Removing the 5-6 jumper from jp18 allows me to boot with the TurboChip but CHKCPU reports L1 is in WT mode. And L2 isn’t detected, dang it.

• Make sure, that JP5 and JP6 are set to 1-2
• JP11: 2-3
• JP16: 1-2 & 4-5
• JP17: 1-2 & 5-6
• JP18: 1-2
• JP19: 1-2 & 3-4 & 5-6
• JP20: 1-2
• JP21: 3-4 (L1WB) or 2-3 (L1WT)
• JP22: 1-2 & 4-5
• check correct setting in BIOS feature setup

Got it! For some reason I had jp5 and jp6 set incorrectly, that and changing both L1 and L2 back to WB in the bios. thank you!

Seeing some better numbers now for L1 cache, slightly slower in L2 and main memory speeds.

Quake seems to prefer L2 set to “Auto”. Maybe nothing new here, just my current observation.

L2 WB = 15.9
L2 Auto = 16.4

This is up from 15.2 before the bios update to support L1 WB. Nice!

Dropped an S3 Trio32 in, upgraded to 2MB after photo, and now hitting 17.1 in Quake!

17fps - getting warmer.

On this board and chipset L2 cache in WT mode is better.
Actually only very few socket3 motherboards produce better performance with WB mode, but when it works is great.

Load up SDD and it squeezes a bit more out of the Trio32, 17.6 up from 17.1.

I know it’s kind of cheating but love to see VESA modes produce results!

pshipkov wrote on 2021-10-11, 21:17:

17fps - getting warmer.

On this board and chipset L2 cache in WT mode is better.
Actually only very few socket3 motherboards produce better performance with WB mode, but when it works is great.

Good to know! How far am I off the mark currently? Do these numbers line up with the 256k cache I have installed?

Trio32 stomps a mud hole on the CL5429 in Duke3D at 320x400.

CL5429 = 24fps
Trio32 = 34fps

I think the highest Quake 1 result with 486 CPU running at 160MHz is 18.8 fps, with Ark1000VL (default Phil's benchmark C).
After the 17'ies it get exponentially steeper.
You will most likely need 1024Kb L2 cache running at tightest BIOS timings for that.

Interesting DN3D result. Didn't expect such a difference.

pshipkov wrote on 2021-10-11, 21:47:

I think the highest Quake 1 result with 486 CPU running at 160MHz is 18.8 fps, with Ark1000VL (default Phil's benchmark C). Afte […]
Show full quote

I think the highest Quake 1 result with 486 CPU running at 160MHz is 18.8 fps, with Ark1000VL (default Phil's benchmark C).
After the 17'ies it get exponentially steeper.
You will most likely need 1024Kb L2 cache running at tightest BIOS timings for that.

Interesting DN3D result. Didn't expect such a difference.

Got it, I’ve got cache and an ARK on the way, I hope, 🤣. Really tempted to pull cache out of another build, must resist…..

Yah def interesting results with Duke3D but to be fair to the CL5429 the scores are within 1fps for both cards at 320x200, so if you just want to play Duke as fast as possible on this board, the CL is fine.

Happy to report the image quality in Windows 98 is also quite good compared to the CL. Unfortunately I’m still stuck in interlaced modes at 1024x768x15bpp. Don’t plan to spend a lot of time in Windows on this setup though.

Alright couldn’t resist, made a compromise and only needed to pull half the sram out of another build, the easy half to access. Now I’m running 1024k cache on the GX4 and pulling 17.9fps in Quake on the Trio32. So close to 18. Although using a VESA mode pushes it over, but we’ll leave it out for now.

Has anyone experimented with the mclk dos tool to try and squeeze an extra frame or two?

For whatever reason it doesn’t recognize my Trio. I’ve tried a couple of times to force a manual config but this ends with a freeze or corrupted screen.

Okay, so I was using the wrong syntax for the parameters. Clocked it up as high as I dared but only saw 0.1 fps difference in the 640x480 modes of Chris’s and PC Player’s benchmarks. Obviously within the range of error but it was consistent. No gain in Quake or Duke in the lower resolutions that I tried.

The pic shows 59MHz but I took it up to ~85Mhz, assuming it was even working.

for vga resolutions perf is cpu bound, so overclocking the video adapter wont make a big diff.

17.9 is a good number. you are above most mobos already, but with this setup you should be in the 18s.
so, what else is not on max ?

pshipkov wrote on 2021-10-12, 14:27:

for vga resolutions perf is cpu bound, so overclocking the video adapter wont make a big diff.

17.9 is a good number. you are above most mobos already, but with this setup you should be in the 18s.
so, what else is not on max ?

Are you sure? According to your graph, I think it’s yours, I’m right on the mark at 17.9 with this combo of hardware, specifically the Trio32.

I should check my own graph before speaking, so i stand corrected.
Somehow in my head S3 Trio VLB was in the 18s. 😀

pshipkov wrote on 2021-10-12, 16:19:

I should check my own graph before speaking, so i stand corrected.
Somehow in my head S3 Trio VLB was in the 18s. 😀

Haha, yah pretty crazy the ARK is basically a full frame ahead.

pshipkov wrote on 2021-10-11, 21:47:

Interesting DN3D result. Didn't expect such a difference.

No Wonder GD542x is suffering from 16bit Bus Connection, or multiplexed 32bit Bus resulting in Maximum Bus Speed of ~16Mb/s
A Fair Comparision would GD543x.

matze79 wrote on 2021-10-12, 18:55:

pshipkov wrote on 2021-10-11, 21:47:

Interesting DN3D result. Didn't expect such a difference.

No Wonder GD542x is suffering from 16bit Bus Connection, or multiplexed 32bit Bus resulting in Maximum Bus Speed of ~16Mb/s
A Fair Comparision would GD543x.

If I can get my hands on a VLB based GD543X I’ll give it a go!