Quake III performance across a range of SGI's

[Irinikus]

Here are the results that I recorded when benchmarking ioQuake3 on various SGI's in my collection, to gauge their relative performance and give others some idea as to what performance to expect when running this game in their machines!

Bring down the console in ioQuake3 and type in the following commands to run the benchmark:

\timedemo 1
\demo four

I used the following settings:

Here are the hinv's and results of the systems tested:

Onyx2:

Tezro:

Octane2:

Indigo2 MAX IMPACT:

O2:

Here's their relative performance represented graphically:

 

[Jacques]

I would dread running this on the R5200 300Mhz O2, it would be a slideshow. Not surprising though given the O2's low fill rate.

 

[Irinikus]

Try it, it would be interesting to see how it compares!

 

[Jacques]

I'll see if I can get it going tomorrow.

 

[Elf]

Edit: With apologies, I no longer wish to have involvement with SGUG or SGI communities in general,
and have also chosen to remove all of my content. Many things have changed since I co-founded, named, and ultimately
then left SGUG. There are many good people around, to whom I apologize for frustrating by removing these things, and
also many petty people that over the years whittled down both the enjoyment as well as sense of obligation I used to
feel to anyone else regarding what was ultimately just a hobby. Unfortunately one of the latter now writes the rules
and so it is time for me to take my things and go.

This message will replace all of my previous forum posts because deleting threads that I started would have removed
other peoples' posts.

 

[fraserN64]

I'm really curious how applicable Quake framerates are to other graphics workloads? While the numbers may differ would the overall workstation distribution follow a similar curve?

I'm most surprised by the Onyx2 outperforming the Tezro. Tezro prices are always high and everyone seems to want one but it got beat by a good margin by the Onyx2. It looks like the Onyx2 had the last/best Graphics Option IR4 which helped.

The O2 was used in TV studios, so it's "optimal" resolution is probably 640x480, anything more and it's probably bus limited.

 

[Irinikus]

These stats may explain the O2's poor performance here:

Pixel fill rateof 33M/sec for the O2, vs. 119M/sec for MAX IMPACT!; Triangle rate of 400K/sec for O2 vs. 1.6M/sec for the MAX IMPACT!

The O2's pixel fill rate is particularly poor! (IR4 has a pixel fill rate of 10,2 billion pixels per second!)

I will carry out some comparisons of output quality in the future. IR4 is not only much faster than the Odyssey, it's output quality is also far superior!

In the mean time, here's a quick comparison between the outputs of the Onyx2 and Tezro:

Here's the output from the Onyx2:

Here's the output from the Tezro:

Take note of the artefacts in the "larva" beneath the weapon in the Tezro's rendering!

Also notice the jaggies on the weapon you're holding in the Tezro's rendering!

 

[Jacques]

O2 R5200 @ 300Mhz 384Mhz with the settings above.

3.3fps
1260 frames 378.2 seconds

 

[Irinikus]

O2 R5200 @ 300Mhz 384Mhz with the settings above.

3.3fps
1260 frames 378.2 seconds

Thanks very much man!!! That doubly confirms the fact that Quake III isn't CPU dependant!

 

[Irinikus]

Anyone else running a MIPS machine with an interesting graphics setup, please run the test and post your results, as well as your machines hinv! Once more results come in I'll update my graph accordingly!

 

[Jacques]

I was actually surprised with the R5k performance, given the much more powerful R12k @400mhz only does 0.2 fps more.

For kicks, reran the benchmark with the same settings at a more 'age' appropriate 640x480. The result is 7.5fps.

I remember playing this on my K6-2 450 and the max I'd run it at was 800x600, anything above that started to bog down.

 

[Irinikus]

I was actually surprised with the R5k performance, given the much more powerful R12k @400mhz only does 0.2 fps more.

For kicks, reran the benchmark with the same settings at a more 'age' appropriate 640x480. The result is 7.5fps.

I remember playing this on my K6-2 450 and the max I'd run it at was 800x600, anything above that started to bog down.

Cool! Try it on your Octane and your other O2 as well!

I'm especially interested to see how the various graphics options available for the Octane and Octane2 perform!

 

[Jacques]

I've sold the Octane, it's boxed up waiting to go to it's new owner

For fun I compiled Quake3 for the RaspberryPi. Running the benchmark with exactly the same settings I get 76.5 fps.

I will pop my R10K 250 chassis into my O2 and test that, but I'm guessing it will achieve about 3.4fps.

Edit - Raspi 3B+ !!

I have to say that as nice as Q3 is, I infinitely prefer Q1 and Q2 for playing. I can happily spend a whole afternoon with Q2 and a bar of dark chocolate and several cups of coffee!

 

[Irinikus]

I've sold the Octane, it's boxed up waiting to go to it's new owner

For fun I compiled Quake3 for the RaspberryPi. Running the benchmark with exactly the same settings I get 76.5 fps.

I will pop my R10K 250 chassis into my O2 and test that, but I'm guessing it will achieve about 3.4fps.

Edit - Raspi 3B+ !!

WoW!!! That shows just how far we've advanced in the last few years!!!

 

[stormy]

There are some horrible driver issues with ioQuake3 and the Odyssey OpenGL driver. Someone on Discord (sorry I forget who, perhaps drmadison?) Actually found the issue and found a possible fix for the ioQuake3 engine. But that is as much as I know. One major performance killer is the skybox rendering which causes a 'hall of mirrors' effect which tanks the fps. I think if these issues were fixed Odysseys performance would be far superior.

Edit: Also @Irinikus what monitor refresh rate is on your Onyx? The SGis also seem to have a locked vsync, causing the FPS to never go over your monitor refresh rate. This is document in Ian Maplesons Quake benchmarks. source: http://www.sgidepot.co.uk/quake2bench.html
You might push over 100 with a decoupled swapbuffer on the Onyx.

 

[Irinikus]

After speaking to Ian, I tried the decoupled swapbuffer on the Onyx, and it made no difference.

The demo used in this test seems to be rather taxing and it never got close to the screens 60Hz refresh rate during the test.

 

[rbz]

Anyone else running a MIPS machine with an interesting graphics setup, please run the test and post your results, as well as your machines hinv! Once more results come in I'll update my graph accordingly!

Is there a quick how-to on how to get Quake3 installed? I know ioQuake3 comes from the old Nekoware stuff, but I have never installed anything from Nekoware so just curious if there is a quickstart to get this running. Q1 and Q2 setup info would be cool too.

Edit: NVM gonna give this a go https://wiki.preterhuman.net/Quake3_(neko_mips4)

 

[Irinikus]

Here's the latest chart including the performance of my latest SGI:

 

[Elf]

Edit: With apologies, I no longer wish to have involvement with SGUG or SGI communities in general,
and have also chosen to remove all of my content. Many things have changed since I co-founded, named, and ultimately
then left SGUG. There are many good people around, to whom I apologize for frustrating by removing these things, and
also many petty people that over the years whittled down both the enjoyment as well as sense of obligation I used to
feel to anyone else regarding what was ultimately just a hobby. Unfortunately one of the latter now writes the rules
and so it is time for me to take my things and go.

This message will replace all of my previous forum posts because deleting threads that I started would have removed
other peoples' posts.

 

[Irinikus]

RISC chips are roughly three times as efficient as CISC, when it comes to performance per Hz (especially back then). Quake III's performance is also more linked to GPU performance.