Very interesting! Could you please post the scores @ 6x100 as well? Then we can really see what the 130mhz fsb does.

I could run them at my own system but it compares much better when it is done at the same computer.

meljor wrote:

Very interesting! Could you please post the scores @ 6x100 as well? Then we can really see what the 130mhz fsb does.

I could run them at my own system but it compares much better when it is done at the same computer.

Yes of course. I have in fact already done just that. 😀

This is with the motherboard cache activated acting like a L3 cache.
6*100 MHz, the memory at 100 MHz CL2.

Super Pi 1m

Sisoft Sandra 99

3dmark 99

Thanx!

The overall improvement isn`t as big as i thought it would be but it is there: 5% on the cpu score, 10% on the 3dmarks.

A very big 30% on the memory is very nice but the least useful i think? On an athlon system you can measure huge differences in memory speed going from sdram to ddr but in actual games the speed increase was barely there....

More homework for you: The max fsb overclock with cache enabled? try 5,5x 112 for 616mhz, i think it will be the best stable setting. 5x120 will be too high for the cache i think.

meljor wrote:

Thanx! […]
Show full quote

Thanx!

The overall improvement isn`t as big as i thought it would be but it is there: 5% on the cpu score, 10% on the 3dmarks.

A very big 30% on the memory is very nice but the least useful i think? On an athlon system you can measure huge differences in memory speed going from sdram to ddr but in actual games the speed increase was barely there....

More homework for you: The max fsb overclock with cache enabled? try 5,5x 112 for 616mhz, i think it will be the best stable setting. 5x120 will be too high for the cache i think.

Even 105 MHz FSB is not 100% stable with the on board cache activated together with the fastest timings. With slower timings and a P133 the board managed 1.5x125=187.5 Mhz with onboard cache activated.

616 Mhz would probably need 2.4V which I think is a bit too much for these CPUs. (This CPU is stable at 5*95 = 570 Mhz with only 2.0V but needs 2.3V for 100% stability at 6*100 Mhz) If 5*120 with on board cache is stable with more relaxed timings it might perform better than 4.5*130 without onboard cache but if I upgrade the memory I think the 4.5x130 MHz setting will be faster again?

Nice work on getting the FSB up but still it is disappointing that it takes that much just to scrape up that much memory side when EDO in theory was capable of over 300mb/s at stock. S5/S7/SS7 memory controllers leave a lot to be desired 😒

This is why people should be thankful that modern cpus have integrated memory controllers that don't bottleneck the system to a crawl.

this is not too impressive, maybe you are not very good at optimizing your system. i got 5:45 with the ga-5ax and k6-2+@100*5.5, onboard cache disabled. enabling onboard cache gives only 4 seconds improvement.

and i have put the ga-5ax in stockpile due to bad agp compatibility experience, and the mvp3 board with 2mb cache being faster:
p55c@100*3: 6:47
k6-2+@100*5.5: 5:17

there are also a few things that i won't risk trying with my collections:
overvoltaging the k6-2+/3+, since amd datasheet lists the absolute limit at only 2.2v, which would surely cause damage, so even 2.1v seems too close to that.
and pushing pci clock over 35mhz, which may risk damaging some pci/agp devices that are less tolerant. 105fsb may be tryable, but my mvp3 board has less flexible fsb options.

noshutdown wrote:

this is not too impressive, maybe you are not very good at optimizing your system. i got 5:45 with the ga-5ax and k6-2+@100*5.5, […]
Show full quote

this is not too impressive, maybe you are not very good at optimizing your system. i got 5:45 with the ga-5ax and k6-2+@100*5.5, onboard cache disabled. enabling onboard cache gives only 4 seconds improvement.

and i have put the ga-5ax in stockpile due to bad agp compatibility experience, and the mvp3 board with 2mb cache being faster:
p55c@100*3: 6:47
k6-2+@100*5.5: 5:17

there are also a few things that i won't risk trying with my collections:
overvoltaging the k6-2+/3+, since amd datasheet lists the absolute limit at only 2.2v, which would surely cause damage, so even 2.1v seems too close to that.
and pushing pci clock over 35mhz, which may risk damaging some pci/agp devices that are less tolerant. 105fsb may be tryable, but my mvp3 board has less flexible fsb options.

Were those Super Pi scores run under Windows 98? Windows XP often improve the scores alot.
But no there are no OS optimizing at all and I did only run the BIOS settings I thought were the fastest, I did not test if other settings were faster.
'The main thing was that a lower speed setting without cache is 10% faster in Super Pi because of incresed FSB with otherwise exactly the same settings and hardware.

