much time I should invest in tweaking it, as the gains would be minimal.

or have I misunderstood something?

Glad you're asking these questions! I'm enjoying the discussion! Maybe we'll both end up with the right answers and nicely tuned cards!

 

[PJVol]

or have I misunderstood something?

The GPU is limited either by power (PPT limit) or by frequency (adrenaline slider Fmax).
You said you can't reach 200W in some workload, which means the GPU is hitting the frequency limit you set.

Therefore, it makes sense to set the Fmax so that it is never reached within the given power limit.

Once you set the desired power limit and then Fmax, choose the Vdd offset moving the voltage slider left, until unstable in TimeSpy, and add some safety margin (~ 5-10mv).

That's basically all the tuning.

PS: It's very important to follow the order of the steps above.

 

[biostud]

The GPU is limited either by power (PPT limit) or by frequency (adrenaline slider Fmax).
You said you can't reach 200W in some workload, which means the GPU is hitting the frequency limit you set.

Therefore, it makes sense to set the Fmax so that it is never reached within the given power limit.

Once you find your optimal power and frequency limits, choose the Vdd offset moving the voltage slider left, until unstable in TimeSpy, and add some safety margin (~ 5-10mv).
Basically, that's all the tuning.

I can't run timespy, it throws me an error that my display is either overclocked or an overlay is interrupting. I haven't even touched power limit, so that just the ~275W. But I will try to tinker with it.

 

[PJVol]

it throws me an error that my display is either overclocked or an overlay is interrupting.

Timespy is the most sensitive benchmark I know of for catching OC instabilities.
My card will crash with anything below 1075-1080mv in GT2 test @ 255-300W PPT.
And @200-220W it needs even more - like 1090-1095mv.
Even though I can play games hours and days without a crash with just 1050mv set. But all this is quite individual, i.e. GPU sample and the platform itself specific.

 

[Insert_Nickname]

I run a 240hz monitor, not that I will ever be close to 240 fps @ 5120x1440p

It's actually a lot less resolution then a "real" 5K monitor (5120x2880p = 14.7MP) and only 89% of 4K (7.37 vs 8.29MP). Depending on the game, 240FPS shouldn't be outside the realm of possibility with a beefy enough card.

As to whether it's worth it, that is a different matter.

 

[biostud]

Timespy is the most sensitive benchmark I know of for catching OC instabilities.
My card will crash with anything below 1075-1080mv in GT2 test @ 255-300W PPT.
And @200-220W it needs even more - like 1090-1095mv.
Even though I can play games hours and days without a crash with just 1050mv set. But all this is quite individual, i.e. GPU sample and the platform itself specific.

Is it the card/driver that crashes or Timespy?

I don't get any freezing, black screen or anything similar. As soon as it starts Graphics test 1 it simply throws some DX error that my display is overclocked or that I'm using an overlay that will disturb the benchmark, and then stop the program. So I don't think my card is crashing.

 

[Stuka87]

If you are just wanting to lower power consumption, there is an undervolt button in the adrenalin software under tuning control.

 

[PJVol]

As soon as it starts Graphics test 1 it simply throws some DX error that my display is overclocked or that I'm using an overlay that will disturb the benchmark, and then stop the program

Yeah, it's likely something is wrong with the driver setup or 3dmark.