[H] users 7950 Overclock Results

So my card isn't broken, it was a driver issue.

i installed new drivers over already installed ones which could have caused problems. Deleted everything and re-installed and i can actually run it maxed out now.

I'll do some OCing and get back here

*EDIT* did a pass at the same settings as the furmark test and it ran fine. Going to sleep but ill run it longer tomorrow.

i get lower load temps in Heaven than Kombuster / Furmark....odd
 
Last edited:
Krayzie135 said:
i get lower load temps in Heaven than Kombuster / Furmark....odd
Click to expand...

Furmark is more designed to just heat up the card.

Heaven is a much more complex gpu process that test many diffrerent GPU processes but doesn't heat the card up as much.
 
SonDa5 said:
Furmark is more designed to just heat up the card.

Heaven is a much more complex gpu process that test many diffrerent GPU processes but doesn't heat the card up as much.
Click to expand...

Correct. While an overclock can fail due to excessive heat as well, sometimes just complex rendering is a more accurate way to test stability, especially when temperautres are kept in check.
 
SonDa5 said:
I do both. Running both time is neat to see.

For total system over clocking I have found that a slightly unstable CPU will not allow my HD7950 to go as fast. So having stable CPU over clock and stable GPU over clock are both important for finding total maximum performance for gaming or benchmarks.

Also when I run both I use HDinfo64 and monitor voltage sensors to find any weakness in my PSU or cooling.
Click to expand...

Its a good PSU test for sure.
 
SonDa5 said:
Furmark is more designed to just heat up the card.

Heaven is a much more complex gpu process that test many diffrerent GPU processes but doesn't heat the card up as much.
Click to expand...

Blkout said:
Correct. While an overclock can fail due to excessive heat as well, sometimes just complex rendering is a more accurate way to test stability, especially when temperautres are kept in check.
Click to expand...

ahh that makes sense.

Paid day off tomorrow which means lots of computer time. Think 1350core is possible? haha
 
Krayzie135 said:
im going to have to take you up on that challenge :cool:
Click to expand...

I don't know what the record is right now for a non-volt modded air or watercooled 7950 or 7970, but I would think 1350-1400 would be near the best in the world.
 
Krayzie135 said:
ahh that makes sense.

Paid day off tomorrow which means lots of computer time. Think 1350core is possible? haha
Click to expand...



On air it may be possible if you have a golden gpu for a few moments on load then the driver will abandon ship.


On water I do think it is possible. Mine can do it on air for not very long but then it gets buggy and driver quits. On water my card can do it but I only run 1350+ for competitive bench marking.

Backing down to 1300/1600 is sweet spot for gaming for me.

Good luck with your over clocking.


I still think you may have some type of low voltage issue with your PSU. When you get into extreme over clocking the high voltage requirements on CPU and GPU really take a toll on PSU performance.
 
ill try swapping a PSU out of another computer and comparing voltages and see if that does anything
 
Is anyone else getting sub-1100 max. OCs on these? It looks like the best my card can do (Gigabyte Windforce, default voltage .993, ASIC 79.0%) is 1025 on any voltage...

Still testing, but I just wanted to know if there are people that aren't getting insane 1200+ clocks (and haven't yet returned them out of nerdrage >:D)

EDIT: "A broken appliance demonstrated for the repairman will work flawlessly." Now that I'm actually trying to post solid numbers, I do get solid 1100! Forget my bitching from before. I'll try the unofficial MSI ocs beyond later, but for now, here are the voltage settings I'm playing with:

Stock Gigabyte OC:
900/1250/.993v - never breaks 53degreesC

Highest OC without touching voltage:
950/1400/.993v/40%fan - 58 degrees (I like the performance boost, but everyone else gets so much higher numbers on stock voltage :'(

1000/1400/1025v/45%fan - 60 degrees

1050/1425/1075v/45%fan - 62 degrees

1075/1450/1100v/50%fan - 63 degrees

1100/1475/1125v/50%fan - ~70 degrees (survived Heaven runs)
 
Last edited:
I think the problem is the absence of a standard for benchmarking maximum overclocks. My Sapphire HD 7950 OC can run Unigine or Kombustor all night at 1025 (stock voltage), yet anything more than 990 makes it crash in Crysis 2. Not immediately, it takes a hour or more at 1000 for example, but eventually it will crash. That's why I state 990 mhz as my max overclock at stock voltage. I wonder what criteria everybody else follws :)
 
My Sapphire 7950 OC has its stock voltage set at 993mv and like you I could not get further than a 950 core without raising the voltage. I believe that many 7950s have their stock voltage set at 1093mv, so that may be why others are getting better results without having to make adjustments.

