matze79 wrote on 2022-05-17, 21:50:

The VIA C3 dosnt work reliable for me at multi 3.0 below 100Mhz FSB. I also have to set it step by step, going instant to 3.0 en […]
Show full quote

The VIA C3 dosnt work reliable for me at multi 3.0 below 100Mhz FSB.
I also have to set it step by step, going instant to 3.0 ends up with freeze.
66*3.0 does not work for me, also freeze.
Just 100/133 FSB works pain free for me on Samuel 2.

So the slowest i can achieve is 300Mhz.

Yes, this topic has mostly been about the VIA C3 Ezra-T, which works down to 150 MHz without issues.
Nehemiah, on the other hand, has issues at FSB 50 when using lower multipliers (it does work fine with FSB 66 or above).
I do have two Samuel CPUs, but I've never tested them (as far as I know, they are very slow). Maybe I'll check them out in the future.
Sadly, I've been very busy lately, so no time for any retro activities. 🙁

Garrett W wrote on 2022-03-13, 10:52:

UT is indeed a good CPU benchmark, be sure to run UTBench which is more indicative of actual gameplay, instead of the flyby intro.

Is it me or is "utbench" actually very flawed and does not reflect the gaming experience in any meaningful way?
I've noticed it on multiple builds, but now that I'm actually testing it again on an ultra-overkill build, something seems very off.

SYSTEM SPECS
MB: Asrock 775i65g rev 3.0
CPU: Pentium Dual-Core E5800 @ 3.73 GHz (FSB 233 x 16)
RAM: 1 GB DDR PC3200 Single Channel
GPU: ATI Radeon "All-in-Wonder" X850XT
SOUND: Creative Audigy 2 ZS

Benchmark results:

1Unreal Tournament  640 x  480 x 16: AVG 122.51 FPS / HIGH 215.00 FPS / LOW 74.44 FPS
2Unreal Tournament 1024 x  768 x 32: AVG 118.06 FPS / HIGH 203.21 FPS / LOW 71.74 FPS
3Unreal Tournament 1280 x 1024 x 32: AVG 115.06 FPS / HIGH 204.44 FPS / LOW 70.89 FPS

This system in particular, runs ALL games/benchmarks at many hundreds of FPS (usually close to 1000 FPS), however, utbench is the only one that actually has a minimum framerate of ~ 70 FPS 😁.
And, sure enough, I ran a practice session with 16 bots, on the exact same level as utbench, and the game is actually unplayable without vsync because of how fast it runs (300+ FPS was the minimum framerate after a 30 minute gaming session).

That's... odd to say the least. Thanks for noting it out, I'd like to look into it a bit more before ruling it out as a benchmark issue.

bloodem wrote on 2022-07-22, 11:20:

This system in particular, runs ALL games/benchmarks at many hundreds of FPS (usually close to 1000 FPS), however, utbench is the only one that actually has a minimum framerate of ~ 70 FPS 😁.
And, sure enough, I ran a practice session with 16 bots, on the exact same level as utbench, and the game is actually unplayable without vsync because of how fast it runs (300+ FPS was the minimum framerate after a 30 minute gaming session).

Interesting findings!

Out of curiosity, how are the frame rates on the default flyby intro sequence?

Joseph_Joestar wrote on 2022-07-22, 12:59:

Interesting findings!

Out of curiosity, how are the frame rates on the default flyby intro sequence?

Did not test the flyby, unfortunately (and the PC is now back in storage, because I'm leaving my hometown tomorrow and returning... home 😁 ).
Will definitely test it again soon, on a different system. If I were to guess, I'd say the flyby intro is most likely extremely fast.

My turn to test a C3 Nehemiah 1200 with a C5P core. It's running on an Asus TUSL2-C, 512MB PC133 and Windows XP. I first used an eBay IDE to SATA converter (as you see on the photo) but it's really an unstable piece of junk, so I switched to a Promise SATA300 card.

First of all some 3Dmark 2000 scores with a Radeon 9600 XT with the Catalyst 10.2 and then a Radeon HD 3450 AGP (there is a PCI-Express to AGP bridge on the rear of the card) with Catalyst 10.7 drivers. Both cards are quite "similar" in performance despite their difference in age but I have to say that the 3450 has much better drivers in terms of visible texture bugs in games...one quirk is that the standard 3450 drivers you find at AMD don't support this card, I used the ones provided on the HIS website. The card itself is a Powercolor branded one, but CPU-Z reveals it's actually made by HIS...

