Mr. K6's 6950 -> 7970 Overclocking Review

[railven]

Well there's at least some kind of benchmarking done on the chips at the AMD factories. I can't believe that they would leave the plant without at least verifying if they work. I really don't know what exactly that entails, but possibly with simple, and not so time consuming tests, they can figure out how good a chip is?

I'm sure all stock cards are tested at the stock settings through a burn-in phase, which most likely all pass. I don't see why they'd test anything beyond that unless it was a factory OC which then they'd probably test at those clocks.

As it goes for Overclocking - your success is ultimately based on your chip. I had a GTX 460 that couldn't stay stable at manufacturer stock clocks/volts beyond 825, and oddly enough the volts were higher than the reference volts from nVidia. It was an EVGA. Long story short, 2 RMAs later I got to keep it and EVGA gave me a GTX 460 SC at the difference in cost (like $20 with some change.) GF got a working card and I got a PhysX Card

Also, are there physical factors known before the chip is actually made that can affect its final quality? Are all wafers made equal, and is a wafer equal over all its surface?

As with all manufacturing processes - not all yields are identical and anamolies can exist from one run to another.

 

[MrK6]

Since I do not want to read the past three pages of hardware insecurity and outrageous claims about some 470s.

Mrk6, did you end up able to give the card voltage and not get clocks you were hoping for ? Or could you not apply voltage in the first place ? Thanks, interested to hear on results.

After listening to Unwinder over at my Guru3D post, it seems more the latter. If voltage is being applied per his assurance, then I'm simply not getting the clocks I want out of this card (topping out at ~1175MHz or so on air, and that's only in some applications). As I mentioned, I'm not paying $550 for a graphics card that responds this poorly, so back it goes and I will try my luck with the Gigabyte. I can say for certain that this is probably the last XFX product I will buy. I'm not dismissing the coincidental nature of my experience with them, but coupled with the past evidence against the company, the half dozen competitors of theirs that have NOT been accused of such practices, and a little common sense, it just isn't worth my time. The results of the Gigabyte will be the final testament to this.

Also, great discussion on binning and you guys cleared a lot of its (usual) misconceptions. To do anything more than a quick stability test is simply not cost effective. :thumbsup:

 

[Quantos]

I'm sure all stock cards are tested at the stock settings through a burn-in phase, which most likely all pass. I don't see why they'd test anything beyond that unless it was a factory OC which then they'd probably test at those clocks.

Perhaps they're able to tell just how good a chip is just by looking at that burn-in test? Even if they're all tested at stock settings, maybe there's an indication of how good the chip is. For instance, maybe the voltage stability differs, or something like that. It's not hard to think there's some intricate statistic that we don't necessarily know about that would give them an idea...

Hypothetical, though... I might be completely wrong

 

[SlowSpyder]

Since I do not want to read the past three pages of hardware insecurity and outrageous claims about some 470s.

Mrk6, did you end up able to give the card voltage and not get clocks you were hoping for ? Or could you not apply voltage in the first place ? Thanks, interested to hear on results.

Saw some posts of people saying to avoid the standard XFX design because they may have taken the best cores and put them in the DD BE version. Just wondering on your results as you have the standard XFX.

Who cares... if there is one thing I learned from the GTX590 launch, it is that no one wants to overvolt an enthusiast level card anyway.

 

[Madcatatlas]

Whats taking so long!? Wtb some graphs and or some numbers

 

[Termie]

Whats taking so long!? Wtb some graphs and or some numbers

If you read above, you'll see he's having trouble overvolting the card.

That being said, OP - why not just put up your stock vs. stock and stock volts vs. stock volts numbers while you're waiting for the RMA? I assume that the stock numbers won't change when you get the Gigabyte card, nor will the numbers at 1125 core, if you want to put those up too.

 

[MrK6]

Perhaps they're able to tell just how good a chip is just by looking at that burn-in test? Even if they're all tested at stock settings, maybe there's an indication of how good the chip is. For instance, maybe the voltage stability differs, or something like that. It's not hard to think there's some intricate statistic that we don't necessarily know about that would give them an idea...

Hypothetical, though... I might be completely wrong

Without a doubt you can tell the quality from a quick burn-in, but that doesn't mean it's economical to use it. Having a test be pass/fail is a lot easier than a qualitative expression.

Who cares... if there is one thing I learned from the GTX590 launch, it is that no one wants to overvolt an enthusiast level card anyway.

Maybe when's already overstepped it's boundaries and shoddily engineered, neither is the case with the 7970.

If you read above, you'll see he's having trouble overvolting the card.

That being said, OP - why not just put up your stock vs. stock and stock volts vs. stock volts numbers while you're waiting for the RMA? I assume that the stock numbers won't change when you get the Gigabyte card, nor will the numbers at 1125 core, if you want to put those up too.

