Lower Geekbench score after installing a 7800GS

[Standard Def Steve]

Hey guys,

I've got a bit of a head scratcher here. I recently installed an EVGA 7800GS video card in my G4 MDD. It's a dual 1.25GHz running Leopard. The installation went remarkably smoothly, from the BIOS flash to the AGP tape mod. I didn't even have to install any drivers or software, it bonged and booted right up. Hardware accelerated Core Image was automatically enabled and everything. Everything's definitely running smoother at 1920x1200, compared to the old GF4 MX.

HOWEVER (nothing's ever 100% perfect, right?) my Geekbench score has decreased from 1073 to 954. I know Geekbench isn't a GPU benchmark, but I was expecting a small increase, since the CPU doesn't have to handle CI in software anymore. Any other G4/7800GS owners experience this?

Thanks!

 

[Unknown_K]

Was the old video card using Core Image, there is probably some CPU overhead with it enabled.

 

[Gorgonops]

Does Geekbench break down the tests in such a way that you can see any disproportionate hits on individual tests?

 

[Standard Def Steve]

Yes, and all of the tests are scoring lower--it's not just limited to a certain category. The multi-threaded tests are taking the biggest hit.

Was the old video card using Core Image, there is probably some CPU overhead with it enabled.

With the GF4 MX, Core Image was enabled in software mode, so there should have been more CPU overhead with the old video card. 

 

[hoi polloi]

Just run it again; you'll get a different number.  It's pretty bad software for benchmarking.

 

[Standard Def Steve]

It's actually been consistent. I ran the test 3 times and it scored 954-957 with the 7800GS. Pop the GF4 MX back in, and it's back up to ~1070.

Is there a program like CPU-Z for PPC Macs? I'd like to keep an eye on the CPU frequency during the benchmark run. I'm thinking that maybe it's throttling, either due to insufficient power, or from additional heat from the new GPU. I mean, we all know that the MDD PSU can barely handle a stock configuration, so the 7800 might be pushing it over the edge.

 

[Bolle]

They didn’t have variable clock speed in the MDD. It does not throttle, it just crashes when it’s getting too hot.

Possible the 7800 clogs up the bus with whatever it is doing which affects overall performance.

The bus in those G4 machines has been a bottleneck more or less since the beginning.

 

[hoi polloi]

Pull up activity monitor or run top, etc.  It should be pretty clear what's going on.  As I said, geekbench is not a good benchmarking tool.  Anything, and I mean any--thing, else running will lower the score.  Try safe booting, for example...

 

[nglevin]

If you really want to get to this level of identifying a performance issue in Leopard with your software, try downloading Xcode 3.1.4 from the download archives at developer.apple.com and look at the OpenGL Profiler (official documentation) that's installed in one of the /Developer/ subdirectories.

The getting started section and monitoring GPU use (to determine if an app is using a software fallback) chapters are the most relevant.

Pull up activity monitor or run top, etc.  It should be pretty clear what's going on.  As I said, geekbench is not a good benchmarking tool.  Anything, and I mean any--thing, else running will lower the score.

Oh yes, top is a good choice for high level CPU latency. Just a simple "top -o cpu" in Terminal should help.

I agree though that benchmark metrics are not as good as what you can confirm with your own eyeballs, it's a game of optimizing to make a graph go up and to the right removed from context of what that even means.

There's always some significant CPU overhead in calling out to an OpenGL driver. Nothing to obsess over.

 

[Standard Def Steve]

Well, I found out why it was scoring poorly. It was the progress bar animation, of all things! I tried running Geekbench minimized and got a score of 1095!

So I'm guessing that before, when it was running Core Image in software mode on the GF4, the animation must've been disabled? I really can't remember. Can anyone running 10.5.8 on a non hardware-accelerated card confirm this?