Well it started off as Ezra 1000Mhz CPUs being the best retro bargain evarrrr because they were a whole $8 cheaper than the unit price of K6-2+-570s, then I pointed out you'd have to run them at a higher clock than 1000 to be as fast, so to somehow prove this wrong, results are pointed to for a CPU running at 1200+, where it wasn't clear what settings were used to obtain said results even, and now we're running on rapidly inflating in price 440BX motherboards, on rapidly inflating in price slotkets, with a not exactly super easy to come across GPU but at least it's not a Voodoo5, to prove how inexpensive the whole deal is. So now we have some results still using settings awkward to compare to the 23 year body of interweb wisdom, where individual results that do compare are going to get cast aside one by one, all to prove (at 1200+ mhz) that I wasn't right to say that a 1000mhz Ezra CPU was slower than a fast K6. .... and that's without even touching the circular logic that a CPU can't score above another CPU because that CPU is faster and the score proves it, unless it's the other CPU scoring it, then it doesn't count.... besides which the tests were picked by the OP as best representing his CPU... and apparently will continue to be picked, until we get down to whatever is actually a representation of 440BX AGP or memory speed, and nothing to do with how fast the CPU is, considering everything is heavily graphically based.

Am I odd that I want one to downclock it because it would make for a great CPU to meet specific CPU speeds and I have a way to avoid slotkets thanks to another Vogons member.

I actually dont know why this pissing match got started, and I find both threads, this one and the K6 to the stars one to be mildly silly since neither is using proper methods for providing accurate and verifiable test results, the K6 one is especially silly in that regard where standard deviation is disregarded totally and 1-2 FPS is suddenly counted as being a valid result rather than the CPU burping or the GPU farting and causing slight deviations in the test results.

The funniest part is that the best use case to build a k6 or a Ezra build isn't to go fast its to go slow. In this use case a Ezra with a 10x multi is better at going slow than a k6.

If you wanted fast there are better choices than either CPU. But if you wanted to go fast you are overclocking. Then it only fair to over clock both CPU in which case Ezra won there too

Warlord wrote on 2022-02-18, 23:01:

The funniest part is that the best use case to build a k6 or a Ezra build isn't to go fast its to go slow. In this use case a Ezra with a 10x multi is better at going slow than a k6...

With SetMul you can set 2x multiplier as minimum on K6-2/3+ on the fly but with Ezra you can set 3x as minimum. I would rather call it a tie 😀

Falcosoft wrote on 2022-02-18, 23:57:

Warlord wrote on 2022-02-18, 23:01:

The funniest part is that the best use case to build a k6 or a Ezra build isn't to go fast its to go slow. In this use case a Ezra with a 10x multi is better at going slow than a k6...

With SetMul you can set 2x multiplier as minimum on K6-2/3+ on the fly but with Ezra you can set 3x as minimum. I would rather call it a tie 😀

Gerwin said otherwise

Warlord wrote on 2022-02-19, 00:06:

Gerwin said otherwise

More precisely?

BTW, from SetMul's documentation (SetMul.txt):

MULTIPLIER OPTIONS
------------------
This table shows the available options for each supported processor:
VIA C3 Samuel 1: 3.0x to 8.0x, 11 choices
VIA C3 Samuel 2 step 0: 3.0x to 8.0x, 11 choices
VIA C3 Samuel 2 step 1+: 3.0x to 12.0x, 16 choices
VIA C3 Ezra: 3.0x to 12.0x, 16 choices
VIA C3 Ezra-T: 3.0x to 16.0x, 27 choices
VIA C3 Nehemiah: 4.0x to 16.0x, 25 choices
AMD K6-2+ / K6-III+: 2.0x to 6.0x, 8 choices (2.5x is excluded)
AMD K7 Mobile (Athlon) 3.0x to 24.0x, 32 choices
AMD K8 (Athlon 64 etc.) 4.0x to 25.0x, 22 choices (integer only)
Cyrix 5x86 1.0x to jumpered multiplier (2, 3 or 4), 2 choices
A VIA C3 may, or may not work reliably at a total core speed below 250MHz.

VIA C3 Ezra-T: 3.0x to 16.0x, 27 choices

AMD K6-2+ / K6-III+: 2.0x to 6.0x, 8 choices (2.5x is excluded)

27 > 8

?