Card's already gone (NCIXUS support = awesome). Hopefully I'll have my Gigabyte tomorrow, if not, Monday. I have a full weekend planned anyway so I'm not sure how much I can get done as it is, we'll see. For now, there are some numbers up there.

 

[railven]

Perhaps they're able to tell just how good a chip is just by looking at that burn-in test? Even if they're all tested at stock settings, maybe there's an indication of how good the chip is. For instance, maybe the voltage stability differs, or something like that. It's not hard to think there's some intricate statistic that we don't necessarily know about that would give them an idea...

Hypothetical, though... I might be completely wrong

I'm sure they do that too. I'm just going off the little I know about the field and the dots I can connect using my level of logic. I'm sure someone like IDC can come in here and open a can of knowledge on us. haha.

I'd guess they have tiered levels of testing. For example, they can probably run a standard burn-in at stock settings. Then a second level at a higher clock settings where they monitor volts and look for high leakage. And so forth.

I can fathom a 3-swings and you're out basic system where chips that pass a third tier of testing can be sold off at a higher cost to AIB for even higher manufacturer settings or sold at stock settings with a beefy heatsink with a promise of high overclocks - of course with a premium.

The ones that fail the initial test will get sent to another set of tests to determine where they can be used (HD 7950 and HD 7930's respectively.) And even there you'll find an HD 7950 that does exceptionally well, but not good enough to carry the HD 7970 name (probably volts.)

 

[Termie]

...
Card's already gone (NCIXUS support = awesome). Hopefully I'll have my Gigabyte tomorrow, if not, Monday. I have a full weekend planned anyway so I'm not sure how much I can get done as it is, we'll see. For now, there are some numbers up there.

Oh, thanks - didn't see that you had a separate spreadsheet linked in the OP.

You know, I'd be curious if those load power consumption figures hold with the Gigabyte card. 412w just seems way too high, especially considering it was way over your OC/OV/unlocked 6950. The 7970 should come in maybe just above a stock 6970, not way above a hyperclocked one, and only 65w higher than a stock 6950, not 130w higher. Even your slightly higher idle power use was suspect - maybe it really was a bad card.

This is just going on Anandtech's review: http://www.anandtech.com/show/5261/amd-radeon-hd-7970-review/27

 

[railven]

@ OP/Thread: I'm really thankful for you taking the time to test these things out.

I do hope the XFX was just a bad sample of what we can expect.

And your videos of the stock cooler, @50% it is probably a little louder than my HD 5870 around 50%. But with the recent grinding noise, I'm not sure haha,

Hopefully the MSI Lightining is in the pipeline for March.

 

[Porter_]

interested in what you find with the Gigabyte card. FWIW my XFX Black Edition DD card doesn't respond to voltage either. i've used both Afterburner and Trixx and can't break past an 1100 MHz wall on the core. it seems like both Afterburner and Trixx are still getting the kinks worked out for our cards, but i'm not optimistic my card will do any better once they do. anyway, looking forward to your results with the Gigabyte card.

 

[Awkward]

trixx can go up to 1.3v

not mine, mine is still being shipped to me :|

 

[MrK6]

Looks like the card is coming in Monday, I might be able to get it on water and some scores up by night some time.

 

[MrK6]

Oh, thanks - didn't see that you had a separate spreadsheet linked in the OP.

You know, I'd be curious if those load power consumption figures hold with the Gigabyte card. 412w just seems way too high, especially considering it was way over your OC/OV/unlocked 6950. The 7970 should come in maybe just above a stock 6970, not way above a hyperclocked one, and only 65w higher than a stock 6950, not 130w higher. Even your slightly higher idle power use was suspect - maybe it really was a bad card.

This is just going on Anandtech's review: http://www.anandtech.com/show/5261/amd-radeon-hd-7970-review/27

That spreadsheet is where you'll find all updates first, so always check that. The power consumption is very odd, even at stock. It could be that Crysis 2 just really pushes the GPU, where as the 6950 might have been limited by its tessellation engine or something. Like you said, the Gigabyte's results will shine some light on this. Also, the higher idle result might be my own fault. I generally let it idle for a minute or so and take the recorded number. It might take 10 minutes or having the monitor off or something to get the GPU to kick down to its 3W state. On the other side though, maybe this shows how irrelevant that 3W idle state is, i.e. if you're using your computer, it's never going to enter it. I'll take a look into it.

@ OP/Thread: I'm really thankful for you taking the time to test these things out.

I do hope the XFX was just a bad sample of what we can expect.

And your videos of the stock cooler, @50% it is probably a little louder than my HD 5870 around 50%. But with the recent grinding noise, I'm not sure haha,

Hopefully the MSI Lightining is in the pipeline for March.