Radeon 9600 XT:

Radeon HD 3450 AGP:

Then I overclocked the C3 using CrystalCPUID which lets you set the multiplier, I didn't change the stock voltage despite what CPU-Z seems to indicate. To my surprise it went from 1200 (9x133) to 1600 (12x133) and is absolutely stable at this frequency! It runs Quake 3 as long as you want and 3Dmark 2000 the same...at 12.5x133 Windows crashes with a BSOD.

The Quake 3 results are not that impressive (maximum quality, 1280x1024) but it remains very playable:
9x133 42 fps
9.5x133 43 fps
10x133 44.7 fps
10.5x133 45.9 fps
11x133 47fps
11.5x133 48 fps
12x133 49.2 fps

And the CPU-Z screenshot once overclocket to 1600MHz, and also the 3Dmark 2000 score:

~PIII 800.

Unfortunately for Centaur/VIA it's exactly that...same performance at double the frequency. But well, I have a soft spot for these underdogs.

Not with such new video card, yes.

Minutemanqvs wrote on 2023-03-29, 15:02:

and then a Radeon HD 3450 AGP (there is a PCI-Express to AGP bridge on the rear of the card) with Catalyst 10.7 drivers. Both cards are quite "similar" in performance despite their difference in age but I have to say that the 3450 has much better drivers in terms of visible texture bugs in games...one quirk is that the standard 3450 drivers you find at AMD don't support this card, I used the ones provided on the HIS website. The card itself is a Powercolor branded one, but CPU-Z reveals it's actually made by HIS...

That's the second time recently I have been hearing that the HD3450s do well for earlier XP stuff. I've had two sitting unemployed for years, looks like they might have a purpose in life after all... though mine are PCIe with Display Port weirdness (got cables)

I wouldn't hold my breath over it. It's more or less equal to Radeon 9600XT/X600XT, but with gimped anti-aliasing (bane of all HD2xxx/3xxx cards), no 16-bit dithering support and bloated drivers with inferior OpenGL ICD. So practically zero performance benefit over X300/X600, but with all the drawbacks.

I guess really it's only gonna get tried if I get more curious, if I fill every PCIe slot I've got with the best cards, then somehow explode them all at once, then fill every slot again with the next best cards, and wow, the same thing happens... then maybe the HD3450s would come into play.... then if I blow up the 3rd lot and the 3450s with them, and it's a cold day in hell, the x300 hypermemory might get a turn 🤣

edit: Upon reflection, that might be a lie, 3 of the boards have onboard graphics I'd rather use and I've got 4 32bit PCI cards I'd rather run than the x300 hypermemory, so I might have to blow all those up first. Pretty much have to get down to being the last working dx8 or better card that will go in a PCIe system

I have exactly 1 PCI card which is a Matrox Mystique 220 and 3 AGP cards, so I do with what I have on hand. The 4350 is clearly a low-end card, but as it's a late low-end card it is still very much usable in older games. I also heard on some youtube videos that cards which have a PCI-Express to AGP bridge often suffer of stuttering compared to pure AGP cards.

Warlord wrote on 2022-02-16, 07:03:

It's common knowledge that K6 450-600 is more like a PII 350-400. I have no idea why some people keep thinking its more than that and it also has a slower FPU than a pentium

Keep in mind most games back in the day were optimized in one way or another for the Pentium's FPU. The biggest offender was Quake 1. Quake 2 is another story as there's 3Dnow! optimized ports for it, but in general the Pentiums were up to twice as fast in games if now 3dnow! support was present.

You're 100% right about the super socket 7 platform, but don't blame VIA. The ALi Aladdin is just as slow and even buggier. Bugginess aside, the BX chipset has up to 3 times the memory bandwidth, and pentium 2 / 3 chips have much faster L2 cache then the regular K6-2 which has no L2 and relies on slow on-motherboard SRAM chips. The K6-2+ and K6-3 are excepted from this, but they don't benefit too much from the on-die cache due to socket 7's memory speed limitations. Keep in mind super 7 platform based on socket 5 which came out in 1995? - and that in turn is based on socket 4 which came out in 1993.... I'd say the K6 series performance is impressive considering what it runs on.

It would be interesting to see the K6-2+ or K6-III run (somehow) on a socket A motherboard, without the limitations of (even then) ancient socket 7 platform.

Minutemanqvs wrote on 2023-03-29, 15:02:

My turn to test a C3 Nehemiah 1200 with a C5P core. It's running on an Asus TUSL2-C, 512MB PC133 and Windows XP. I first used an eBay IDE to SATA converter (as you see on the photo) but it's really an unstable piece of junk, so I switched to a Promise SATA300 card.

Interesting! I'm using a C5P core as well (but it's revision A, so newer?) on a VIA 694T motherboard. I don't think my system is very stable beyond 1400, and at 12x it just freezes, so the furthest I went so far is 133x11.5. But I don't care about overclocking all that much.

