Tertz wrote:

Would be more comfortable to see one table with all settings-results from the fastest to the slowest for Ezra-T 1000. As you also have Nehemiah 1200, its results are similarly interesting.

Agreed. Something maybe like this:

Tertz wrote:

Would be more comfortable to see one table with all settings-results from the fastest to the slowest for Ezra-T 1000. As you also have Nehemiah 1200, its results are similarly interesting.

I can put this in my to-do list...

Tertz wrote:

For video card you may use a standard S3 Trio64V+ or VirgeDX (72 MHz) as most people would use something close by the performance in case of Win9x-DOS build. It's possible also to run 2nd series of tests with ISA, as this may add to slowdown, while there were fast and slow ISA cards to choose from.

I can't decide if I want this build to have any 3D capability... Judging by Gona's Compatibility Matrix, the C&T F6xxxx has very good VGA compatibility.

And further testing using NSSI and AIDA actually confirms the onboard VGA to be on the PCI bus, rather than on the AGP bus. If I kept the onboard VGA, I'll also save an expansion slot... Hmm. The fastest on PCI I have is a Voodoo or an Nvidia chipset, but they are not as compatible as S3 cards.

Regarding ISA, I have thought of multiple VGA cards in this build using a Tseng, Cirrus Logic or a Paradise chipset...

Tertz wrote:

Tests that may be done. CPU: Speedsys, Speedtst (with /L), Snooper - they work on wide range of CPUs. Gaming: doom, quake, 3dbench, pcpbench - what Phil used for database. FPU tests for stock and overclocked states: Super Pi Mod. Any other which have a base for comparision.

I'll keep those benches in mind. I used NSSI and AIDA to confirm that the Ezra-T has L2 cache - it's strange that Speedsys didn't report this. The CPU has 128k of L1 cache, and 64k of L2 cache ; a rather strange combination.

Tertz wrote:

j^aws wrote:

This is the slowest I can achieve in Speedsys - 9.46pts, using the slowest timings in the BIOS.

200MHz; L1D; BPD; Slowest BIOS timings
and ICD slowdown option seems as not available with this

By disabling L1 cache, ICD (Instruction Cache) should be disabled too, and make it redundant.

j^aws wrote:

I can put this in my to-do list...

Besides tables for speed settings for C3 CPUs you have, you may add other info like known compatibility with MBs (and their revisions, BIOS), software, etc. There are no much retro users with these CPUs to say practical data.

I can't decide if I want this build to have any 3D capability... Judging by Gona's Compatibility Matrix, the C&T F6xxxx has very good VGA compatibility.

For DOS testing may be used 2D card.
While for general usage, if FPU is good, then 3D games will play well, - if to find PCI card with good 3D performance, and preferably not forgeting about 2D. From this point, to use MB with AGP slot is easier way - there is a lot of cheap GF2-3 AGP cards with excellent pefrormance in both 9x and DOS. GF seems to have high DOS compatibility, but if you want more it's possibly to setup S3 PCI and Nvidia AGP at the same time.

Tertz wrote:

Besides tables for speed settings for C3 CPUs you have, you may add other info like known compatibility with MBs (and their revisions, BIOS), software, etc. There are no much retro users with these CPUs to say practical data.

Most S370 boards and Slot 1 boards with the latest BIOS seem to work with Ezra-T and slower CPUs from my tests. Slotkets seem to be finicky when it comes to Slot 1 boards. From what I recall, I don't think I've come across a board that didn't POST with these CPUs...

Tertz wrote:

For DOS testing may be used 2D card.
While for general usage, if FPU is good, then 3D games will play well, - if to find PCI card with good 3D performance, and preferably not forgeting about 2D. From this point, to use MB with AGP slot is easier way - there is a lot of cheap GF2-3 AGP cards with excellent pefrormance in both 9x and DOS. GF seems to have high DOS compatibility, but if you want more it's possibly to setup S3 PCI and Nvidia AGP at the same time.

