MilkyWay@H Benchmark thread Spring 2014 - Summer 2016

[Assimilator1]

Lol @ lawsuit , I emailed the ebay guy I bought it from & just asked where he got it from, he said he bought it from a guy who had removed it from an unused development machine! Doh! I'm not too bothered, it performs as well as a HD 7950 ......... being an extremely rare card maybe it'll have collectors value some years down the road??

Think I've got ya re CPU usage, so basically none of your CPU tasks are being stalled due to being below 100% utilisation, cool .
Supposed to be an average of 5 WUs but I don't think it'll make much difference to CPU times . I see you've left the maths to me again (average 3221s).

(Btw I'm in the process of overclocking my 4820k, got it at 4.2 GHz atm, (nearly 18hrs OCCT stable so far) but it's already running rather hot, core 3 apparently peaked at 72C today! although currently it's at 66-68C. I think I can probably back off vcore a little as the bios was cranking it up with the multiplier.

 

[TennesseeTony]

I see you've left the maths to me again

I'm never sure which column you want.

 

[Assimilator1]

I sometimes forget myself with the CPU tasks , but I remember by just going for the column with the longer times.

(Backed off vcore to 1.18v, tested for 17 hrs at 4.3 GHz, all clear , now testing with the same vcore at 4.4 GHz..... damn, failed at 2mins!)

 

[Assimilator1]

Hey Tony, what happened to your 2nd score table? It's been replaced by a lovely lady .

 

[TennesseeTony]

We'll have to blame TinyPic.com for that.

 

[Assimilator1]

Lol.

Got my CPU stable at 4.34 GHz @1.18v , that leaves a wide temperature margin for the summer. Their seems to be a large jump in temps if I have the vcore at even just 1.2v, doesn't look like my cooler can handle much on this CPU (IIRC it's a TR UE120).

 

[Assimilator1]

Ah, I thought 50 secs was too good for that card. Those aren't the work unit's we are trying to benchmark.

Go to and login to your account at Milkyway and then click the milkyway preferences about mid page:

Now, you want to edit the default preferences. For two reasons. First, so only the 1.02 project gets crunched by the GPU, and 2, the 1.02 project pays the most in points.

Deselect the bottom three for now, run at least 5-10 of the Milkyway@Home (that's the 1.02 we're talking about), and don't forget to press Update, like I always do.

Don't forget to go back and recheck those other three sub-projects, if that's what you want to do. Most of us use the CPU power on something else, since the GPU is so much more efficient at running Milkyway.[/SIZE]

I know this was posted a while back, (I forgot :$), but he can't enable the v1.02 app as all WUs error out on HD 6900s (& 5800s) after Cat 13.1, as per the discussion I had with him just prior.

Btw, I'm finally benchmarking my CPU at it's 'new' speed, I reckon I'll sneak past ya to the top

 

[Assimilator1]

Updated time for my i7 4820k @4.34 GHz, HT on. Averaged from 15 WUs - 3596.7s

Well that's weird, despite an ~11.3% increase in clock speed, crunch time has only come down by ~2% , anyone any idea why?

Although I have mostly only o/c by raising the multiplier (I increased the bus speed by a whole 1 MHz, 1% ) I wouldn't of thought it would be a memory bandwidth issue as I have quad channel RAM!

Hmm, 7 later WUs finished about 70s quicker than the above average! Wth??

 

[Assimilator1]

Table updated with new times, & a clarification on how the Titans times are derived.

 

[ElFenix]

tahiti was such a beast at this stuff.

 

[Assimilator1]

Yep, & still is! (that reminds me, I've been meaning to buy an HD 7970! lol)

The 7970 GE version has a DP power of 1075 GFLOPS, no single modern AMD GPU can beat that, & I don't think NVidia cards can either.
I don't know about the upcoming RX 400 series card though, wiki only has 1 entry for that chip atm & it doesn't list the DP power.

 

[Orange Kid]

So.....you going to start all over again with the change from v102 to v136 and no more 106.88 credit WU's?

 

[Assimilator1]

What? what?

 

[waffleironhead]

Just checked in there. My 6950 can crunch again because they are using the modified fit app for everything now.
Points seem to be low now though.

 

[Assimilator1]

Cool you can use your 6950 again .

Orange Kid
Ok I see on the front page that it says the mod fit project will be the main 1, but I don't see it saying the MW v1.0x will be phased out, just the dedicated mod fit project, whatever he meant by that....

 

[Assimilator1]

ttt

 

[TennesseeTony]

Do they still have these tasks? For 106.88 points? All I get these days are 133.66 point tasks, and they behave quite differently, in that 1.) they are longer and 2.) they 'reset' the percentage back to zero several times in the progress bar.

 

[Assimilator1]

Err, I hope so! Lol, let me have a look.....

 

[TennesseeTony]

So.....you going to start all over again with the change from v102 to v136 and no more 106.88 credit WU's?

Up to version 1.43 now. Lot's of changes going on for Milkyway in the past year or so. That's a good thing, I suppose, but not good if you want to track performance of various cards. I'm sad that we don't get to see the newer cards on the older tasks. I don't think much would have changed though, 7970/280X would still rule them all.

 

[Assimilator1]

Yea I saw that the MW apps are up to v1.4x, & the MW N Body to v1.62.

What I couldn't find out for certain is if the v1.4x apps only have 133.66 credit WUs, but it looks very likely, I went through dozens of result pages from your top PC & those of the top MW cruncher & only found 133.66 credit WUs.
Well the change is annoying but at least it means everyone's crunching the same credit WUs , looks like I do have to start a new benchmark thread though!
Am I right in saying the N body app is for the CPU only?

Btw Tony your the No 5 top producer in MW! Congrats

 

[Assimilator1]

Hey Tony, looking at several pages of results on your computer 705185 it seems the WU times vary by quite a bit, mostly ranging from 110-125s (with a few outliers down to 105 & 130s). Are you running multiple WUs at once? (for better output).
I'm just trying to see if the new app has this variability (making it a poor benchmark), or whether it's just something on your host that causes the varying times.

 

[TennesseeTony]

4 per card. They seem variable one at a time too.

 

[Assimilator1]

Could you do 1/2 dozen individually & post the times?

 

[waffleironhead]

I just ran a bunch of MW wu's.
As stated earlier, they run different than before. It seems they break the work up into chunks(based on card Double precision maybe?) My HD6950 seems to restart each WU maybe 3-4 times before completion, and looking at the log file it says 3 chunks. Anyway... wu run times.
101.33 seconds Running 1 at a time
179.49/2= 89.74 seconds running 2
328.11/4= 82.0275 seconds running 4
612.44/8= 76.55 seconds running 8

 

[Assimilator1]

Well even if they break the work up, as long as the times are consistent with the same credit, that's what counts .
What does matter is that the old 106.88 credit WUs aren't available anymore!, I will have to start a new table & thread .
How many single WUs did you run to get that time of 101.3s? (quite an improvement by running multiple WUs btw).