To be honest I've been really enjoying the setup. I've paired it with a Voodoo 5 card and an SB32 soundcard in mATX form-factor, giving me a fairly balanced and versatile compact system.

The motherboard comes with built-in UltraDMA100, so while this is basically a 440BX-tier system, it feels a lot snappier. Yes, you can add any kind of IDE controller, but with mATX, expansion slots are at premium. Also the board POST is very fast compared to most 440BX boards I've used. Usually by the time my monitor syncs with the videocard, I'm already booting into Windows. To go one step further, I went with 98lite Micro. I like it, but not entirely sure it's a must-have at this performance level.

Another thing is, the chip runs very cool. I was a bit afraid to turn down the fan RPMs on my Tualatin, but here I did it, and the chip is routinely at ~30-35 degrees Celcius after hours of use.

And finally, the slowdown options are there too. Not as versatile as Ezra, but I can reach mid to high 386 levels with everything turned off in setmul on a 133 MHz FSB. Unfortunately, the chipset doesn't allow switching between 66 and 133 MHz FSB through software, otherwise I think it would be even better. It seems to play Wing Commander 1 decently, but realistically I'm only going to need slowdown for a few newer titles.

Currently my only gripe is that the motherboard's (it's a Chaintech 6VIA5T) front panel pins are located right next to the ISA slots, so you just can't fully secure something like Terratec EWS64XL (and it's not a particularly long card). Clearly they only had short ISA modems in mind with this layout.

Also for some reason the motherboard doesn't display my FSB correctly in POST (shows 266 Mhz vs actual 133), but this doesn't affect anything. Tried updating the BIOS and also using BIOS patcher, didn't help.

Looking at your video, with Q3A at lowest settings, the Ezra-T scores 61.6 at the default clock of 1Ghz versus the 66.7 of the juiced up K6-III+ @ 633Mhz. So I would say that it's roughly equivalent to the fastest AMD offering (with the speed throttling advantages you mention). Were you able to reach 133 x 10 (1.33Ghz) on any of your Ezra-T CPUs? Can you please be more specific about the voltage used to get to 1.26Ghz? Thanks and great work.

mockingbird wrote on 2023-07-05, 12:52:

Were you able to reach 133 x 10 (1.33Ghz) on any of your Ezra-T CPUs? Can you please be more specific about the voltage used to get to 1.26Ghz?

Admittedly, I have a fairly small sample size (only 7 x Ezra-T 1 GHz at this point), but they all work flawlessly at 1.26 GHz (most at the default voltage of 1.45V, only one or two need a bump to 1.5V).
As for 1.33 GHz, only one of my Ezra-T CPUs is fully stable at this frequency, and it needs 1.55V to achieve 100% stability (it's still nice and cool when running at that voltage, though).

mockingbird wrote on 2023-07-05, 12:52:

So I would say that it's roughly equivalent to the fastest AMD offering (with the speed throttling advantages you mention).

Well, I would't say that...
As mentioned, based on my experience, the Ezra-T is basically guaranteed to run at 1.2 GHz (FSB133 x 9), if not more. 440BX motherboards are also (almost) guaranteed to run perfectly at FSB133 too (even the cheaper ones).
At this frequency, this platform is hands down faster than any SS7 platform out there (no matter how much you overclock and fine tune the latter).
You also get the stability and compatibility of the 440BX chipset - everything just works, I haven't encountered a single issue with any of the motherboards I tried.
And, yes, you will also have a much more granular speed flexibility.
Last but not least, at the time of this writing, it's also cheaper compared to a SS7 platform.

On the other hand, what you are seeing in my SS7 video (Asus P5A + K6-3+ @ 633 + GeForce 2 Ultra) is basically the 'best case scenario'. You need the right CPU which can handle that frequency (most modded K6-2+ 570 CPUs seem to handle 633 MHz well, but this is FAR from being the norm), you also need the right motherboard (in my experience, the Asus P5A is the fastest SS7 motherboard out there), and you need to fine tune it, change registers and debug all sorts of issues (like nVIDIA cards not being 100% stable out of the box). And even after all of that work, you will still encounter incompatibilities and other unforeseen problems which seem unsolvable (as an example, I am still unable to play Lion King on any of my Asus P5A rev 1.04 boards - the game runs at 5 FPS and stutters like hell no matter what I tried - and at this point I think I've basically tried everything).
And, again, this is the best case scenario, NOT the norm. Most SS7 platforms will be much slower than what that video shows and will also be much more finicky.