I have another 440BX build with AGP, and I'm thinking that will be a better candidate for 3D, and focused on running demoscene stuff.

I managed to get some random benches with a few PCI cards today:

Results below >>

Ezra-T was clocked at 8x150 - 1200MHz:

Surprisingly, the MX200 did very well and was the fastest. And the ARK Quadro64, looking the least capable, kept up with the pace. Still undecided on what to choose...

j^aws wrote:

Ezra-T was clocked at 8x150 - 1200MHz

So it's P2 450 MHz level in DOS on overclocked 1200 MHz and on standard it would be P2 400 level. Maybe a little faster, as PCI, not AGP cards were used.
With data of SuperPI the performance will be more clear.

Tertz wrote:

With data of SuperPI the performance will be more clear.

superpi 1m
http://hwbot.org/benchmark/superpi_-_1m/ranki … or_2327&cores=1

superpi 32m
http://hwbot.org/benchmark/superpi_-_32m/rank … or_2327&cores=1

More can be found here on hwbot.

Anyway I don't think superPI is the best benchmark for this cpu, A mmx Pentium out does it in superPI

The 7th slowest superPI 1m run on a 233mhz mmx pent on hwbot beats the best run with this cpu

j^aws wrote:

Surprisingly, the MX200 did very well and was the fastest. And the ARK Quadro64, looking the least capable, kept up with the pace. Still undecided on what to choose...

These are DOS tests, the video card speed hardly matters. I would say: enable write combining for the VESA modes and use the video card with the best compatibility: Voodoo3 or MX200.

Warning: Don't touch that VRM of a working voodoo3 PCI, it gets very hot.

Jade Falcon wrote:

superpi 1m

At 1200 MHz FPU seems be near AMD K6-2 400 MHz level. Presumably, P2 450 level in 3D games is maximum (if you did tests with fastvid previously).
Real fps in Win9x can be checked with Voodoo 3000 at resolution like 640x480. For example, in Unreal in software and Glide. Maybe Phil will say, for comparision, what fps he've gotten with his similar purpose and performance machine on K6, where was used a littler faster Voodoo 3500 (to down/overclock the cards to match would be good). He've said results only for 1024x768, while for CPU comparision lower resolution is better. To know fps on both CPUs with Geforce2 cards is also useful.

The best run on hwbot with superpi 1m with this cpu was at 1550mhz and it's about the same as a mmx pent at 300mhz in superpi.
That's hardly a p2-450 or k6-2 400.
But 3D is definitely hole other game. My 1200mhz Nehemiah goes neck and neck with my old 650mhz tbird setup.

It's really a fun cpu. I found with a savage 2000 the gap between my 650 tbird and the Nehemiah widens if I push the card. and over clocking the bus makes a big difference with the cpu. In fact the c3 scale far betting then most cpu do with over clocking

Jade Falcon wrote:

My 1200mhz Nehemiah goes neck and neck with my old 650mhz tbird setup.

Nehemiah is 25% faster at same clock. Where Ezra at 1200 MHz gives P2 450 level, Nehemiah should do P3 560 MHz what's close to your Athlon 650 MHz. In some 3D games the difference may to be higher, as C3 have worse FPU than Pentium 2-3. Higher resolutions and slow video may reduce it.

I found with a savage 2000 the gap between my 650 tbird and the Nehemiah widens if I push the card.

It's good to measure concrete fps in games on C3 and other CPUs. For comparision, I'd choose more standard Intel CPU and Nvidia card. On resolutions not higher than 800x600/16 bit.

^^
This is what NSSI shows for the Ezra-T, 150x8, 1200MHz:

You can see Dhrystone and Whetstone comparisions with other processors. BTW, it can't detect the correct MHz speed for the Ezra-T. The FP is close to a PII 450, and Integer close to a PIII 800...

