RacoonRider wrote:

vetz, I just contributed benchmark results for all my Socket 5/7 motherboards to both your and Phil's benchmark. I wonder why no one else is interested enough to run a couple of tests.

Maybe this is because of `intrests`, time, and sense. Most people only want to have to build a system and want to playing games on it (they dont care about 5% more performance..)

It just take a lot free of spare time to help with these benchmarks.. For myself, iam actually not in the mood now to go test my boards.. I already have to have some patience to get things done here..
I have still a lot of things to repair and solder here.. But my biggest problem is waiting when i can buy a professional weller de-solder station for about 1200 euros.. And i dont only buy it for my repairs.. I can use it for other de-solder projects too.

This isn't unusual when you have very set requirements. For example all my Matrox cards are AGP. It's much easier to just benchmark what you have. One of the reasons I wanted to make sure that the benchmark for the VGA database is as easy to use as possible. People don't have the time to figure it all out themselves 😀

Mau1wurf1977 wrote:

This isn't unusual when you have very set requirements. For example all my Matrox cards are AGP. It's much easier to just benchmark what you have. One of the reasons I wanted to make sure that the benchmark for the VGA database is as easy to use as possible. People don't have the time to figure it all out themselves 😀

For a comparison like this to be valid and still get help from other users there has to be some requirement rules to keep data integrity. Results would be all over the place if all types of graphics cards were accepted. You would not know what caused the fluctuations. There is a compromise that has to be made between data integrity and accessibility for the benchmark itself. I wanted to avoid just going with one type of card like others have done in the past. This is the reason I thought this through and I benched all the Matrox PCI cards before starting. Maybe I could've done this with the S3 cards as well, but here I've seen too much fluctuations between the different models (even within the same chip-series just another manufacturer) that it's pointless. The Matrox PCI cards are both quite common and cheap/easy to acquire so they made for the best compromise I could come up with. I honestly don't think the requirements are very strict (feipoa for instance had a requirement for a G200 PCI 16MB for his benches)

Each benchmark methodoligy has its place though. The philisophy I undertook was to compare CPU performance while trying as much as possible not to let the surrounding hardware interfere with the results. I do still have some things to re-test in this regard. While the results with a throw whatever hardware you have at it approach are certainly interesting, they do not narrow in on the cause of dissimilar results between platforms, i.e. is it the motherboard, RAM speed, chipset, graphics card, cache, etc which caused one system to perform better/worse. This approach is still extremely useful because it shows what scrores are capable with various combinations of hardware and compares the system as a whole whereby it would be difficult and time consuming for one individual to carry this out alone.

If one wanted to compare the endless possibilities of the Pentium 1 era, for example, my personal preference would be to
1) Compare all CPU varieties while holding all other hardware constant
2) Compare all major GPU varieties while holding all other hardware constant
3) Compare all chipsets varieties while holding all other hardware constant

The results of this approach are very difficult to achieve unless one individual tester has all combinations of hardware. To have multiple users test introduces uncertainty due to variations of supposedly the same hardware, as Vetz pointed out. Or the user may be missing one element which was different, or set the CMOS differently, etc.

I've added two more boards. An Epox VIA MVP3 and my Compaq (430VX).

I'll add ECS SI54 AIO (Socket 5 SiS) tomorrow.

vetz wrote:

I'll add ECS SI54 AIO (Socket 5 SiS) tomorrow.

Added. One of the slower boards. Uses Phoenix BIOS (first Socket 5 board I've seen with that)

no asus p5a?

i have one and have a millenium pci compaq 4mb but the only suitable cpu is a p75 wich will not do 100mhz or a p120 wich has way to much bent pins to get straight without loosing my patience 😀

meljor wrote:

no asus p5a?

i have one and have a millenium pci compaq 4mb but the only suitable cpu is a p75 wich will not do 100mhz or a p120 wich has way to much bent pins to get straight without loosing my patience 😀

I have the board, will be tested soon.

Thanx!

Two new boards added:

ASUS P/I P55TP4XE (Intel 430FX) with 256 and 512kb PB
ASUS P/I P55SP4 (SiS 551X) with 256kb and 1024kb async

ASUS P5A crashed in Quake benchmark, so need to figure out what is causing it before making the results public (it won't beat the HX boards). Will also retest my Compaq with 512kb cache.

Nice, a whole 3% performance boost when using 1024K vs. 256K on the ASUS P/I P55SP4 board. Does this board have an undocumented 40 MHz FSB setting, and if so, does it work with the original Cyrix 6x86?

feipoa wrote:

Nice, a whole 3% performance boost when using 1024K vs. 256K on the ASUS P/I P55SP4 board. Does this board have an undocumented 40 MHz FSB setting, and if so, does it work with the original Cyrix 6x86?

Haven't tested yet, will look at the board after work today. My guess is no 40mhz FSB, but support for Cyrix 6x86.

feipoa wrote:

Nice, a whole 3% performance boost when using 1024K vs. 256K on the ASUS P/I P55SP4 board. Does this board have an undocumented 40 MHz FSB setting, and if so, does it work with the original Cyrix 6x86?