I love the SS7 platform, but I love it BECAUSE of the aforementioned issues (and due to nostalgia, since I have many good memories related to this platform). However, I imagine that many people just want to play games without any headaches. 😀

bloodem wrote on 2023-07-05, 14:14:

Admittedly, I have a fairly small sample size (only 7 x Ezra-T 1 GHz at this point), but they all work flawlessly at 1.26 GHz (most at the default voltage of 1.45V, only one or two need a bump to 1.5V).
As for 1.33 GHz, only one of my Ezra-T CPUs is fully stable at this frequency, and it needs 1.55V to achieve 100% stability (it's still nice and cool when running at that voltage, though).

I think I will keep mine at 1Ghz then. Technically, I can do it in software since my BIOS doesn't have multiplier control and the ones jumpered on the motherboard are ignored (and are too low anyways). SMB and Setmul work fine though (which is fortunate, since my clock generator is either counterfeit silicon or defective), so I can take it down to 2ooMhz (but cannot go above 1Ghz or 100mhz FSB -- I have no idea what the clock IC really is, only that it is labeled ICS9148BF-26, and I could not find something equivalent from ICS/Renesas from that time that was pin compatible so it is a mystery to me).

Looking at this table, which setting do you think resembles a 486/DX-33 the most? If I go lower to say 150Mhz (50Mhz FSB x 3), how much slower is that in your experience with regard to the chart (486 SX territory perhaps)?

Well, I would't say that...
As mentioned, based on my experience, the Ezra-T is basically guaranteed to run at 1.2 GHz (FSB133 x 9), if not more. 440BX motherboards are also (almost) guaranteed to run perfectly at FSB133 too (even the cheaper ones).
At this frequency, this platform is hands down faster than any SS7 platform out there (no matter how much you overclock and fine tune the latter).

Hey, you don't have to convince me... I completely agree -- best case scenario with K6-III+ isn't realistic for most people, and the puny VRMs on these SS7 motherboards aren't all that appealing either. BX motherboards are a class above anything from that era. Even the most inexpensive ones have proper VRMs.

Last but not least, at the time of this writing, it's also cheaper compared to a SS7 platform.

Well, technically, yes... But finding the needle in the haystack motherboards with clock generators that are RayeR SMB-compatible makes Ezra-T more expensive, in my opinion. I contacted RayeR, he said he would consider adding more support, but don't hold your breath (my words)...

Yes, you're absolutely right about SS7 motherboards... Didn't like ALI at all, and VIA might be better, but I don't like their pre-Athlon southbridge controllers (buggy IDE). You are far more generous in your nostalgic feelings for them. I had both a P5A and P5A-B (latest revisions) and was glad to get rid of them.

I have my Ezra-T tested now and working properly, I'm going to pair it with a Voodoo3 2000, and it's going to serve as a good DOS/Win98 "slow" system... For Win98 "fast" I have the AsRock AGP board you benchmarked here with the C2D (as well as system already built with a P5PE-VM), but I'm going to start experimenting with PCIe platforms for Wndows 98 (VIA and Intel) and the GeForce 6800... I don't see any downsides to this, because the Win98 "slow" system can compensate for any games that don't function with the drivers required for the 6800 (though the 6800 drivers do support table fog).

What's left then for complete coverage in terms of games compatibility is a real 486 (have one) and 386 (in the process of being built). The prior for better VESA compatbility than the Voodoo3 can offer (Trio64 VLB), and the latter for older titles that require it. If I had a Voodoo2 SLI, I could use that with the Ezra-T paired with some generic S3 Virge and then I wouldn't need the 486, but the Voodoo2 SLI is like a white elephant.

I draw the line at XT and 286. Nice to watch videos about them while sitting on the loo, but no need to own one in real life... Let the Apple and Commodore people have them.

There is a survivorship bias in what we see of 440BX boards now, the earlier less flexible ones more likely got scrapped sooner. However, some of the early and crappy ones are still around. They have inflexible and weak voltage regs, don't go below 1.8 or so, and rarely do better than 115 FSB, I can't tell you if this is due to improvement in the BX silicon over the first year, or improvement in board design, because you don't get older silicon on newer boards and vice versa. Do not forget that 440BX was launched for PII @ 100Mhz FSB. With 450Mhz PIIs these boards are fine for their day and solid as you want, but try to get them running PIIIs and faster than 100 and you are struggling. Even with power regulated powerleap adapters and a bunch of workarounds these boards are a pain in the arse. The "famous" P2B had these problems in the earliest revisions, however community support did manage to alleviate some of these with known mods, voltage reg transplants etc. There are plenty of good boards released with baked in PIII support from after this, but the "all BX are golden" myth will bite you in the arse if you subscribe to it.