And as requested, I've tabulated my earlier Speedsys results:

gerwin wrote:

j^aws wrote:

Surprisingly, the MX200 did very well and was the fastest. And the ARK Quadro64, looking the least capable, kept up with the pace. Still undecided on what to choose...

These are DOS tests, the video card speed hardly matters. I would say: enable write combining for the VESA modes and use the video card with the best compatibility: Voodoo3 or MX200.

Warning: Don't touch that VRM of a working voodoo3 PCI, it gets very hot.

There are some really slow PCI cards that I haven't included in my tests, and they can make an impact at higher resolutions in DOS.

I've tested a few more cards, and from running old demoscene stuff, it looks like an Nvidia M64 trumps the above for compatibility... Probably going to choose that card.

j^aws wrote:

This is what NSSI shows for the Ezra-T, 150x8, 1200MHz

Real mode, - probably it's for 16-bit operations. Performance in Win9x apps may differ.

I've tabulated my earlier Speedsys results

Nice. Similar table for Nehemiah 1200 is also interesting for comparision.

it looks like an Nvidia M64 trumps the above for compatibility... Probably going to choose that card.

Nvidia cards are good in this, and are one of the fastest (PCI and AGP) in DOS up to Geforce 3. In SVGA games this may to be important, besides benchmarks.

I'm surprised to see an SB-Link on an industrial board considering they're pretty rare on even consumer boards. Also, I agree that the processor is really nice. 😀

Okay, so I haven't forgotten about this - just been really busy lately.

After testing the board and CPU, I wasn't sure on what I wanted this build to focus on. Eventually, I decided to stuff it full of soundcards that are mainly geared towards non-wavetable synthesis. Below, a sneak preview on the current status:

I managed to get six soundcards playing nice with each other. Various incarnations of Yamaha OPL: YMF289, YMF262 and dual-YMF3812. Alongside them, supported by a MOS SID 6581(Innovation Replica) and Philips dual-SA1099 (Game Blaster).

Wavetable synthesis is supplied by the GF1 on the GUS MAX, and the EMU8000 on the AWE32. The rest will come from external synths using Roland MPU-401 and SoftMPU for Intelligent Mode.

For graphics, I've been back and forth. Currently installed is a Voodoo 5500, but I had a GeForce MX400 until recently.

Installed from top to bottom :
CT3980
GUS MAX 2.1
CT1330
CT1300
SSI-2001
YMF-724F
V5500

That's it for now. I'll be testing a bunch of batch files for all this to work smoothly. Any thoughts will be much appreciated...

So, getting six soundcards installed was tricky, and I nearly forgot how I did it. I'll be documenting some rules that helped me:

- For the AWE32 CT3980, install it correctly. Which means get to know CTCU, CTCM and AWEUTIL. Ensure to also install the Device Driver CTCM in CONFIG.SYS. You can disable various parts of this card using CTCU. Importantly, you still need to initialise the card using CTCM, even when parts are disabled. Otherwise, ports are hogged from other cards.

- For the GUS MAX, this card can easily live with others due to its non-conflicting ports. However, it needs to be initialised, otherewise it may block FM on other cards for some games.

- For the Sound Blaster Pro 1, removing jumper from port 220 disables the card, so that others can work alongside it and use its DMAs and IRQs.

- For the Game Blaster, if all other cards are disabled, it can live on port 220, which is the most compatible. It also needs no DMA, nor IRQs; minimising conflicts.

- For the SSI-2001 Innovation Replica, it is the simplest to configure. It needs an uncommon port of 280, and no IRQs, nor DMAs.

- For the YMF724 using SB-Link, forcing a PCI IRQ in BIOS helps conflicts. I force low IRQs of 3 and 4 to this card and the Voodoo 5500. The Sound Blaster Pro 2 ports, IRQs and DMAs can be set differently using SETUPDS.

