Benchmarking tactics.

PCTwin

New member
I have been benchmarking for fun for quite a while now, but I usually stop on a 24/7 stable overclock and don’t take it any further. Comparing my scores to those on the Hwbot.org UK top 100 list usually makes me feel weak and timid, but it came to me the other day that these guys can’t possibly run these machines 24/7 on the likes of LN2 or DICE, so what’s going on here.

Tactics. Basically the cpu’s and gpu’s are overclocked to suicide levels just for the benchmarks and nothing more, if they survive, it’s a bonus. But where does that leave me. I have a two loop watercooled system with the gpu and cpu on separate loops, so how do I get into the UK top 100, never mind into the points.

Let’s look at the cpu tests first. Hwbot gives you cpu rankings for the following benchmarks.

CPU-Z

PCMark 2005

Pifast

SuperPi (1m)

SuperPi 32m

wPrime 1024m

wPrime 32m

Now let’s talk tactics. There are four short tests and three long tests on this list.

The three short tests are CPU-Z, Pifast, SuperPi (1m) and wPrime 32m.

On the short tests I don’t need a 24/7 stable system, nor even a long term stable system. It only needs to be stable for a short time, long enough to run the test and validate the score somehow. This usually takes the form of a screenshot or online validation.

Using a Q9550 as an example, I managed to get 4.552 GHz stable long enough for CPU-Z to validate online. That was it though, when I did a screen grab and tried to open paint, the system crashed. I had to stabilise the system to go online and grab the validation screen.



Although this didn’t bag me any points it did put me 25th in the UK Q9550 hardware ranks. This is not a stable overclock, it won’t run for more than a few minutes in my current setup, but it counts and that’s what matters.

Pifast, SuperPi and wPrime are a different matter. CPU-Z doesn’t make the cpu do much at all, but these programs need the cpu stable enough to do some work. For Pifast expect the benchmark to last between 20 to 30 seconds on a Q9550, SuperPi lasts between 10 to 30 secs and wPrime lasts about 8 to 10 secs. Pifast took a few tries at 4.505 to get a screenshot, twice I got below the 21 secs mark but couldn’t get a grab. But persistence pays dividends and eventually I got a screenshot at 21 secs.



At the moment this ranks 8th in the UK and bags me 3.7 hardware points. Again it’s not stable for long but, if you can get a screenshot, it counts. SuperPi reacted the same way, taking a few tries before I got a screenshot at 10.328 secs. Another thing I noticed at this stage is that SuperPi was giving me a higher score in Windows 7 RC1 64bit than it was in vista 32bit, so I had to consider this as well.



wPrime 32m on the other hand gave me a better score in Vista 32bit, although I had to drop the OC to 4.437. This gave me score of 8.636 secs which bagged me 4.8 points and ranks 9th in the UK.



That covers the short benchmarks, but the same tactics apply to the long tests as well.

The long benchmarks are SuperPi 32m, wPrime 1024 and PCMark05. My tactics here are the same as before. The OC only needs to be stable long enough to run the tests and verify them. SuperPi 32m was the most problematic of these, expect this to last between 11 minutes to 14 minutes on a Q9550. wPrime 1024 gives you your results in seconds and doesn’t convert the result to minutes and seconds until you enter your score on Hwbot.org, expect results in the region of 270 to 320 seconds range.

I managed to get SuperPi 32m to run at 4.351 MHz which gave me a result of 11mins: 12 secs: 860 msecs. This bagged me 1.3 hardware points and ranks 16th in the UK top 100.



wPrime 1024 again scored better in Vista than in Win7, with the cpu at 4.445MHz, finishing in 273.14 secs which equates to 4 mins 33 secs 840 msecs on Hwbot.org giving me a score of 5.1 points, ranked 7th in the UK top 100.



I have left PCMark05 to last because this is the longest of the cpu tests. Not only that but it also depends on the OC of the graphics system to pick up extra points if I overclock the graphics card as well. I may well take another shot at this test as I ran it without OC’ing the gpu, anyway at the moment I managed a score of 12470 which gives me 0.5 global points and ranks 23rd in the UK top 100. I didn’t score any hardware points, just global points. It also ranks 472nd global.



Just about anyone who's submitted to HWbot.org probably knows all this already, but I just thought I'd share this with you all.
 
+1. Excellent post.

I'm new over at hwbot and even newer here... but it strikes me that there should be a category that somehow takes O/S uptime into the equation. I mean, I can run around a 40% overclock... but only for 45 seconds or so... what good's that when I have to use someone else's machine to buy replacement parts for mine?