I purchased mine at launch and need the following voltages to be game and benchmark stable:

993mv = 950 core
1093mv = 1025 core
1149mv = 1075 core (my 24/7 setting)
1200mv = 1125 core

These voltages are "target," ie what I set in Afterburner. When I monitor voltages in GPUZ, my 1149mv target voltage exhibits a measured voltage of ~1090mv.
 
By golly, I think you're right.

A lot of people do mention comparing their 7950 results to 7970 "stock" voltages, which are all the way up in ~1100 volts. Hell, I could do pretty high at 1100 volts too, but at that point, the tradeoff for more heat and noise is just not worth it.

Considering the nice bump I already get at 1100 on Bf3 (beats the 670 I had by a nice margin), I'm pretty satisfied now.
 
Just got my card (MSI Twin Frozr 7950) a couple days ago. Haven't had too much time to run a bunch of tests. So far, working mostly on the memory I'm up to 1050/1800 (!). I've just been running Heaven benchmarks to test performance along the way, then maybe crank it to 4xAA/16xAF/mod. tesselation and let it run for 20 mins.

Performance has been increasing with every 25mhz step in memory and I haven't noticed any artifacts yet. Looking around, I may have gotten lucky with the memory on this card....:D The core locked up once when I took it to 1100mhz in Heaven. Seems pretty ballpark to me, although the default voltage on this chip is 1.039v with 87.7% ASIC quality. Looks like most chips default between .993v and 1.093v, so I may have room to work with.

Much more stability testing and ocing to come!
 
A good test for the memory OC is Witcher 2, you will see blue artifacts when the memory isn't stable. It often looks like blue streaks or blue squares.Its pretty noticeable when its happening so it makes it a good test for AMD cards.
 
Are you guys actually seeing performance gains with memory OCs more than 1375Mhz? I couldn't even squeeze out a few more average FPS on Heaven going from 1375Mhz to 1575Mhz.
 
dblinkhorn said:
Are you guys actually seeing performance gains with memory OCs more than 1375Mhz? I couldn't even squeeze out a few more average FPS on Heaven going from 1375Mhz to 1575Mhz.
Click to expand...

At 1050/1250, my average fps was 92.2. (score: 2323)
1050/1575- 98.1 (2472)
1050/1800-99.0 (2494)
1050/1850-100.1 (2522)

However, the core is much more important for sure.
1075/1800-102.2 (2574)
1125/1800- 106.2 (2675)

My memory topped out at 1875 where it finally artifacted in Heaven. ~1850 seems to be my highest stable so far Got a 3d11 score of P8973 with the card at 1100/1800. Had to bump core voltage to get stable above 1085. Right now at 1125 testing at 1.062v in Trixx.

Rest of my rig for comparison is [email protected] and 6gb ram.
 
dblinkhorn said:
Are you guys actually seeing performance gains with memory OCs more than 1375Mhz? I couldn't even squeeze out a few more average FPS on Heaven going from 1375Mhz to 1575Mhz.
Click to expand...

Try a different benchmark, Heaven sucks as a benchmark, but its good for stability testing. Not all benchmarks are the same though, some are more core speed dependent some respond better to memory overclocking.
 
There doesn't seem to be any one perfect stability testing method, for sure.

Like others mentioned, Heaven can pass some tests that BF3 will fail. (Maybe I'm not running Heaven long enough?) But by far the most picky has been the original Crysis. Passing Crysis has pretty much been the overkill test that will pass the others I've been running for the past day.

So, so far I've gotten to 1150/1525 at 1.225v, 60% fan, 73C. Not great numbers, but at least now I know it can reach the average OCs. I'm going to need more time to test beyond that, since, beyond this point, I'm running the GPU hotter and louder than I'd like to, and I have to be even more stingy about voltage and temp/performance.

EDIT: Never mind. 1155/1525 is as high as this card's gonna go, on any voltage. Overclocking is never guaranteed, as they say. But next time, I am going to buy a card I know cannot possibly have been opened and returned.

Heaven bench:
00001.jpg

Do we have a Heaven thread on [H]?
 
Last edited:
I envy you guys with dual fans, my single fan does the job but the results of higher OC are incredibly good, i just run into a heat issue.