I have a very simple AUTOEXEC.BAT with all relevant PATHs, and within it, I call upto six batch files configured for each sound card. In these individual batch files, I have all relevant environmental variables, such as SET SOUND, SET BLASTER etc... Once all tested, I'll probably setup a menu-driven startup.

The PnP resource manager in the BIOS needs to be tweaked carefully. I like all the Sound Blaster cards to use the popular port 220, and this is achieved by initialising/ activating the relevant card. I keep both the PnP AWE32 and YMF724 on IRQ 5, and the SB Pro 1 on IRQ7.

Finally, I use the default DMA1 on all these SB cards, so the only change in the BIOS I need when changing cards is setting DMA1 to Manual/ Auto. I force DMA0 to Manual, so nothing uses it, and the AWE32 grabs the first available DMA of 1. It does the same for the first available IRQ of 5.

The BIOS settings for the GUS are left Manual for IRQ11 and DMA7. It uses port 260.

Once setup, flicking through these six soundcards is fairly trivial. I'll setup an external toggle switch to enable / disable the SBPro1 with a simple case-mounted switch. Haven't looked into a suitable external mixer yet; undecided on this...

Phew, that seemed a lot, but in practice, it's easy to switch sound cards...

EDIT:
Reminder that HIGHSCAN option for EMM386 in CONFIG.SYS causes issues with AWE32 locking, so avoid this entry.

A small sidenote after further testing: The YMF724 needs the GUS not initialised - the reverse of when using the ISA cards for FM.

I can't resist filling all slots, and the last slot has always been the most difficult - currently has the V5500. Realising the focus of this build and future complementary builds, I wondered if I could get seven soundcards installed. So I dragged out my neglected TerraTec EWS64XL, mainly to checkout its 4236b CrystalFM, and managed to make it the seventh working soundcard!

I just used the basic PnP Crystal CWDINIT and CWDMIX DOS tools, forcing the 'o' switch when initialising. This can't get the two MPUs working. Anyone have a working set of DOS files, preconfigured to work with an EWS64XL 1.2? I have the installation CDs buried in some random box, and finding them is going to take a while. Thanks in advance.

VIA C3 Samuel 2 and Erza/Erza-T is known to have the FPU running at 1/2 of the CPU clock, if your C3 runs at 1200 MHz, the FPU runs only at 600 MHz

VIA C3 Nehemiah has the FPU running at fullspeed, so this is why Nehemiah is considerably faster, but also Nehemiah has a deeper pipeline (16 stage vs 12 in Erza/Samuel 2) and many other changes like cmov instruction support and the replacement of 3DNow! with Intel's SSE

I wonder about the VIA C3 Samuel 1 (known as VIA Cyrix III) which is even slower than Samuel 2, but extremenly rare.

VIA Samuel 1 uses a older process (180 nm), higher voltages (1.8-2.0 volts) and doesn't have L2 cache at all. Ranged from 433 MHz up to 733 MHz

^^ I think I might have a Samuel 1 somewhere. Do those have Cyrix silkscreened prominently over VIA?

For slowing down, the Nehemiah is lacking a speed range around 486DX-33 to a P133. It behaves more like a PIII when using slowdown techniques.

j^aws wrote:

^^ I think I might have a Samuel 1 somewhere. Do those have Cyrix silkscreened prominently over VIA?

For slowing down, the Nehemiah is lacking a speed range around 486DX-33 to a P133. It behaves more like a PIII when using slowdown techniques.

I have one of these, I ran Phil's VGA Benchmark suite and entered the results here:https://docs.google.com/spreadsheets/d/1lvF9n … wpNU/edit#gid=0 It's entry 437 of the now locked database. This was on a very fast ASUS CUSL2 (intel 815) motherboard with a 3DFX Voodoo Banshee so I suspect you'll not make it go any quicker than this for DOS games. It is a very slow processor indeed. As you can see it's sharing positions with 300MHz and under CPU's on the benchmarks, Cyrix MII's @ 250MHz are faster overall.
Here is a photo of my chip.