Warlord wrote on 2022-02-19, 00:21:

VIA C3 Ezra-T: 3.0x to 16.0x, 27 choices […]
Show full quote

VIA C3 Ezra-T: 3.0x to 16.0x, 27 choices

AMD K6-2+ / K6-III+: 2.0x to 6.0x, 8 choices (2.5x is excluded)

27 > 8

?

For me "better at going slow" meant that which can be set to run at slower speeds, not how many steps are available. The Athlon Mobile has 32 steps. Does this mean that it is better at going slow than both the K6 and Ezra? I doubt it...

Also Re: VIA C3 Ezra / Ezra-T CPU

More speed steps allow you to dial in much more closely a desired speed. K62 have many gaps and cannot exactly duplicate certain speeds where as a C3 can.

Warlord wrote on 2022-02-19, 00:31:

Also Re: VIA C3 Ezra / Ezra-T CPU

I'm not convinced. More granularity is only important at the low end of speeds. It is virtually irrelevant if you can get few MHz differences at the high end. There are no speed sensitive software that can run at PII equivalent of 400 Mhz but do not at 450. And on SS7 platform you can use lower FSB and with a VIA southbridge you get twice the granularity with throttle. So K6-2/3+ with a VIA chipset is very flexible at the low end.

Well, if i set L1 and L2 off and clock down, the machine gets too slow for VIA Sound Drivers (MS-DOS)

matze79 wrote on 2022-02-19, 01:01:

Well, if i set L1 and L2 off and clock down, the machine gets too slow for VIA Sound Drivers (MS-DOS)

That's why it is important to get as slow as possible without disabling L1 cache. Most TSRs related to soundcards cannot work reliable with disabled L1 caches. Warcraft 2, Descent etc. can be set to perfect speed with a k6-2+ without disabling L1 cache (even when SB Live's SB 16 emulation is used). Even Ultima VII works well with throttle and L1 cache enabled (but with an ISA ESS 1868 soundcard) .

BitWrangler wrote on 2022-02-18, 22:01:

Well it started off as Ezra 1000Mhz CPUs being the best retro bargain evarrrr because they were a whole $8 cheaper than the unit price of K6-2+-570s, then I pointed out you'd have to run them at a higher clock than 1000 to be as fast, so to somehow prove this wrong, results are pointed to for a CPU running at 1200+, where it wasn't clear what settings were used to obtain said results

I never said it was the "best retro bargain evarrr" (I firmly believe that this title still belongs to certain Athlon XP builds).
Furthermore, a year ago, the K6-2+ 570 used to sell for $100 or more. It's only now that they are more affordable, since that seller in the US has thousands. Still, he only sells packs of 3, which I guess many people don't want.
Then there's the whole motherboard situation. Decent SS7 motherboards can end up costing A LOT more than the actual CPU (depending on where you live). And then there are all the quirks that the SS7 chipsets have, which I believe are well known at this point.

BitWrangler wrote on 2022-02-18, 22:01:

and now we're running on rapidly inflating in price 440BX motherboards, on rapidly inflating in price slotkets, with a not exactly super easy to come across GPU but at least it's not a Voodoo5, to prove how inexpensive the whole deal is.

It's not inexpensive, but I would argue that even if we assume that it's at the same price, it's still a better deal: you are getting a stable, fast and very flexible platform that allows you to play most games released between 1982 - 2000.
Regarding the GPU... it still baffles me that in this day and age people still don't understand how CPU benchmarks work and why it's important to use a powerful GPU, to avoid any video card bottlenecks.
Anyway, you'll be pleased to know that you will get the same results (at lower resolutions) with a cheap GeForce 2 MX (and now that you've mentioned it, I will certainly make a separate video about it). Yes, depending on the game, the GeForce 2 MX will be slower at 1024 x 768 or even 800 x 600. But at 640 x 480, on these specific platforms, it will usually match the GeForce 2 Ultra in most games (again, it's weird that I have to explain how CPU/GPU bottlenecks work).
Having said that, you don't seem to have a problem with GeForce 3 Ti 200/GeForce FX5900XT cards being used, but somehow an overclocked GeForce 2 Ti bothers you. 😁

BitWrangler wrote on 2022-02-18, 22:01:

So now we have some results still using settings awkward to compare to the 23 year body of interweb wisdom, where individual results that do compare are going to get cast aside one by one, all to prove (at 1200+ mhz) that I wasn't right to say that a 1000mhz Ezra CPU was slower than a fast K6.

