LN2 test - Bulldozer FX-8120 || GIGABYTE 990FXA-UD7 || 4x Radeon HD 7970

[massman]

Hey all,

Short post because the CPU was not good.

As we all know, in between now and April 29th, there's pretty much nothing to do for the extreme overclocker since Ivy Bridge will kick ass. So instead, I wanted to see how good a full AMD setup really performed in 3DMark11. Before, I already tested Bulldozer with 4x GTX580, but that ended up in one big LOL/LMAO/ROFL-fest since it could hardly compete with any other similar setup. So, hopes were not that high to begin with. In any case, I started with a simple run on air cooling to check initial performance.

I pushed the CPU to 4GHz (small OC) on simple air cooling. The result was mindblowing:


(no tesselation tweaking)

Very promising start.

Without too much stress, we could then pull down the CPU to -180°C and see how high it could clock. Turns out ... not so high. 6150MHz stable through Physics was about the maximum! One core kept bugging in the tests, so ... little we can do about it. With 6G only, it was already quite obvious that this setup wouldn't be doing too well. Anyway, had some fun pushing the NBclk and DRAMclk in physics. 990FXA-UD7 still is my favorite AMD board .

Eventually, I ended up around 18.2k (tesselation disabled).



At 6GHz, this score isn't that bad (for Bulldozer) relatively speaking. 20K wasn't that super-easy with the GTX580s, with 7970s it should be quite possible. But, well, you need a good CPU.

Playing with memory and physics is still quite fun, though. I managed a little higher memory clocks through Physics. Should be interesting to see what happens when the memory is under LN2 too!





Hardware used

- AMD Phenom FX-8120 (retail)
- GIGABYTE GA-990FXA-UD7 (F5 bios)
- 4x Sapphire Radeon HD 7970
- 2x 2GB GSkill Flare
- Coolermaster Silent Pro M 1000W
- Thortech Thunderbolt 1200W

Some pictures of the session:













Till the next time!

[BeepBeep2]

FX is rather strong in Vantage and 11

[Hondacity]

UD7 would be my favorite too if I had 4gpus lol. Maybe the added PCIE lanes limited the max clocks?

[FlanK3r]

- AMD Phenom FX-8120 (retail)



but nice results, Il try next month my FX again, I need new FX 81xx looks promising for OC. Still I told, FX is not so bad CPU, this CPU has only weak x87 performance, FP and some single threads (not all). Its only half-product. Vishera will be better.

[BeepBeep2]

but nice results, Il try next month my FX again, I need new FX 81xx looks promising for OC. Still I told, FX is not so bad CPU, this CPU has only weak x87 performance, FP and some single threads (not all). Its only half-product. Vishera will be better.

Try the one you said was too hot to run Cinebench past 4.8 GHz.

[massman]

UD7 would be my favorite too if I had 4gpus lol. Maybe the added PCIE lanes limited the max clocks?

Not sure. I used similar settings like I used for the 7G runs with the media sample. This chip couldn't do 4.5G on air either ... I reckon it's particularly shyte.

Or, well, Leeghoofd tested it before. He probably degraded it a GHz or so.

[dinos22]

Or, well, Leeghoofd tested it before. He probably degraded it a GHz or so.

Lol good one

[blossa]

Nice one! Did you try max NB-clocks for this session? I managed to get +3.7GHz working but with a lot lower mem-clocks. Not sure what is the best though.

[massman]

I got about 3.6G IMC stable with 1.5V Vnb, which is similar to the other chips I tested. I reckon +3.7G would work too, but then I'd have to really push on the voltage.

Performance-wise, at least in 3D11 Physics, you'd want a combination of high NB and high MEM. Ideally, I'd like to see what kind of performance I'd get with the memory divider one notch lower and HTT ~370MHz instead of 330MHz, but so far I haven't found the combination of settings that allow high HTT, high dram and high NB. Performance significantly increased when switching from the 1866 to the 1600 divider, btw.

[blossa]

I had the same problem. If I remember right, around 350-355MHz was benchable, but then I did not get good NB and mem. That was on UD7, I have not checked if UD5 behaivs the same.