[Irinikus]

The IR4 system doesn’t seem to be locked to half the monitors refresh rate, the Odyssey systems seem to be, as I’ve never seen ioQuake3 run @ more than 30 FPS on an odyssey system.

I will do some testing in this regard across all of the systems in the future.

[mapesdhs]

Thanks for that excellent summary!! The skybox thing reminds me of the transparency issue which makes Quake1 behave strangely on some systems.

If you ever get the chance btw, Quake2 in a CAVE is very cool, with motion tracked gun position.

Ian.

[jwhat]

Hi Irinikus,

I added a second RM11 (so now 2 x RM11) into my IR4 Chassis and this is now getting significantly better results than when I only had 1 x RM11:

Quake 2 (Q2) & Quake 3 (Q3) as per your setup.

For Q2:
setenv DECOUPLE_SWAPBUF y
timedemo 1
map demo1.dm2

For Q3:
\timedemo 1
\demo four


O350 with 24 CPU (8 x 800 MHz + 16 x 1GHZ) + 50 GB RAM / IR4 (with 2 x RM11) @ 1280x1024

Q2: 689 frames, 11.7 seconds: 59.1 fps (== refresh rate of monitor)
Q3: 1260 frames 30.8 seconds 40.9 fps 12.0/24.5/67.0/8.7 ms

From recollection, I think with single RM11 I was getting lower performance than V12 (Infinite Performance).

Back when I did first round of IR4 testing I said "I also ran the Quake 3 demo on O350 but results were funny so will redo later" ... the funny being that performance was well below what I would have expected. Adding 2nd RM11 seems to have addressed this.

So my results Q3 are:
- Octane (2 x 600) + V12 - 18 fps
- Fuel (800 HHZ) + V12 - 23.4 fps
- O350 (various CPU configs) + V12 - 26.2 fps
- O350 (24 CPU) + IR4 with 2 x RM11 - 40.9 fps

So now showing level of performance significantly better than V12 systems.

I now have 3 x RM11, still looking for extras to bring IR4 up to full complement...

Cheers from Oz,


jwhat/John

[Irinikus]

Cool!

Very nice!!!

You will also notice that besides getting better performance than a V12, you also get antialiasing and un-dithered textures!!!

The V12 seems to naturally employ allot of dithering, resulting in poor texturing!

It's interesting that my Onyx2 with 4xR14K@500MHz and 2xRM11 doesn't lag your O350 system by much. (I get 37.6 FPS)

Quake3's obviously not CPU-bound!!!

[jwhat]

Hi Irinikus,

yes definitely dependent on graphics / rendering capabilities and not CPU.

I expected that it would be very similar performance wise to your Onyx2, given the identical IR setups.

I was even thinking that Onyx2 might out perform CPU'ed up O350 as it has more tightly couple architecture than the O350 which is integrated via Numalink.

Have a good Christmas.

Cheers from Oz,


jwhat/John

[mapesdhs]

Q3 is probably fill rate limited. Remember though that none of these ports are remotely properly optimised for SGIs, id said they were 'quick & dirty' ports. Would be quite something if the same thing were written in Performer.

Ian.

[jwhat]

Hi Ian,

I hope you are well.

These performance discussions always seem to stir you ;-)

It will be interesting to see how adding RMs impacts the Q3 bench mark, as per you comment on fill rate.

If you do stumble across any extra RM11 please give me a PM.

Best wishes for Christmas.

Cheers from Oz,


jwhat/John

[jwhat]

Hi Irinikus and Ian,

here is my final update on Quake Benchmarks, having put additional 2 x RM11 into my IR4 chassis (so now running full complement of 4 x RM11 on main pipe):

Q2: 689 frames, 4.2 seconds: 164.9 fps (greater than refresh rate of monitor)
Q3: 1260 frames 21.2 seconds 59.3 fps 10.0/16.9/35.0/2.6 ms

HW config as per prior tests.

Total aside, when I went to boot up machine once I had added extra RM11s, the main chassis keep hanging at l1 boot. This set off a bit of panic on my side, as having got to IR4 nirvana, now the machine was failing... On opening up the chassis I saw I had not updated the DALLAS on this host with the cz7asm replacement. I put in one of these with new battery and machine come good again. No doubt failing DALLAS battery had resulted in corrupted NVRAM which was stopping L1 from starting up correctly. So my Onyx 350 gets to live another day...


Cheers from Oz,


jwhat/John.

[mapesdhs]

Thanks for that! Just curious, are all of these at 1280x1024? Also, are you using the DECOUPLE_SWAPBUF setting with the Q3 tests aswell? (I would)

I've not heard of the cz7asm thing, way behind on much of this now, so more reading up to do.

Ian.

[jwhat]

Hi Ian,

I did test with: "setenv DECOUPLE_SWAPBUF y"

The results surprised me as I got significant fps increase:

Q3: 1260 frames 17.5 seconds 72.0 fps 4.0/13.9/30.0/4.5 ms

I was surprised as previously this did not seem to make any difference (as also reported by IRINIKUS). Maybe this was because I was using uppercase 'Y', I suspect, rather than lowercase 'y', so user error....

And yes is at: 1280x1024

Here is link to DALLAS DS1742 Replacement Thread: https://forums.irixnet.org/thread-3360.html
I believe that GlitchWorks (who also make replacement DALLAS DS1687 for Octane) also now has one as well.
I am always posting on this as it has been such a life saver for all my O350s and Fuel.

EDIT: rerun on Fuel (800MHz + V12) + Octane (2 x 600 + V12) with lowercase 'y' and got small improvements:
- FUEL
-- Q3: 1260 frames 49.9 seconds 25.3 fps 19.0/39.6/81.0/9.0 ms
-- Q2: 689 frames, 13.8 secondsL 50,0 fps
- OCTANE
-- Q3: 1260 frames 51.8 seconds 24.3 fps 19.0/41.1/78.0/9.3 ms

Cheers from Oz,

jwhat/John