There ought to be something like a media recoding test, where you have to say... transcode a 200GB 1920x1080 x264 mkv file into a 9GB mpeg2 vob file... something that actually takes a bit of time to do...
 
Hwbot only uses Wprime version 1.55. After writing this post I had to re-run all those benches again and replace all my results.
 
It dosn't always follow that rigs are not stable with high clocks.

My P35 rig is stable 24/7 at 5.1ghz and can bench at 5.3ghz/5.5ghz depending on the bench.

The x58 in my sig will bench all day long at 4.8ghz with 3d benchies, and at 5ghz+ with other benchies.

Sometimes you just have to accept that a given CPU has a maximum clock that it will go to with a given method of cooling. The colder it is, the higher it will go.
 
What are you trying to gain, the people in the UK top 100 will have systems specifically for benching. They will buy many chips i.e celerons etc. And clock them, do as many tests as they can to achieve lots of points.

What I do is compare the results are only cooled by water or air, this way you get a better indication of how good your overclock is compared to everyone else in the world.
 
what do people gain from their games, enjoyment. I bench cause it's fun for me. I'm currently 29th in the UK with 190 points. As I only have access to water or air that's all I do compare against.
 
name='kitfit1' said:
It dosn't always follow that rigs are not stable with high clocks.

My P35 rig is stable 24/7 at 5.1ghz and can bench at 5.3ghz/5.5ghz depending on the bench.

The x58 in my sig will bench all day long at 4.8ghz with 3d benchies, and at 5ghz+ with other benchies.

Sometimes you just have to accept that a given CPU has a maximum clock that it will go to with a given method of cooling. The colder it is, the higher it will go.

I know rigs can be stable at high speeds, the E8600 is a great chip for that, but my p35 + E7400 is barely stable at 4.5gig on water. The speed of the ram in use at the time also helps. there's lots of contributing factors to a stable overclock, but thats for a different post.
 
I have a question, is it possible the go higher than 435 FBS on a P43 chipset motherboard?

I cant say as I have overclocked that many p43 mobos so I cant answer that.

Try starting a new thread in the overclocking section. you may get a better response there.
 
tnx anyway
smile.gif
 
To get back on the thread, I got a tactic for ya'll
tongue.gif


When doing your benching with the likes of SuperPI - even tho you need a cpuz in your screenshot, close it whilst you do the bench run and open it when you're going to take the screenie
wink.gif


It can shave those minor numbers off I've found.
 
best tactic for benchmarking is xp.

cpu scores are much higher (or lower;)) on xp vs later ones, only use win 7 for vantage and unigine which need the directx.

i have an old 40gb ide drive i use, yea boot is a bitch waiting all those 30 seconds but its well worth those extra points.

you are spot on about stability, it only needs to be stable enough to run the bench, after that could care less. i too dont have a lot of money to throw at hwbot and use it more to compare myself to others with my hardware who aren't using subzero cooling.

now my 9550 tops out around 4.6ghz but i wouldnt call it stable above 4.45ghz to tell you the truth but i know that its the same for everybody else submitting on there, its not a stability submission its about balls out performance.
 
XP I would use for most of the CPU tests, apart from pcmark05.

3dmark 01 and aquamark are definately xp but the other gpu benches run better in vista or 7 with the latest drivers.
 
You really got to bear in mind how much you're willing to strip down the services and extra crap in each of the OSes too. Windows 7 and to a lesser extent Vista, have a bunch of "helpful" services that get in the way of benching. You need to give them time to settle down after booting to before they show their colors.

But in fairness, when you like to use XP, you have little of these things to worry about.

EDIT: With wPrime, don't forget to set the thread use in the advanced settings. That way you won't be wondering for an hour like an idiot (points to self) why your a fraction of the fantastic scores everyone else in the world is getting.
 
idk, other than 06 and later i have found the older drivers give better results than the new ones which seem to be giving all the gains in dx10 and 11. this is the big issue rawz was on about in another thread, getting the right driver for the card for each bench is the big issue with the 3d tests because for a 5 or 10 year old benchmark the latest driver isnt always the greatest.

xp vs win7, i have tweaked both as much as i can and xp wins everytime in the things that will run on it. sometimes with the 3d tests the drivers might be better on 7 but its not as big a gain in points as what i lose from the cpu score.
 
I know it's a pita, but technically, from a performance pov, you really need to look at each 3d benchmark, the year of the card you're using, and the revision of the drivers.

Too far away from where your card came out, the driver makers are looking at performance for newer cards and compatibility for yours. Then you need to think about which directx is applicable in those dates.
 
Back
Top