Personal computing discussed

Moderators: renee, morphine, SecretSquirrel

 
APWNH
Grand Gerbil Poohbah
Topic Author
Posts: 3200
Joined: Thu Oct 31, 2002 9:10 pm
Location: Boston MA
Contact:

3DMark11 Scores on this new system I built (AMD 7950)

Mon Jun 25, 2012 10:18 pm

So I had ordered a Gigabyte Windforce 7950 3GB about a week prior to the Nvidia 670's release.

I regretted that decision some, because the 670 consumes less power and was more performant. I actually went and bought a 670 and that's what's running in my machine (a SB 2500K currently at 4.0Ghz on a Z68 board). A secondary reason was I anticipated I was going to get one of those Korean overclockable 2560x1440 monitors to run at 120Hz and the AMD cards were unable to drive that display mode at the time. There have since been developments that have changed this (so both AMD and Nvidia cards can run those settings now).

I basically had the 7950 lying around on my desk till a friend of mine decided he wanted a new computer. Wow that's perfect, I have a great graphics card to start off that build right here, I said. Newegg was going to charge me a restocking fee to return the 7950 (which is certainly not defective as I will describe soon)...

So i slap together the new machine (on which I am typing now), it was a fantastically difficult and space-constrained Mini-ITX build, in the Silverstone SG06, with ASUS's P8Z77-I Deluxe Z77 board. I took a ton of pictures which I will hopefully remember to post-back here. (somebody post a reply to the thread if in 2 weeks it's not linked) With my new camera. So I have some cool pics to show people this awesome work I did. I'm not sure how coming up with smart ways to push pins and plugs in, in extremely cramped spaces, helps, but it sure makes me a more experienced system builder.

The 2600K is clocked to 4.5Ghz with the voltage set to auto in the UEFI/BIOS and it reads 1.32V in CPU-Z. A pretty average CPU, if I try to drop the voltage using offset setting it always BSOD's, but at the Auto setting it is rock stable in P95 and IBT.

I load up the drivers from ASUS... I have Virtu MVP installed. HyperFormance is turned on, because why not.

I run 3DMark11 with the CPU at 4.5 and the 7950 is STOCK. Result is over 10K 3DMarks. Now every single report I find anywhere for the 7950 card is in the 8000 range. but this machine at stock pumps out over 10000.

Something's not right! I say.

I go into CCC, unlock the overclocking options (since this massive cooler keeps the card's temps quite low, and I know it's a pretty good model to OC) and bump her up to 1050/1450 (you know how some people have their 7950s pushed to 1200/1700? Yeah.)

11687 3DMarks. I double-check to make sure it is actually 3DMark11 not Vantage. Oh it's not vantage, the demo's are all DX11-pretty and the runs are quick.

So what the hell's going on? I'm positive when I had this 7950 installed in my 2500K system (which I was running at 4.5Ghz also BTW) it never achieved higher than about 8400 points. But 10K? Surely the 4 additional threads couldn't be it. The Intel 3000 couldn't be it.

Did AMD make driver improvements? driver improvements or hacks that yield this large a change in performance? The framerates were there, the tests ran SMOOTH.

But what's bugging me is what could account for the dramatic difference. Is it the Z77 chipset somehow??

I'm going to run this benchmark again, this time with Virtu MVP off to see what changes. And I guess I'll grab a few more benchies to test with. But I gotta ship the box out soon, and I really want to find out just how it's ever possible for me to be pushing close to 12K on 3DMark11 when my other box (Gigabyte Windforce GTX 670 @stock, 2500K @4.0) can only hit 8641.

Here's one more somewhat related bit of strangeness. Does anyone else get the x-scale of the Similar Systems graph in the 3DMark11 results page all jumbled together? I can't read the numbers. Either way the orange bar for my 7950 result isn't visible. Perhaps it is just so far off on the right with so few other similar scores.

I had *switched* the 7950 *out* for the 670, but right now the 7950 is *trampling* my 670.

Any ideas?

Edit: Forgot to mention, all 3D11 benchmarks run at Performance setting, (1280x720) benchmark tests only.
Image

Here's a link to 3dmark results page for a run at stock clocks (with hyperformance enabled) http://3dmark.com/3dm11/3741784

For example, I'm looking at the results from my 670 machine and it can only hit 57.2FPS on GT3...
 
APWNH
Grand Gerbil Poohbah
Topic Author
Posts: 3200
Joined: Thu Oct 31, 2002 9:10 pm
Location: Boston MA
Contact:

Re: 3DMark11 Scores on this new system I built (AMD 7950)

Mon Jun 25, 2012 11:06 pm

I think I found the "culprit".

With Virtu MVP Hyperformance OFF the GT3 test dropped to 46.83 FPS from 68.82 FPS.

That's 31.9%.

That is more than significant.

These be head-scratching WTF numbers.

So I pose this question to you, members of the most intelligent tech hardware forum on the interwebs.

