Bulldozer FX-8150 || GIGABYTE 990FXA-UD7 || MSI 4x GTX580 Lightning

[Massman]

Getting ready, scores and more information is for the next coming days.

SETUP:

- 2x Bulldozer FX-8150
- GIGABYTE GA-990FXA-UD7 (F5 bios)
- 4x MSI GTX 580 Lightning
- 2x Dominator GTX2
- Silent Pro 1000W
- AX1200

Big thanks goes out to the boys at Tones.BE (http://tones.be/) as well as GIGABYTE HQ and of course the legendary Macci from AMD !

PICTURES:

http://hwbot.org/blog/wp-content//setup22.jpg

http://hwbot.org/blog/wp-content//setup12.jpg

http://hwbot.org/blog/wp-content//eeehr.jpg

[Sam OCX]

nice showoff

[Massman]

I'm still installing ... nothing more I can do at the moment. Wait for the first 3DMark11 at stock speeds :D

[Massman]

+6.25% in CPU clock = +14.77% in Heaven DX11 :D

http://hwbot.org/blog/wp-content//4g.png

http://hwbot.org/blog/wp-content//4.25g.png

[Leeghoofd]

Good to see you have dual channel going now ... want to try some new rams :D ??







ps : don't kill the PSU ;)

[Massman]

4250 MHz: +6.25% in CPU clock = +14.77% in Heaven DX11
4500 MHz: +12.5% in CPU clock = +23.79% in Heaven DX11
4750 MHz= +18.75% in CPU clock = +29.07% in Heaven DX11

http://hwbot.org/blog/wp-content//4500.png

http://hwbot.org/blog/wp-content//4500-2.png

http://hwbot.org/blog/wp-content//4750.png

http://hwbot.org/blog/wp-content//4750-2.png

[Massman]

So far, though, I haven't found any clock wall ... :p

[Christian Ney]

So far, though, I haven't found any clock wall ... :p

haha in da face Intel :D

[dinos22]

LOL at the clock wall remark

what about a performance wall :D

[Massman]

There's no wall. There's crazy scaling !! :D

[dinos22]

just skip to 6Ghz and start testing "scaling" hehe

[Leeghoofd]

does he get another cpu if he kills it ? you know Massman when he's on a roll :p

[Massman]

Ran my first 3DMark11 with this setup and the score turned out better than anticipated. Maybe we had some performance issues at LITS2011 before ...

http://hwbot.org/blog/wp-content//15225.png

[Massman]

Before I go LN2 on the CPU, I need to return to the F5 bios to check some performance issues. I'm getting consistently 100p lower in Heaven DX11 on F6/G2 BIOS than I got with F5. I did reinstall the operating system (did that with F5 too), so maybe it's a software issue.

http://hwbot.org/blog/wp-content//4750-21.png

[Massman]

F5 BIOS is roughly 200p faster in 3DMark11. I gives a boost in the graphics score of the benchmark.

Last score for today. System prepped for LN2 run tomorrow. 7G, here I coooooooome !!

http://hwbot.org/blog/wp-content//15572.png

[dinos22]

the problem areas are still there the way it looks to me

GT1 & PT/CT scores

don't forget we didnt run NB higher than 2000 or so too and memory was left at auto

[dinos22]

also 1100T or 1090T will beat BD scores imo

http://hwbot.org/submission/2180076_dinos22_3dmark11___performance_4x_geforce_ gtx_580_15233_marks

look at this run with lower GPU speed and CPU frequency

look at GT1 1090T eek

http://hwbot.org/image/607812.jpg

[Massman]

Nice catch, Dino!

GT1 seems to be the most problematic game test for Bulldozer. The three others are roughly on par with Thuban. Given the Bulldozer architecture, I'd say the PT score is fine too for a 4 module chip. Combined isn't too big of an issue.

// Edit:

If GT1 would be so low due to the it's CPU-intensive nature, increasing the clock frequency of the CPU should yield in great scaling of GT1. Or at least greater scaling than the three other game tests.

GT1 4500 MHz = 86.28 FPS
GT1 4625 MHz = 87.74 FPS
GT1 4750 MHz = 87.85 FPS
(on F5 bios, 4750 = 89.61 FPS)

I don't understand why GT2 is getting worse when increasing the CPU frequency, though ...

http://hwbot.org/blog/wp-content//14994.png

http://hwbot.org/blog/wp-content//15216.png

http://hwbot.org/blog/wp-content//15456.png

// edit2:

Also, GT1/GT3 ratio of my scores compared to Dino's is very strange:

http://hwbot.org/blog/wp-content//14994.png vs http://hwbot.org/image/673847.jpg

[matose]

I still don't understand why you are playing 4xSLI with Bulldozer. Are you a masochist or something? :)
SB-E at 5GHz will score way better that FX-8150 at over 7GHz. With 4xSLI you need a lot of CPU power for the VGAs to scale which you could not get with Bulldozer...

[dinos22]

why not man

[Massman]

Lol!

I test Bulldozer for the same reason I play with IGP or like to play with the Athlon XP once in a while. I'm not interested in breaking world records or getting top scores, I'm just interested in seeing how the platform behaves and how you can maximize the results.

