The New 3DMark Benchmark - Testing Smartphones to Multi-GPU Gaming PCs

[thujone]

171514 icestorm
19102 cloud gate
7003 fire strike

3570 @ 4.3
7950 @ 1100/1600
13.2 beta 5 drivers

http://www.3dmark.com/3dm/114844?

 

[PrincessFrosty]

3Dmark - Taking all the nuances and preferences with video cards and inserting all that subjective preference into a single score which is meaningless since 1995

I'd like to know the weighting they give scores based on things like average frame inconsistency, for example when ATI get better frame rates but micro stutter causes the actual perceived frame rate to be lower how is this taken into account with the final scores - wait it's not? Right...

Raw horsepower in specific circumstances has been a bad way to measure game performance ever since gaming became more complex than a few polys on the screen, it leads to the kind of mindless dick waving we're seeing in this thread.

 

[Termie]

3Dmark - Taking all the nuances and preferences with video cards and inserting all that subjective preference into a single score which is meaningless since 1995

I'd like to know the weighting they give scores based on things like average frame inconsistency, for example when ATI get better frame rates but micro stutter causes the actual perceived frame rate to be lower how is this taken into account with the final scores - wait it's not? Right...

Raw horsepower in specific circumstances has been a bad way to measure game performance ever since gaming became more complex than a few polys on the screen, it leads to the kind of mindless dick waving we're seeing in this thread.

You really don't get it, do you?

When different systems and different uses get the same score with the same videocard, the results show exactly the opposite of what you're saying. It's completely objective. This is not an nvidia versus ati thread, and if you try to make it one I'll ask the mods to lock it. It's a GTX670 vs. GTX670 thread, or HD7870 vs. HD7870 thread, or Intel HD4000 versus HD4000 thread.

3dMark is by far the best tool I've come across in 7 years of system building to diagnose CPU and GPU problems and test overclocking.

 

[hyrule4927]

Some results from my PC:

Noticed a weird whine coming from my GPU during Ice Storm, but it seems to only occur at 600+ FPS, so I guess I don't need to be overly concerned. :whiste:

Also, I'm assuming a driver update might bring up my score a little bit, but I've been sticking with 12.10 because every more recent driver hurts my Folding@Home performance.

 

[Termie]

Some results from my PC:

...

Noticed a weird whine coming from my GPU during Ice Storm, but it seems to only occur at 600+ FPS, so I guess I don't need to be overly concerned. :whiste:

..

Yup, I get some serious coil whine in both my system with that first test as well - the FPS is just too high, as it's made for phones and tablets, not computers. That being said, it would look pretty impressive running on a phone!

 

[Termie]

171514 icestorm
19102 cloud gate
7003 fire strike

3570 @ 4.3
7950 @ 1100/1600
13.2 beta 5 drivers

http://www.3dmark.com/3dm/114844?

Some results from my PC:

This is interesting...two systems running an HD7950, the first at 1100/1600, the second at 1035/1250, scoring 25% apart in the graphics score. That is way more than the theoretical bump provided by the 6% delta in core clock. Could 3dMark be that sensitive to memory clocks, where there is a 28% difference?

I might have to play around with that a bit...

 

[hyrule4927]

When I have a little more time on my hands I can bump the core clock up to around 1200 and rebench (gotten it up there before, but I've reinstalled Winows since then and have to figure out the right voltage again). My memory can barely even do 1350, so not much room for experimentation there.

Updating my drivers might make a difference too.

 

[thujone]

yeah... memory clocks are what got me over 7k in fire strike. i usually leave them clocked low... started increasing them just to see if i could squeak out a few extra points to clear 7000

 

[rgallant]

Yup, I get some serious coil whine in both my system with that first test as well - the FPS is just too high, as it's made for phones and tablets, not computers. That being said, it would look pretty impressive running on a phone!

first time ever hearing coil whine on my 580's

 

[PrincessFrosty]

You really don't get it, do you?

When different systems and different uses get the same score with the same videocard, the results show exactly the opposite of what you're saying. It's completely objective. This is not an nvidia versus ati thread, and if you try to make it one I'll ask the mods to lock it. It's a GTX670 vs. GTX670 thread, or HD7870 vs. HD7870 thread, or Intel HD4000 versus HD4000 thread.

3dMark is by far the best tool I've come across in 7 years of system building to diagnose CPU and GPU problems and test overclocking.