WTF? :o :wink:

What on earth were these folks able to discover that allow them to make some kind of software to "optimize the rendering pipeline" to the tune of THIRTY PERCENT?

At first I thought whoa! Since FPS is a bad unit of measure lets look at the average frame time:
68.82FPS = 14.53 ms/f
46.83FPS = 21.35 ms/f

100% * (21.35-14.53)/21.35 = 31.9%

Still, the point stands. AMD couldn't come up with ways of "optimizing the rendering pipeline" in their driver??

This really goes to show that software is just as important as hardware.

I have finally found some information that confirms my findings (that is, Lucid Virtu Hyperformance magically gives you large tens of percent more frames per second): http://www.tweaktown.com/articles/4651/ ... ndex5.html

Time to take a look if my Z68 board can support Virtu MVP...

TL;DR: Update your Lucid Virtu software. It contains black magic that boosts your FPS. And not in a fake way either, the FPS really was significantly higher and I could actually see the framerate was smoother as I watched the demo run.
Last edited by APWNH on Mon Jun 25, 2012 11:26 pm, edited 1 time in total.
 
APWNH
Grand Gerbil Poohbah
Topic Author
Posts: 3200
Joined: Thu Oct 31, 2002 9:10 pm
Location: Boston MA
Contact:

Re: 3DMark11 Scores on this new system I built (AMD 7950)

Mon Jun 25, 2012 11:20 pm

I'm gonna keep posting because there is more to discuss.

Why does Futuremark consider Hyperformance Enabled test results as invalid?

If the machine computes the same frames at a faster rate why would that disqualify it?

What they are doing here is no longer evaluating the raw game graphics performance of a system anymore.

I have updated scores using Hyperformance on my Z68 box: http://3dmark.com/3dm11/3741941

With the 670 the graphics score delta is much higher than even with the 7950. The Graphics Tests ran at an improved speed that was basically *absurd*. However the final score comes up short because the 2500K can't handle the last two physics-related tests as well.

7950 with Hyperformance On was no comparison to the 670 without it. But once it's enabled the 670 comes back on top.

This whole business of improving framerates by this much is still really fishy for me. I'm going to fire up BF3 with Hyperformance on and see what kind of FPS I pull. What makes no sense to me is why neither Nvidia nor AMD were able to perform these driver optimizations to squeeze out this extra performance. Even if it only had such a big impact on 3DMark scores alone, it would make a difference for marketing purposes. :-?
 
APWNH
Grand Gerbil Poohbah
Topic Author
Posts: 3200
Joined: Thu Oct 31, 2002 9:10 pm
Location: Boston MA
Contact:

Re: 3DMark11 Scores on this new system I built (AMD 7950)

Mon Jun 25, 2012 11:58 pm

And now I discover the real answer which is here

http://www.anandtech.com/show/5728/inte ... -biostar/3

And from Lucid:

http://lucidlogix.com/download/WP-Elimi ... 181011.pdf

So this makes some of the comments I made earlier quite stupid. And it is clear (to me) now why Futuremark invalidates framerate numbers produced when enabling these features.

I still really like what they're doing though. Really awesome software techniques. I still think that our GPU vendors should have worked in these features into their drivers, and in fact the reason why they didn't pursue it is exactly because of marketing reasons. Applying techniques that render the FPS measurement useless is not good at all from a performance metering perspective!
 
Rand
Gerbil
Posts: 92
Joined: Mon Dec 27, 2004 2:34 pm

Re: 3DMark11 Scores on this new system I built (AMD 7950)

Tue Jun 26, 2012 12:08 am

I don't recall the details specifically, and undoubtedly someone else will come by with more information but from what I recall the frame rate displayed with HyperFormance mode enabled doesn't accurately reflect the actual frames that are computed, it's throwing out frames that it deems unimportant (and we can only hope those frames genuinely aren't important) and not displaying them so it's doing considerably less work.

As far as I'm aware there is still some debate as to whether Hyperformance actually results in visibly smoother gameplay and more consistently low frame times. I'm not sure I've read much that's really investigated it in any real detail to prove things one way or another. Lucids track record with their promises isn't particularly great though, so I'd tend towards being wary of trusting Hyperformance without playing around with it quite a bit.

Either way, Hyperformance frame rate display doesn't track with frames that are actually displayed. What it says and what it shows isn't the same thing.
 
Rand
Gerbil
Posts: 92
Joined: Mon Dec 27, 2004 2:34 pm

Re: 3DMark11 Scores on this new system I built (AMD 7950)

Sat Jun 30, 2012 7:16 pm

I found a decent review of this at BeHardware http://www.behardware.com/articles/858-1/lucidlogix-virtu-mvp-in-action.html. Their opinion was decidedly negative, which is to say no real world benefits at all unless you're really looking for a cheat that can artificially increase your benchmark scores.
Worth reading though, they did a pretty good look into it.

Who is online

Users browsing this forum: No registered users and 1 guest
GZIP: On