Unfortunately, the camera doesn't pick up the sound pressure that well. The quality is relatively accurate, but the overall intensity is "fuller" in person, I would say. In either case, it's an improvement over previous generations, so anyone upgrading should expect a positive change.

interested in what you find with the Gigabyte card. FWIW my XFX Black Edition DD card doesn't respond to voltage either. i've used both Afterburner and Trixx and can't break past an 1100 MHz wall on the core. it seems like both Afterburner and Trixx are still getting the kinks worked out for our cards, but i'm not optimistic my card will do any better once they do. anyway, looking forward to your results with the Gigabyte card.

According to Unwinder, AB's voltage control is completed, and there will be no more updates. That's one of the reasons I was a little more certain in blaming the card here over the programs. As you said, the Gigabyte will at least give another data point, although it will be hard to derive any resounding conclusion from only two data points.

trixx can go up to 1.3v

not mine, mine is still being shipped to me :|

I tried AB, Trixx, and GPUTweak. They all had the same result, so I'm guessing they all access the same voltage controllers.

 

[MrK6]

Got the Gigabyte 7970 in today, so far it's flying through everything I've thrown at it at 1125 on stock voltage, so it's definitely a better card than the XFX I returned. I'll probably do some quick tests tonight and then put it on water and run the whole suite first thing tomorrow. Anyone want to take a stab at what the max OC on water will be?

 

[Termie]

Got the Gigabyte 7970 in today, so far it's flying through everything I've thrown at it at 1125 on stock voltage, so it's definitely a better card than the XFX I returned. I'll probably do some quick tests tonight and then put it on water and run the whole suite first thing tomorrow. Anyone want to take a stab at what the max OC on water will be?

Congrats. Definitely sounded like that XFX was messed up.

Honestly, I'm not so sure that launch-day non-reference cards are the way to go. After reading up on the XFX cards, it seemed like they weren't quite ready for prime time. I've never had any problem with reference cards, except maybe their being a bit loud.

 

[Elfear]

Got the Gigabyte 7970 in today, so far it's flying through everything I've thrown at it at 1125 on stock voltage, so it's definitely a better card than the XFX I returned. I'll probably do some quick tests tonight and then put it on water and run the whole suite first thing tomorrow. Anyone want to take a stab at what the max OC on water will be?

Glad your Gigabyte is working out better for you. Magic 8 Ball says 1350/1600 under water.

 

[ader098]

Mr. K6, are you still planning on using the mcw82 on the 7970? If you are, then you should be aware that universal gpu blocks require a spacer since the 7970 gpu die is surrounded by a shim.

 

[Tempered81]

Anyone want to take a stab at what the max OC on water will be?

1325/1800 @ 1.3v set in AB

 

[wahdangun]

Congrats. Definitely sounded like that XFX was messed up.

Honestly, I'm not so sure that launch-day non-reference cards are the way to go. After reading up on that XFX cards, it seemed like they weren't quite ready for prime time. I've never had any problem with reference cards, except maybe their being a bit loud.

yep, in every forum that i read(guru3d, techpowerup, and tomshardware) it look like every xfx board is a bad overclocker

 

[Troll Trolling]

Hi,
could also try some undervolting?
Stock voltage HD 7970 overclocks very well, so I think it actaully got a lot more voltage than it really needs.

Thanks

 

[SlowSpyder]

1325/1800 @ 1.3v set in AB

Hehe... I'm just getting started. Playing Skyrim @ 1075MHz on the core, stock volts. I have Sapphire Trixx installed, I think I'll be gunning for 1.3GHz soon.

I had two Sapphire VaporX cards... for the first time in a couple of years, I can hear my graphics card fan. Not loud, but a low hiss from the blower as I game. Not too bad, but I wonder how things will go with more clockspeed and voltage. That's probably what will limit my OC, the noise.

 

[Panopticon]

In regular gaming I haven't had the fan go above 41% that is at 1125/1575. You can set it manually and see how those noise levels sit with you, for me its not a problem at all.

 

[SlowSpyder]

In regular gaming I haven't had the fan go above 41% that is at 1125/1575. You can set it manually and see how those noise levels sit with you, for me its not a problem at all.

It's not bad, not annoying. Just different after two years of nearly completely silent video cards.

 

[3DVagabond]

Hehe... I'm just getting started. Playing Skyrim @ 1075MHz on the core, stock volts. I have Sapphire Trixx installed, I think I'll be gunning for 1.3GHz soon.

I had two Sapphire VaporX cards... for the first time in a couple of years, I can hear my graphics card fan. Not loud, but a low hiss from the blower as I game. Not too bad, but I wonder how things will go with more clockspeed and voltage. That's probably what will limit my OC, the noise.

Quit friggin around! Put the juice to that card. Make her talk, dude!!!