jakethompson1 wrote on 2020-07-09, 00:30:

I've had the ESA-486 industrial motherboard for a few years now and don't think I've posted Speedsys for it. It uses an SoC with […]
Show full quote

I've had the ESA-486 industrial motherboard for a few years now and don't think I've posted Speedsys for it.
It uses an SoC with a Cyrix 486DX4 core.
As you can see, there is no L2 cache. It does have 128MB of onboard PC100 SDRAM.
These hard drive numbers look pretty bad, don't they? Not sure why because I have UltraDMA enabled in the BIOS.

AFAIK, enabling UDMA or DMA in BIOS usually just allows it to be used but does not explicitly enable it . For that, under DOS, you would need to load something like the third party UDMA.SYS .

darry wrote on 2020-07-09, 01:41:

AFAIK, enabling UDMA or DMA in BIOS usually just allows it to be used but does not explicitly enable it . For that, under DOS, you would need to load something like the third party UDMA.SYS .

Ah, I was thinking the BIOS would install an int13h handler would accommodate it, but I guess not.
I assume SPEEDSYS does go through the BIOS and doesn't talk to the IDE io ports directly?
The UDMA on this board is apparently quite specialized. The ZFx86 website has a patch for the 2.2.x Linux kernel that I've been able to compile and use for UDMA. A plain Linux kernel won't do it. They also have a Win95 and Win98 driver.

I have not checked the full thread of this on 12 pages, but what I saw is very interesting, but I can't see which benchmark tools are you using for this. Has anybody put the Cyrix/TI 486SLC in coparison? I know, it is a "486" CPU for 386 systems, and a quick benchmark in Checkit 3.0 says, that it has only about 50% as the saved value of a 80486-33 system, even with L1 cache enabled by the Cyrix tools. I have such a machine, and I could run the benchmarks if they are compatible, and I also would like to do the comparison with the original 80386-33, and I coud compare with ULSI387 Math Copro and IIT387, and as soon as I get hands on a Cyrix FastMath 387 and an original Intel 80387.

1ST1 wrote on 2020-11-21, 08:54:

I have not checked the full thread of this on 12 pages, but what I saw is very interesting, but I can't see which benchmark tools are you using for this. Has anybody put the Cyrix/TI 486SLC in coparison? I know, it is a "486" CPU for 386 systems, and a quick benchmark in Checkit 3.0 says, that it has only about 50% as the saved value of a 80486-33 system, even with L1 cache enabled by the Cyrix tools. I have such a machine, and I could run the benchmarks if they are compatible, and I also would like to do the comparison with the original 80386-33, and I coud compare with ULSI387 Math Copro and IIT387, and as soon as I get hands on a Cyrix FastMath 387 and an original Intel 80387.

If you download the pdf files, then you will also see more detail around the testing methodology and the various benchmarks used (both under DOS & Windows).
I think one of the reasons why the Cyrix SCL/DLC CPU's were excluded, is because they don't fit on a Socket 3 platform.

The intention of this exercise was to keep (as far as possible) all the other components the same, apart from swopping out the CPU's - this was to obviously to ensure that there were no other variables in the testing that was performed.
Granted, there were some CPU's that were tested on other motherboards, but with the same chipset.

I seem to recall that @feipoa was planning on performing a similar benchmark exercise on the 386 platform, but I'm speaking under correction.
If you had said the Cyrix 486DLC, then I would say you perhaps have an interesting point, but I wouldn't regard a Cyrix 486SLC as being in the 486 class, since it runs on a 16-bit external data bus (like your 386SX CPU) which severely cripples its performance. At best, you will get the equivalent performance of a 386DX-33 with a 486SLC CPU (provided you can get the L1 cache to operate properly on the motherboard), but nowhere near 486 performance.

That being said, Texas Instruments (being a second source for Cyrix back then) also made a Cyrix 486DLC equivalent, called the TI486SXL with 8kb of L1 cache (both in 132-pin and 168-pin packages and even clock doubled versions), The 168-pin equivalent one could probably test on a similar setup to see how it stacks up against your Intel/AMD 486DX counterparts.