I did no such thing. I gave a reason why I think 3DMark is useless and then I gave an example to back this up, a single score might be "objective" but it only reflects a narrow set of properties of the card in a narrow set of circumstances.

You're not a mod and you're in no position to threaten me with action, stop pretending to be.

CPU and GPU problems have always best been found by stressing the components with applications such as Prime95, SuperPi and Furmark, all of which stress the relevant components more than 3DMark does.

 

[Termie]

3Dmark - Taking all the nuances and preferences with video cards and inserting all that subjective preference into a single score which is meaningless since 1995

I'd like to know the weighting they give scores based on things like average frame inconsistency, for example when ATI get better frame rates but micro stutter causes the actual perceived frame rate to be lower how is this taken into account with the final scores - wait it's not? Right...

Raw horsepower in specific circumstances has been a bad way to measure game performance ever since gaming became more complex than a few polys on the screen, it leads to the kind of mindless dick waving we're seeing in this thread.

I did no such thing. I gave a reason why I think 3DMark is useless and then I gave an example to back this up, a single score might be "objective" but it only reflects a narrow set of properties of the card in a narrow set of circumstances.

You're not a mod and you're in no position to threaten me with action, stop pretending to be.

CPU and GPU problems have always best been found by stressing the components with applications such as Prime95, SuperPi and Furmark, all of which stress the relevant components more than 3DMark does.

Fair enough - don't use 3dMark if you don't like it. Furmark is far too inexact to actually determine overclock scaling - it's good only for determining overclock limits. And I don't think anyone's seriously used SuperPi for benchmarking since 2005, when single-core scores actually had some relevance. In fact, it's the perfect example of the kind of specific and unrepresentative test that you criticize.

And next time, try to use more graceful terms than "mindless dick waving" for legitimate benchmarking that you have technical issues with. You thread crapped, and regardless of whether I'm a mod, I've been around here far longer than you and know that they don't tolerate that kind of commentary.

 

[guskline]

Termie: Please check your PM.

 

[Mir96TA]

I didn't realize I was that slow!

 

[Mir96TA]

FireStrike

I guess I have some room to O.C

 

[Rhoxed]

Rig #1 in sig
http://www.3dmark.com/3dm/151038
6516

 

[An00bis]

Lame.

Disgusting, I'm really appalled by 3dmark's shitty visuals and I feel TRAUMATIZED just from thinking about the fact that I'll have to watch this crappy benchmark for 1 hour... (duke nukem back in 1996 looked better and ran better than all 3dmarks put together including this one) oh well, I hope that the buccaneers will help me in the following weeks if the find a FIX like they did for 3dmark11, altough 3dmark did let you skip those retarded demos.

 

[Stu @ MSD]

It's a business mate, not a charity. They exist to make money like the rest of us. Why not register and run your benches in 5 mins flat?

As for duke nukes looking better? Love certainly is blind. Lol

 

[Rikard]

This is interesting...two systems running an HD7950, the first at 1100/1600, the second at 1035/1250, scoring 25% apart in the graphics score. That is way more than the theoretical bump provided by the 6% delta in core clock. Could 3dMark be that sensitive to memory clocks, where there is a 28% difference?

I might have to play around with that a bit...

I finally got around to try this out a bit. With my 24/7 OC 1100/1250 MHz on my HD7950 I score 6627 (7679) in Firestorm (Graphics). Increasing only the clock to 1150 MHz I get 6804 (7929), i.e. a 3% increase in both scores for an increase in clock by 5%. If I in addition increase the memory by 16% to 1150/1450 I get 6946 (8113), i.e., a 5% (6%) increase in the score compared to 1100/1250 MHz. I did not check how the CPU speed would scale yet.

One weird thing happened with that going to 1175/1450 MHz that is never causing a problem for me elsewhere caused a driver failure. The tests I ran after that scored much lower, but I got back the higher and more reasonable scores after a reboot. Then the driver failure happened at much lower OC, so I am not sure if this is even related to OC or if there is something else that is messing things up every now and then. Since I never saw this in Heaven for example I am tempted to say it is a bug in 3DMark. This is with Cat 13.1 stable.

 

[Termie]

I finally got around to try this out a bit. With my 24/7 OC 1100/1250 MHz on my HD7950 I score 6627 (7679) in Firestorm (Graphics). Increasing only the clock to 1150 MHz I get 6804 (7929), i.e. a 3% increase in both scores for an increase in clock by 5%. If I in addition increase the memory by 16% to 1150/1450 I get 6946 (8113), i.e., a 5% (6%) increase in the score compared to 1100/1250 MHz. I did not check how the CPU speed would scale yet.