It doesn't make sense that GT1 slacking 15-20FPS is only due to a lack of CPU power. For one, all three other GTs are perfectly on par with Thuban. For two, increasing the clock frequency of the CPU doesn't really make GT1 scale; the scaling is identical to the other tests.

[matose]

Try 2 things:

1) Deactivate 1 core per module and test again to see if it's an arhitecture problem
2) remove 1 VGA card to see if the CPU is underpowered

[sin0822]

Did you try adjusting the CPU NB to better relate to your clocks? Under LN2 supposedly you can really turn up CPU NB for a lot better performance, though i didn't try.

Odd issue.

[dinos22]

NB doesnt clock well on BD compared to older CPUs with a little messing around i did in london

[sin0822]

Yea it wont go higher than 2.7-2.8 on air, i just saw a chart AMD made somewhere it said to expect 3-4ghz CPU NB clock under LN2, so i was wondering.

But i know for high GHZ you keep it low, but i guess if you were aiming for performance you might wanna raise it, no? I guess this is the only time anyone goes for BD performance.

[Massman]

NB does make a difference, but not 15FPS. Jumping from 2000NB to 2500NB gave me +~3FPS in GT4.

[Massman]

Running 3-way, GT1 goes down 10%. The decrease in GT4 is slightly larger, but that is what we expect from a more GPU-bound subtest.

http://hwbot.org/blog/wp-content//3way.png

// edit: played with affinity.

- C0 = 24.6 FPS
- C1 = 24.7 FPS
- C0+C1 = 42.01 FPS
- C0+C2 = 51.14 FPS
- C0+C1+C2 = 60 FPS
- C0+C2+C4 = 67.39 FPS
- C0+C1+C2+C3 = 71.2 FPS
- C0+C2+C4+C6 = 75.21 FPS
- C0+C1+C2+C3+C4+C5 = 80.44 FPS
- C0+C1+C2+C3+C4+C5+C6+C7 = 82.81 FPS

// edit2:

Looking at this this score makes me even more confused

submission

GT2, GT3, GT4 are on par with Bulldozer and Thuban. For some reason GT1 is the only subtest that is getting major performance hits. Also, GT1 isn't supposed to scale with frequency it seems. Stelaras has better GT1 with 1GHz less CPU power and 50MHz more GPU speed.

submission

[dinos22]

in terms of score ratio comment you really need to be testing exactly the same settings to see if it is just the extra frequencies on various parameters that contributed to strangeness

[_mat_]

Yea it wont go higher than 2.7-2.8 on air, i just saw a chart AMD made somewhere it said to expect 3-4ghz CPU NB clock under LN2, so i was wondering.Tried to get it up an friday and I'm stuck @ about 2.8. CPU was just at -20 °C, but it didn't show any scaling before so it was just for safety reasons.

[Massman]

in terms of score ratio comment you really need to be testing exactly the same settings to see if it is just the extra frequencies on various parameters that contributed to strangeness

The ratio "strangeness" is not a worry. It's probably due to software and/or NB clocking.

[Massman]

Two new top scores for AMD platform, I think. 7G was rather unstable ... need to try again.

http://hwbot.org/blog/wp-content//3922.091t.png

http://hwbot.org/blog/wp-content//17973t.png

[sin0822]

nice job!

[Christian Ney]

Really nice testing you did.

want to see 4k and 18k though :D

[Turrican]

scales nicely.

[Sam OCX]

so at 10GHz, AMD might have a shot ... bring on the high leakage samples :D

[Hondacity]

I have a feeling somebody will try lapping these gigahurtz monsters lol

[Massman]

Boy oh boy, did I waste a lot of time with inefficient settings ...

http://hwbot.org/blog/wp-content//18111.png

http://hwbot.org/blog/wp-content//18251.png

Fiddling in the BIOS gave me this:

http://hwbot.org/blog/wp-content//19199t.jpg

[Hondacity]

nice nb

[fuzz3l]

Nice PJ!

What brings you the perf gain in Physiks (1400p or 14%) and Combined (600p or 8%)?
Thats huge for just a BIOS option...

_
Christian

[Massman]

4G NB also worked, but couldn't get the clocks up to 7G then.

http://hwbot.org/blog/wp-content//6500-4000-physcs-2.png

[Massman]

Nice PJ!

What brings you the perf gain in Physiks (1400p or 14%) and Combined (600p or 8%)?
Thats huge for just a BIOS option...

_
Christian

I got the boost when I changed two things, but didn't have enough LN2 to check which of the two was causing the actual gain. I did:

- run NB 1:1 with HT Link
- lower memory multiplier from 1:4 to 3:10

I spoke about this gain with Macci and the best we could come up with is that choosing a lower memory multiplier is tightening up the memory in some way. This probably in combination with a highly overclocked CPU and NB. I didn't see this kind of bottleneck on air cooling.

I think Leeghoofd is picking up the gear tomorrow evening and can't keep the cards for much longer anymore either, so further testing will be for another time, I think.

[just_nuke_em]

How many volts did you have on the cpu-nb if you don't mind saying?

[Massman]

1.6V for 4G

[Massman]

FYI, this is the data I kept from air pretest runs.

http://hwbot.org/blog/wp-content//4wayairtesting.jpg

[dinos22]

awesome job mate keep at it :D

[Hondacity]

linear scaling? show a graph please

:)