However, based on my own tests, a Cyrix 486DLC 40 MHz CPU runs just below that of an Intel 486DX 33 MHz CPU (in terms of integer performance). Considering that most 386 motherboards did not have VLB slots, this would also hinder overall performance compared to an equivalent 486DX-33 setup with a VLB grahpics card.

@feipoa did perform some comparative tests between a Cyrix 486DLC & TI486SXL - Cyrix/Ti 486DLC vs 486SXL

No, it's a 80386DX-33 machine upgraded with the TI486 (Cyrix) CPU. So it's fully 32 bit on the mainboard, but not the ISA cards. Maybe soon I have two interesting real 486DX machines to benchmark. First I need to restaurate them. These machines have no setting for clock speed or voltage, so mainboard always runs at 33 MHz. But they are EISA bus, I have EISA VGA cards for them, and they have 64 bit interleaved memory bus, up to 64 MB with an access speed of 30ns, no L2 cache. Memory bus is similar to Pentium 4 machine, these machines always require to install two PS/2 simm modules at the same time. They made that because besides the 486 the mainboard also could be populated with an i860 Risc CPU as a "copro". The one mainboard has the original i486DX-33, the other has a voltage adapter socket with an Intel or AMD DX4-100 (I forgot, that's made more than 25 years ago) and maybe I can upgrade the DX-33 to a Cyrix-DX2-66.

I will check the PDFs and let's see...

jesolo wrote on 2020-11-21, 12:19:

The intention of this exercise was to keep (as far as possible) all the other components the same, apart from swopping out the CPU's - this was to obviously to ensure that there were no other variables in the testing that was performed.
Granted, there were some CPU's that were tested on other motherboards, but with the same chipset.

Has anyone performed the inverse test? Use the same CPU with different motherboard chipsets? I imagine that the biggest differences would be in I/O and WB cache support. Probably a more difficult exercise because you would need to find a way to control for l2 cache size, WB cache support, wait states, bios versions, and the 3 common bus types found on socket 3.

I'm curious because I see some huge (4x) performance variation in PCI performance with the early socket 3 implementations. What kind of performance variation was there for ISA and VLB on the socket 3 platform? How can you measure that without testers having the same add-in card?

douglar wrote on 2020-12-02, 15:26:

jesolo wrote on 2020-11-21, 12:19:

The intention of this exercise was to keep (as far as possible) all the other components the same, apart from swopping out the CPU's - this was to obviously to ensure that there were no other variables in the testing that was performed.
Granted, there were some CPU's that were tested on other motherboards, but with the same chipset.

Has anyone performed the inverse test? Use the same CPU with different motherboard chipsets? I imagine that the biggest differences would be in I/O and WB cache support. Probably a more difficult exercise because you would need to find a way to control for l2 cache size, WB cache support, wait states, bios versions, and the 3 common bus types found on socket 3.

I'm curious because I see some huge (4x) performance variation in PCI performance with the early socket 3 implementations. What kind of performance variation was there for ISA and VLB on the socket 3 platform? How can you measure that without testers having the same add-in card?

I did perform a couple of tests with one or two 486 CPU's on different motherboards.
However, my tests were to see how your "hybrid" motherboards compared to a "standalone" 486 motherboard (by "hybrid" I mean a motherboard that can accept both a 386 or 5V 486 CPU).

What I've found was that the "hybrid" OPTI 495XLC chipset does hamper performance.
For example, on a DataExpert EXP3406 motherboard (which has the OPTI 495XLC chipset), I managed to get a maximum of 10.43 fps in the Doom timedemo3 test with a 486DX-33 CPU.
On an Asus VL/I-SV2GX4 motherboard (which utilises the SIS 85C471 chipset), I managed to get 13.90 fps and even 14.48 fps on another unbranded motherboard with the same chipset.