OK, the board has an undocumented 40mhz FSB setting, option for fixed PCI bus speed and it does recognize the Cyrix 6x86 correctly. So in all it should be the perfect Cyrix 6x86 80mhz board, BUT I have problems running all the VGA benches with L2 (both async and PB) cache enabled. I dunno if it's my board or whatnot. I think I need more testing to find the problem, but I'm waiting for a DIP cache remover before proceeding. Here is a screenshot of Speedsys:

Do you have an SRAM tester?

Did you try lowering your L2 or main memory timings?

Is it only at 40 MHz that you have issues with the L2 cache?

feipoa wrote:

Do you have an SRAM tester?

Did you try lowering your L2 or main memory timings?

Is it only at 40 MHz that you have issues with the L2 cache?

No, I don't own a SRAM tester, but there have been no problems with Intel Pentium CPUs and it affects both SRAM and PB cache types. I'm using the latest BIOS as well. I can't find any documentation showing official support for the Cyrix 6x86 with this board.

I did try to lower settings to their minimum, but this had no effect. Increasing FSB to 66mhz also had no effect. Only thing that have made a difference so far was disabling L2 cache.

There might be some Cyrix 6x86 register bit which needs to be altered via software. You can use ibmm1.exe as a quick way to change these. Just looking at the datasheets, these might be worth changing,

LINBRST
If set, the 6x86 will use a linear address sequence when performing burst cycles. If clear, the 6x86 will use a “1+4” address sequence when performing burst cycles. The “1+4” address sequence is compatible with the Pentium’s burst address sequence.

WT_ALLOC
If set, new cache lines are allocated for both read misses and write misses. If clear, new cache lines are only allocated on read misses.
[IIRC, LOCK_NW needs to be 0 to alter WT_ALLOC]

NO_LOCK
If set, all bus cycles are issued with the LOCK# pin negated except page table accesses and interrupt acknowledge cycles. Interrupt acknowledge cycles are executed as locked cycles even though LOCK# is negated. With NO_LOCK set, previously non-cacheable locked cycles are executed as unlocked cycles and therefore, may be cached. This results in higher CPU performance. See the section on Region Configuration Registers (RCR) for more information on eliminating locked CPU bus cycles only in specific address regions.

SLOP
If set, the LOOP instruction is slowed down to allow programs with poorly written software timing loops to function correctly. If clear, the LOOP instruction executes in one clock.. This bit is only availabe in 6x86 CPU’s with DIR1 of 17h through 21h.

For some reason it now works using the original Cyrix 6x86 P80 cpu (at 2x40mhz) with 1024kb async L2 cache enabled!:

I did not change any register bits with software.

With ASUS P55SP4, Matrox Mystique 220 & 32MB of EDO RAM. Optimized settings:
3Dbench2: 100,7
PCPBench: 23,8
Doom: 1907 realtics (39 FPS)
Quake: 11,6 FPS

Previously I had been testing with a Cyrix 6x86 PR166 and a Cyrix 6x86L PR200.

Surprised by the low Doom and Quake score.

Thanks for checking this. Could the motherboard be running the PCI bus at half speed (20 MHz) to cause these low Quake and Doom scores? If so, is there a means to disable the 1/2 FSB-to-PCI multiplier? Did your one other Socket 5 board with a 40 MHz bus exhibit such poor scores with a 40 MHz bus?

By comparison, the Cyrix 5x86-80 without branch prediction enabled indicated a Quake score of 11.9 fps, although I am not sure if the settings you used for this are the same as what I used in the U4BC - everything default, sound on, console text off.

feipoa wrote:

Thanks for checking this. Could the motherboard be running the PCI bus at half speed (20 MHz) to cause these low Quake and Doom scores? If so, is there a means to disable the 1/2 FSB-to-PCI multiplier? Did your one other Socket 5 board with a 40 MHz bus exhibit such poor scores with a 40 MHz bus?

By comparison, the Cyrix 5x86-80 without branch prediction enabled indicated a Quake score of 11.9 fps, although I am not sure if the settings you used for this are the same as what I used in the U4BC - everything default, sound on, console text off.

This is with the PCI bus speed jumper set to "32M" (whatever that means?). It has two settings, either CPUCLOCK/2 (20mhz) or this 32M setting. With the default setting at 20mhz I got 2035 realticks in Doom and 11 FPS in Quake. PCPBench and 3Dbench were unchanged.

I used Phil's VGA Benchmark settings, which have Quake and Doom in fullscreen mode without the GUI. If you used the default view settings that makes any comparison difficult.

Did your one other Socket 5 board with a 40 MHz bus exhibit such poor scores with a 40 MHz bus?

The only other board that I own and which works with 40mhz bus and the Cyrix6x86 80P is the ASUS P54TP4. I haven't tested it yet. I plan to do it in the near future and also compare the different 6x86's at the same 80mhz speed. We'll see how much the internal core improvements of each version give as juice pr. mhz.

ASUS P/I-P55TP4N 256kb PB and 512kb PB results added.

Feipoa: Did you miss this thread?