mockingbird wrote on 2023-07-05, 14:46:

I have no idea what the clock IC really is, only that it is labeled ICS9148BF-26

Why do you think it's fake? The ICS9148BF-26 is the reason why SoftFSB works. This clock generator can be found on a handful of motherboards (personally, I only saw it on the Gigabyte 6BXC and the Amptron PII-3100B). I think some older revisions of the Asus P2B might also have it, but never seen one myself.

mockingbird wrote on 2023-07-05, 14:46:

Looking at this table, which setting do you think resembles a 486/DX-33 the most? If I go lower to say 150Mhz (50Mhz FSB x 3), how much slower is that in your experience with regard to the chart (486 SX territory perhaps)?

Eh, I had actually started working on a batch file with a user friendly menu, so that I could easily choose any speed I wanted, but sadly I never had time to finish it. I performed many tests to identify the proper settings for each CPU performance level in more 'real world' scenarios, but all that data is on a disk that is currently in storage.
The table in that thread relies on Speedsys, which is one of the worst benchmark tools to properly measure the performance of fast machines that have been slowed down. For more accurate results, 3DBench 1.0C can be used. A 486SX would score ~ 19 - 21 FPS in this test, while a 486DX-33 would score ~ 26 - 28 FPS.
Either way, if you downclock the Ezra-T to 150MHz (50MHz x 3) without disabling any caches, it will still be way too fast (Pentium MMX equivalent speed, IIRC). So you will need to experiment with the "L1D" and "ICD" setmul arguments (in combination with the CPU's frequency) and see which combo will give you 25 - 30 FPS in 3DBench 1.0c

mockingbird wrote on 2023-07-05, 14:46:

What's left then for complete coverage in terms of games compatibility is a real 486 (have one) and 386 (in the process of being built). The prior for better VESA compatbility than the Voodoo3 can offer (Trio64 VLB), and the latter for older titles that require it. If I had a Voodoo2 SLI, I could use that with the Ezra-T paired with some generic S3 Virge and then I wouldn't need the 486, but the Voodoo2 SLI is like a white elephant.

You really don't need a 486/386 if you're only interested in games. Trust me, I have quite a few 386/486 PCs, and while they are very nice to tinker with, you are better off playing games on the Ezra-T which can be tuned for any speed point, starting with a slow 386. If you encounter any games that, for some reason, don't work with the Voodoo 3, you can always get a secondary S3 PCI video card and it will surely work with virtually anything you throw at it.

mockingbird wrote on 2023-07-05, 14:46:

I draw the line at XT and 286. Nice to watch videos about them while sitting on the loo, but no need to own one in real life... Let the Apple and Commodore people have them.

Me too (until recently, at least). I just purchased my first 286 and have really enjoyed playing with it and upgrading it as much as possible (still have some upgrades left to do). But, yeah, let's just say that it's far from being my first choice for playing any games. 😁

BitWrangler wrote on 2023-07-05, 15:30:

There is a survivorship bias in what we see of 440BX boards now, the earlier less flexible ones more likely got scrapped sooner. However, some of the early and crappy ones are still around. They have inflexible and weak voltage regs, don't go below 1.8 or so, and rarely do better than 115 FSB, I can't tell you if this is due to improvement in the BX silicon over the first year, or improvement in board design, because you don't get older silicon on newer boards and vice versa. Do not forget that 440BX was launched for PII @ 100Mhz FSB. With 450Mhz PIIs these boards are fine for their day and solid as you want, but try to get them running PIIIs and faster than 100 and you are struggling. Even with power regulated powerleap adapters and a bunch of workarounds these boards are a pain in the arse. The "famous" P2B had these problems in the earliest revisions, however community support did manage to alleviate some of these with known mods, voltage reg transplants etc. There are plenty of good boards released with baked in PIII support from after this, but the "all BX are golden" myth will bite you in the arse if you subscribe to it.

I've upgraded quite a few older 440BX boards so that they could handle voltages down to 1.3V (some of them very old and cheapo, like the Amptron PII-3100B - which actually has the "PII" in its name). After also recapping them, almost all worked flawlessly at 133 MHz. All in all, I have tested more than 100 x 440BX motherboards by this point (at least half being of questionable quality), and so far I only found a few that fail to work at FSB 133. The Epox EP-BX3 is one of those boards, which is only stable up to 124 MHz.
I have NEVER seen a 440BX motherboard that is only stable up to 115 MHz.

As for the survivor bias... it should also apply to SS7 boards, but it doesn't. 😁 Most boards were crappy back then, and most SS7 boards that you can find today (for insane prices) are still crappy.