Think im gonna get a artic cooler
 
BlackThought said:
I envy you guys with dual fans, my single fan does the job but the results of higher OC are incredibly good, i just run into a heat issue.

Think im gonna get a artic cooler
Click to expand...

Why not waterblock?
Silent
keep temps far lower than air ever could
possibly push your overclock well past 1200 on the core.
 
I have a gigabyte windforce 7950, with asic of 68%.
On stock voltage of 1.093, I can do 970 mhz max. With voltage mods, I can do 1100 @ 1.23 v.
The temperature is not the issue.


Here's a good test for those who have found out their max in long term tests:
Of course put powertune to +20% as you want to minimize throttling.
Get the latest OCCT 4.3.1 ( http://www.ocbase.com/index.php/download )
Set direct x 11, resolution to 480 x 480, shader complexity to 8, and error checking on. Don't limit the framerate. Also try 400 x 400, sometimes that errors quicker. Weird enough, max resolution wouldn't find errors... so I think that the smaller resolutions having higher fps allow the image to be cycled more so that OCCT finds errors (by seeing differences between frames that keep getting rendered over and over).

This was the only way that my card would fail. Without that test, I could do 1010 mhz @ stock V in other games, with very random crashing...sometimes only after hours.

In the old days with my gtx 260, furmark or occt 3 was good to find errors for my max clock. However, with the 7950, OCCT dx11, 480x480, shader 8, really does the trick!

I'm curious as to those that have ridiculous overclocks can pass this specific setting mentioned above.
 
I have also found that the sapphire 900 mhz 7950 bios, number 1 in tech power up vga bios archive has a different max TDP. ( http://www.techpowerup.com/vgabios/115188/Sapphire.HD7950.3072.120116_1.html )

Running 1100 mhz @ 1.23v w powertune +20%:

OCCT 3 480x480 shader 8, shows throttling even with PT +20% in MSI AB or HWinfo64 with gigabyte bios. I max out at around 18 amps for 12v input according to gpu-z.


Running 1100 mhz @ 1.23v w powertune +20%:

With the sapphire 900 mhz 7950 bios, 7970 bios, and the sapphire 950mhz 7950 bioses, I get higher FPS in OCCT 3 480x480, no throttling observed, and can max out at 22 amps for 12v input!

So, it seems like those bioses have a higher TDP setting... any one else observe this??

Also in heaven 3 with extreme tess, 16x af, 8x aa, the min fps for gigabyte bios is 19.5, with the other bioses, 22 fps. I think that is why some people see improvements using the 7970 bios (although that bios kills hdmi output on our 7950s). In another forum someone mentioned memory timings, but memory bandwith is not that severe even if I try different memory clocks, I get similar min fps.
 
SonDa5 said:
TPU doesn't have the newest Sapphire HD7950 950mhz edition BIOS. I put it up on OCN if you want to try it. It is dated from May of 2012.

http://www.overclock.net/t/1281711/hd7950-revised-with-reference-hd7970-pcb-club#post_17724563
Click to expand...
C
Yep, I tried that one. I noticed that it gave a bit different VRM temps than the Sapphire 900 mhz bios. Instead of being balanced by 5 degrees C, on the 950 bios, one would be 10-15 C higher!

The 950 mhz bios also had a similar effect with higher powertune TDP, I couldn't get it to throttle either.
 
Got one of these for my girlfriends PC. Its the MSI Twin Frozn III. It has the 7970 PCB with 1 6pin and 1 8 pin. 89.1% ASIC.

Stock volts is 1.031.

Since she is using a Lian-li V351b case (Heavily Modified). I can overclock it to 950mhz on the GPU (on stock volts!!!), but memory tops out on the slider at 1575.

In her case it gets to around 72c while playing Dayz after 4 hours.

Not a damn bad card at all. I would push it more, but I got the case's airflow about as good as I can ever get it, either way VERY impressed with the card.

Also I replace the Thermal past MSI uses with some of the good stuff from Noctua.....dropped her initial temps down by 5-6c easy.
 
Last edited:
P7914 3DMarks
GRAPHICS SCORE
7817
PHYSICS SCORE
8818
COMBINED SCORE
7469

Just have my 7950 at 950/1325 modest speed bump
 
Any idea if Gigabyte will void warranty if I remove the heatsink and put it back on?

Dissapointed that Gigabyte doesn't put a heatsink on the VRM's. At stock volts, OCCT/furmark/Kombustor can get the vrms to 85C. At 1.23v, they get up to 102!