You got it all wrong. At the default 1 GHz speed / FSB100, on a 440BX, the Ezra-T is still usually faster or at least AS FAST as an overclocked K6-2+. 😀

BitWrangler wrote on 2022-02-18, 22:01:

.... and that's without even touching the circular logic that a CPU can't score above another CPU because that CPU is faster and the score proves it, unless it's the other CPU scoring it, then it doesn't count....

I'm saying that certain synthetic benchmarks should be taken with a grain of salt (especially when the K6-2/3+ surpasses a Pentium 3 in those specific benchmarks 😁 ). I mean, does anyone actually think that there is any game in which a Pentium 3 Katmai 600 would be slower than an (overclocked) K6-2/3+? I hope not...
All in all, If you can think of any Win 98 3D game where you think the K6-2/3+ is faster than the Ezra-T running on 440BX, please do point it out and we can compare it as well (based on my experience so far, I don't think such a game exists, though).

BitWrangler wrote on 2022-02-18, 22:01:

besides which the tests were picked by the OP as best representing his CPU... and apparently will continue to be picked, until we get down to whatever is actually a representation of 440BX AGP or memory speed, and nothing to do with how fast the CPU is, considering everything is heavily graphically based.

I picked the most common games/tests that basically all reviewers have been using for 20+ years now. Had I picked anything else, you would have surely complained that I chose "games that nobody plays just to make the VIA C3 look good". Am I right, or am I right? 😀 Furthermore, I specifically picked games that have built-in benchmarks, thus giving us repeatable results (which is the only way for properly doing comparisons: no matter what you are comparing - repeatability is vital!). As mentioned before, if you have any other game suggestions, feel free to tell me and I will test those as well (but they MUST have built-in benchmarks, otherwise things get way too complicated).
Now, regarding the "heavily graphically based" statement - What on earth would you have wanted me to test... Excel performance? 😀 We are on a gaming focused forum, right? The word "games" is actually in its title.

And I never said that it was just the CPU, I specifically mentioned that the "winning combo" is the VIA C3 + 440BX chipset. So, yeah, a big part of the Ezra-T's appeal and performance is most definitely the 440BX chipset, can't argue with that.

So let me set the record straight: I DON'T think that, from a historical standpoint, the VIA C3 is particularly impressive - NOT AT ALL. In fact, let's not forget that these CPUs were launched in 2002/2003, at a point in time when AMD and Intel were already miles ahead. Also, the K6-2/3(+) was sold between 1998 - 2000, so there shouldn't even be a comparison between the VIA C3 and the K6 CPUs when it comes to the impact that they had on the market at the time.

However, from a retro gamer's perspective, it just so happens that a VIA C3 + 440BX combo gives you the best of both worlds (good Win98 gaming performance, excellent stability and compatibility and the ability to hit any possible speed point down to a 286). And, arguably, all for a cheaper price (at least in some parts of the world).

Bottom line: if anyone thinks I should change anything regarding how I'm performing these benchmarks and what games I'm testing, do let me know and we can work something out.

Falcosoft wrote on 2022-02-19, 01:03:

That's why it is important to get as slow as possible without disabling L1 cache. Most TSRs related to soundcards cannot work reliable with disabled L1 caches. Warcraft 2, Descent etc. can be set to perfect speed with a k6-2+ without disabling L1 cache (even when SB Live's SB 16 emulation is used). Even Ultima VII works well with throttle and L1 cache enabled (but with an ISA ESS 1868 soundcard) .

I'm guessing you never tested a VIA C3. That's actually the idea: you DON'T need to disable the L1 cache to hit a gazillion speed points (especially since there are games which enable the L1 cache at startup). You can selectively disable the branch prediction, instruction cache, decrease the multiplier down to 3x and more importantly, on a motherboard like the GA-6BXC, you can also drop the FSB down to 50 MHz using software. I have more than 20 SS7 boards, none of them can do this (their FSB is adjusted either through DIP switches or jumpers). Maybe there are some that might at least offer a SoftFSB BIOS menu, but in 20 years I never came across one of these (so if there are, they are ultra-rare).

Yes, on a VIA chipset, Throttle works very well (in conjunction with many CPUs like the Athlon XP "Thoroughbred" or even the K6 series), at least I've never had problems with it (although, I saw reports on this forum that Throttle has issues with joysticks and other peripherals).
But you get the same or even more flexibility with a VIA C3 Ezra-T + GA-6BXC combo (without needing Throttle at all), with the added bonus that you have Pentium3-like gaming speed at the high end, plus all the undeniable advantages that come with a 440BX chipset.