One weird thing happened with that going to 1175/1450 MHz that is never causing a problem for me elsewhere caused a driver failure. The tests I ran after that scored much lower, but I got back the higher and more reasonable scores after a reboot. Then the driver failure happened at much lower OC, so I am not sure if this is even related to OC or if there is something else that is messing things up every now and then. Since I never saw this in Heaven for example I am tempted to say it is a bug in 3DMark. This is with Cat 13.1 stable.

Thanks for the detailed testing. Your scaling is very much what I'd expect. The only thing I'm surprised about is that your starting Graphics score of 7679 at 1100/1250 is so much higher than the user who earlier posted an HD7950 graphics score of 6404 with 1035/1250 clocks. Your score just builds from there, so it wasn't simply the huge memory overclock you had that was setting your card apart.

Very puzzling - typically 3dMark is very consistent in this regard. Maybe we'll need to see more HD7950 and HD7970 scores to see what's happening here.

As for the driver crash, it could be a bug, or it could be that 3dMark is presenting an even higher load than other programs. I know that in the past, I would always use Heaven to test my overclocks, because it would bring down my system pretty quickly even where 3dMark11 would sail right through an overclock. That's why I appreciate that the new 3dMark Firestrike test really pushes the limits with DX11 effects. It may look a little wonky, but it's a serious load and therefore a good test of OCs.

 

[Rhoxed]

score i posted above


1100/1450 = 7399 Graphics score

 

[Termie]

I finally got around to try this out a bit. With my 24/7 OC 1100/1250 MHz on my HD7950 I score 6627 (7679) in Firestorm (Graphics). Increasing only the clock to 1150 MHz I get 6804 (7929), i.e. a 3% increase in both scores for an increase in clock by 5%. If I in addition increase the memory by 16% to 1150/1450 I get 6946 (8113), i.e., a 5% (6%) increase in the score compared to 1100/1250 MHz. I did not check how the CPU speed would scale yet.

...

I'm wondering if something might be inflating the scores - perhaps Virtu MVP? User Rhoxed scores 7399 with nearly the same clocks, and user Hyrule was slightly below that with 1035/1250 clocks.

Not saying you meant to do this - I'm just trying to figure out if 3dMark might be inconsistent or affected by other software. Your scaling is exactly what I expect though - about 3% for every 5% core, and 1% for every 3% memory. Hence the reason Rhoxed is about 15% faster than hyrule.

1100/1450 = 7399 Graphics score

Some results from my PC:

...

Just FYI, I'll post my GTX670 and HD7870 scores at different clocks to give a few more examples of scaling as soon as I can.

 

[Rhoxed]

Pushed card from 1100/1450 to 1150/1550
so 50/100 oc



1150/1550 = 7733

going to run at hyrules clocks.

EDIT: - 3 different driver versions could explain the scores.... i will try the betas

EDIT 2: Tried the Beta 5's no change from the driver I was previously using at 1100/1450 clocks

EDIT 3: At clocks of 1035/1250 i get a graphics score of 6940

1035/1250 = 6940
1100/1450 = 7399
1150/1550 = 7733

 

[Termie]

Ok, got some Fire Strike Graphics Score numbers for my GTX670:

1084/6000 (reference GTX670 clocks): 6326
1189/6200 (default clocks of my GTX670 FTW): 6799
7.5% increase for a 9.6% core clock increase and 3.3% memory clock increase

1200/6500: 6957
2.3% increase for a ~1% increase in core clock and 4.8% memory clock increase

1200/6700: 7004
0.66% for a 3% memory clock increase

1215/6700: 7053
0.7% for a 1.25% core clock increase

1215/6800: 7072
0.25% for a 1.5% memory clock increase

Conclusion - it seems 3dMark helps fine tune (at least for purposes of 3dMark) where the right combo of core and memory clocks is. The initial memory OC'ing seems to have a very good effect, and then really trails off. Ideally I would have just tested core clocks and various intervals, and started over with memory clocks. Maybe later!

 

[YBS1]

Weeeeee!!

 

[kami]

3770k @ 4.4GHz, 16GB DDR3@1600, single GTX 680 at stock speeds

Ice Storm: 170,821
Cloud Gate: 24,245
Fire Strike: 6,553