I also want to put heatsink on the VRM's fastened with thermal tape. I was going to use this VRM heatsink from my dead 6950.

I might also change the paste to Prolimatech stuff I have.
 
robnitro, do you play BF3? Can you card keep at 1.1Ghz with 1.123V without crashing?

Mine has 71% ASIC, 1.031V, 900Mhz stock. Gigabyte Windforce.

Can do without voltage mod only 950Mhz max.

1000Mhz --- 1.056V

1050Mhz -- 1.100V

1100Mhz -- 1.142V

All these with latest 12.7 RC1 ( 19 July ). With older drivers, add +2 voltage scales up, for instance, 1000Mhz I needed 1.068V at least to prevent BF3 crashes.
 
Last edited:
How do you tell 1.031 v? What program tells you that? I know mine is 1.093v, and under load it can drop as low as 1.02v in GPU-Z or HWinfo64.

Haven't tried BF3, I mainly play racing games and some Metro 2033. I believe at stock 1.09 I was able to run them at 1020 or so. Maybe the most random crashes happened... but it didn't seem like it was oc related until I went to occt:
My stability testing is different.

Take occt 4.3.1 and run dx10, 480x480 res, shader complexity 8, fps limit leave at 0 (unlimited), error checking ON. Of course, run powertune +20% to limit throttling.
If that passes, try 320x320 res.

It won't burn out the card, powertune limits the TDP.

OCCT 3 wouldn't show errors, and if I ran OCCT 4.3.1 in dx9 or at a higher res, errors wouldn't happen, despite knowing that speed is randomly unstable.

Can you check that with your high stable clocks and see if you get any errors? Thanks in advance, it would help me understand some things and come up with an idea as to what is the best way to check stability.
 
Last edited:
I have 2x 7970 HiS reference.

1st is 77% ASIC - default voltage 0.993v
2nd is 72,5% ASIC - default voltage 1.031v

I can run both at 950mhz without changing the voltage and I can run both at 1100 changing the voltage, cannot go further because of my power supply.
 
Drangueos said:
I have 2x 7970 HiS reference.

1st is 77% ASIC - default voltage 0.993v
2nd is 72,5% ASIC - default voltage 1.031v

I can run both at 950mhz without changing the voltage and I can run both at 1100 changing the voltage, cannot go further because of my power supply.
Click to expand...

Do you play BF3

If so what are your FPS with vsync off?
 
hodor. said:
My Sapphire 7950 OC has its stock voltage set at 993mv and like you I could not get further than a 950 core without raising the voltage. I believe that many 7950s have their stock voltage set at 1093mv, so that may be why others are getting better results without having to make adjustments.

I purchased mine at launch and need the following voltages to be game and benchmark stable:

993mv = 950 core
1093mv = 1025 core
1149mv = 1075 core (my 24/7 setting)
1200mv = 1125 core

These voltages are "target," ie what I set in Afterburner. When I monitor voltages in GPUZ, my 1149mv target voltage exhibits a measured voltage of ~1090mv.
Click to expand...

I may have similar scaling (Asus DCUII). I'll have to try, it will go to ~1025MHz on stock 1.093v

speedyaxon said:
At 1050/1250, my average fps was 92.2. (score: 2323)
1050/1575- 98.1 (2472)
1050/1800-99.0 (2494)
1050/1850-100.1 (2522)

However, the core is much more important for sure.
1075/1800-102.2 (2574)
1125/1800- 106.2 (2675)

My memory topped out at 1875 where it finally artifacted in Heaven. ~1850 seems to be my highest stable so far Got a 3d11 score of P8973 with the card at 1100/1800. Had to bump core voltage to get stable above 1085. Right now at 1125 testing at 1.062v in Trixx.

Rest of my rig for comparison is [email protected] and 6gb ram.
Click to expand...
Thanks for that analysis! I won't even bother trying to overclock over ~1600MHz on the ram.
Did you change the memory voltage?

Has it been figured out where the core overclocking quits reaping the full benefit in a similar fashion?
 
BlackThought said:
Do you play BF3

If so what are your FPS with vsync off?
Click to expand...

Vsync off, resolution 5884x1080 (because of the bezel compensation)

Caspian Border:
graphics on AUTO = 55 to 85 fps (some options on ultra and AA = off)
graphics on HIGH = 60 to 95 fps (all High and AA = off)
graphics on ULTRA = 40 to 65 fps (all Ultra and AA = on)
 
You must log in or register to reply here.
Back
Top