I have lots of spare parts and the CPU is not a good clocker anyway. The only thing I care if it gets damaged is the board in it self and Im pretty sure it wont as I keep my eyes on the warm bits.
So far even the old Samsung disk I use because its already on its last legs (and seem to manage overclocked PCI bus without data corruption) survives.

noshutdown wrote:

this is not too impressive, maybe you are not very good at optimizing your system.

Let's keep it friendly please 😀

Since all settings are the same the absolute performance doesn`t matter much, it is all about the differences so i think it is a nice test.....

Btw, agp stability is all about the drivers, brands of the agp cards, bios revisons and revisions of the boards itself so it just depends how good it is on each board. (my via boards are crap for example compared to my Asus p5a`s, except for the aopen ax59pro with via chipset)

I have replaced the 128 MB module with a 256 MB module. The first one I tried, a double sided module with chips marked M-tec did not want to run at 130 MHz CL2 (or it did run and was seemingly stable). After a few hours of testing, register errors showed at the next restart. I replaced the M-tec module with a double sided Infineon module and let the system loop GL-Quake during the whole night. It seems rock stable!

I did also benchmark GL-Quake but the results are not very useful since I diddnt bother to inactivate v-sync.

K6-2+ @585 Mhz, GA-5AX, 256MB SDRAM @130 MHz CL2, Geforce2 MX400, Windows 98SE.
GL Quake 1024*768 v-sync on: 16bit - 60.0 FPS, 32bit - 57.3 FPS

I will run some more benchmarks later today or tomorrow.

I did run the same benchmarks as I did with 128MB to see if the increased amount of memory (or different module) had brought any benefits, it had.
Exactly the same settings just a different module with the double amount of memory.

Sisoft Sandra 99: No gain but we can see that the system has become more bloated with more processes running in the background.

3dmark 99: No gain in 3d score, some gain in CPU score.

Super Pi: Here we see a huge gain, a bit of a surprise after not seeing any gain in memory bandwidth.

I did also run a quick Quake 3 demo001 bench. The Geforce2 MX400 seem to be a good match for the CPU.
640*480: 1346 frames, 29.2 seconds: 46.0 fps
1024*768: 1346 frames, 30.7 seconds 43.8 fps

Seems like I really did sell off one of the olden goldies back in 2004 (I needed the money back then).
Can't blame it. I had a TNT2-Vanta installed in the system, and not a gForce2 card.
I thought that it was worthless back then, as I used it for a fileserver and then got my first usb HDD.
And yes. The GA-5AX was by far the most stable board that I have ever owned.
Slow, yes. On the other hand, it could run for day's at times without any reboot, running Win98/XP.
The cpu was an old K6-2 running at 500MhZ and the GFX was a TNT2-Vanta AGP 16mb.
It had a SB-Pci 128 card, and a cd-rom burner from Acer. All with 80-pin cables on the HDD controller.

I overclocked my factory underclocked Geforce 2 MX400 from 260 MHz core/mem to 290 MHz core/mem.
GL-Quake now manages to keep a constant 60 FPS (with V-sync on) at 1024*768 32bit color 😀. No gain in Quake 3, it seems 100% CPU limited at this point even at 1024*768.
Some OS tweaking gave the system ~100 more CPU points in 3dmark99 and the GPU overclock gave ~100 more 3D points, nothing worth screenshotting.

I think this is as fast this combination of CPU, memory, motherboard and video card will get in Windows 98.
I will source a K6-III+ too see if it brings some improvement, but from the benchmarks I have seen the performance gain at the same speed isnt more than 1-5% (in games).

Did the K6-3+ arrive?
Did it set the board on fire?
Did it bring some benefit?

Yes
No
Yes

Here are some benchmarks with the K6-3+ at 5*120 = 600 MHz at 2.0V. All other settings should be as before.

Nothing much has happend in Sandra. The memory is a bit slower at 120 MHz FSB vs 130 MHz FSB, hardly a surprise.

In 3dmark 99 we see decent gains.

SuperPi 1m see a huge gain, this is a new personal best with a socket 7 system running Windows 98.

The FPS in GL-Quake did not improve but Quake 3 gained 4.5% at 1024*768.

The quirks of old systems are hard to understand sometimes.

I am using the K6-3+ with a single double sided PC133 Infineon 256MB memory module running at 120 MHz CL2 and the performance is good.
If I add another identical stick the performance takes a nosdive, the boards cache is inactivated so thats not the issue, the CPU L2 caches the memory.
I lose ~20% in CPU tests and ~15% memory performance going from one stick to two. Perhaps this board would be more happy with a single 512MB stick.
Or perhaps the second stick is just bad. More testing is needed 😀