On a 486DX2-66 I got the following between the DataExpert EXP3406 vs the Asus VL/I-SV2GX4:
15.05 fps vs 24.27 fps.

All of the above tests were performed without EMM386.EXE loaded.

I must add that I did use two different graphics cards on the motherboards, a Tseng ET4000AX (1 MB) ISA card on the "hybrid" motherboard and a Cirrus Logic CL-GD5422 (1 MB) ISA card on the "standalone" motherboards.

It's interesting that you posted that. The recent 486 tower that I won has an Informtech IT486SLC MINI, based on a 495SLC. Might have to look for something better to replace it with.

This are some snapshots from my Toshiba T4850CT (486 DX4/75, 8MB RAM download/file.php?id=88419&mode=view ) laptop benchmarks:

Probably a dumb question.

If the DX4 is a triple clocked 33mhz DX and the AM5x86 a quadruple clocked DX then wouldn't the DX4 really be 99mhz and the 5x86 132mhz?

From what I understand, normally the front-side bus clock rate is at or around 33.3333 MHz, so more like 99.9999 MHz. People tend to round out the decimal places.

feipoa wrote on 2020-12-27, 03:57:

From what I understand, normally the front-side bus clock rate is at or around 33.3333 MHz, so more like 99.9999 MHz. People tend to round out the decimal places.

It always makes me think of an LP record.

This is a very interesting project and thread. I have the one and only DX2 with WB L1, the SX955. I've only tested it in VLB motherboards so far, but in all but one case, performance is actually worse with WB enabled. The Asus PV/I-486SV2G shows measurable improvement with WB enabled, but it is very little (1-2% at best). Don't go out of your way to acquire this CPU. I do have a SiS-based PCI board I haven't used yet, so that'll happen soon. I bought it because it has 8 cache sockets, which is rare for PCI SiS boards.

I haven't done a thorough comparison, but I can report that the PC Chips M912 seems to be a little faster than the M919 tested here when equipped with Award BIOS and fully optimized. This is based on my testing with the DX2-66 and a S3 Trio64 2MB DRAM. I do barebones testing with VLB IDE to SD Card, and an ISA ethernet card to load XT-IDE BIOS with ROM address cached. DOS 6.22 gets loaded in safe mode.

I've been looking for the perfect benchmark platform, and the M919 is the best candidate yet, but only revision 1.5. I happen to have one, but it's broken beyond my ability to repair. Revision 1.5 has traditional socketed cache which makes it possible to use 512kb L2. That board would allow you to compare all 486 CPUs and both VLB and PCI video cards with consistent settings for all.

Interesting that you mention that. I also have an m915 that is broken beyond repair. I've been using it as a parts board. At what PCI clock rate does the m915 run at when the FS B is at 40 MHz? Is it like the M919 in this regard, that is it implements an automatic 2/3 clock reduction for the PCI bus?

anybody noticed the "new 486 world record" ?
https://www.youtube.com/watch?v=qPRSHI4JoWI

180Mhz on an X5 with over 20fps in Quake

CPU Galaxy has done it again. Achieving 200Mhz on an AMD 5x86 running the FSB at 66!

https://youtu.be/LI1_RlVLhu8

I didn’t think it was possible to run the PCI bus and IDE controller stable at this speed but it worked well enough in the vid.

Does he say how many AMD 5x86 CPUs he had to test before he found one that ran could run Quake at 200 MHz?

feipoa wrote on 2021-05-09, 02:45:

Does he say how many AMD 5x86 CPUs he had to test before he found one that ran could run Quake at 200 MHz?

Great question! Looked like an ADW.

Looks like he is running with a peltier... -8.3c at the very beginning of the vid.

Edit.... I'm dumb.. it says Peltier right in the video title.

Ahh, I guess he didn't find a magic Am5x86 which runs at 200 MHz with air cooling. So he used his 180 MHz air cooled chip at 200 Mhz w/peltier? Sorry, I haven't watched his video yet (too busy this weekend).