Yeah, I'm not sure what people are expecting here. I really enjoy this thread because BloodEm has taken the time to think a few things through and tests the same software with the same settings on the same platform with somewhat conservative options/settings (and proper, old drivers for the GPU!). Overall, a controlled environment. Point proven or not, I think this is useful data, and just fun to watch.
I too find it a bit perplexing that we are talking about pushing the system, at 1600x1200 no less, which seems like a rather arbitrary choice as I can't imagine many people will play using that kind of resolution on retro hardware. This is a CPU test.
Job well done bloodem!

That being said, I have a couple of suggestions should you continue with this:

• Consider entering the results in a spreadsheet for easier readability. I had to open three tabs to compare Ezra-T, Nehemiah and Katmai 600, moving all three videos forward by increments, it might get silly if you end up doing a lot more of these!
• Before moving on to K6-3 and variants, would you consider adding a few more Pentiums in the mix? A Coppermine at 600 would be really interesting to see how it does clock for clock, but perhaps finding the equivalent to the Nehemiah at 1.46 would also be interesting (I do have a personal desire to see this as I run a Nehemiah 1.46 semi-regularly), I suspect a Pentium III 733 might not be far off the mark.
• Instead of Unreal's flyby, perhaps UT's UTBench might be more realistic as a benchmark. I realize saying this that a) UT is a fast paced multiplayer shooter instead of the slower/exploration type shooter that is Unreal and b) UT is certainly more demanding, but I'm not certain Unreal's flyby is representative of anything. Then again you might ask, is 3DMark representative of anything?
• Would you consider running the Nehemiah slightly lower to compare with the Ezra-T clock for clock?

Garrett W wrote on 2022-02-19, 09:10:

I too find it a bit perplexing that we are talking about pushing the system, at 1600x1200 no less, which seems like a rather arbitrary choice as I can't imagine many people will play using that kind of resolution on retro hardware. This is a CPU test.

Exactly. At 1600 x 1200, that stops being a CPU test and it becomes a GPU test (at 1600 x 1200 the GeForce 2 becomes the bottleneck in most titles).

Garrett W wrote on 2022-02-19, 09:10:

[…]
Show full quote

• Consider entering the results in a spreadsheet for easier readability. I had to open three tabs to compare Ezra-T, Nehemiah and Katmai 600, moving all three videos forward by increments, it might get silly if you end up doing a lot more of these!
• Before moving on to K6-3 and variants, would you consider adding a few more Pentiums in the mix? A Coppermine at 600 would be really interesting to see how it does clock for clock, but perhaps finding the equivalent to the Nehemiah at 1.46 would also be interesting (I do have a personal desire to see this as I run a Nehemiah 1.46 semi-regularly), I suspect a Pentium III 733 might not be far off the mark.
• Instead of Unreal's flyby, perhaps UT's UTBench might be more realistic as a benchmark. I realize saying this that a) UT is a fast paced multiplayer shooter instead of the slower/exploration type shooter that is Unreal and b) UT is certainly more demanding, but I'm not certain Unreal's flyby is representative of anything. Then again you might ask, is 3DMark representative of anything?
• Would you consider running the Nehemiah slightly lower to compare with the Ezra-T clock for clock?

Yes, I realize that this will get very difficult to follow, which is why I will also create graphs as soon as I find the time.
I will definitely test more Pentium 3s (the 700 MHz Coppermine will follow with & without overclock) + a 600 MHz Celeron to see how it compares to the Ezra-T, since someone mentioned it in a previous post.
Noted, I will include UTBench in future videos!
Good idea, I will run the Nehemiah at the same clock speed as the overclocked Ezra-T (1.26 GHz) to see how it compares. It should still be at least 25% faster.

bloodem wrote on 2022-02-19, 08:03:

I never said it was the "best retro bargain evarrr" (I firmly believe that this title still belongs to certain Athlon XP builds).
Furthermore, a year ago, the K6-2+ 570 used to sell for $100 or more. It's only now that they are more affordable, since that seller in the US has thousands. Still, he only sells packs of 3, which I guess many people don't want.

When it comes to the price point, this is really much less relevant compared to measured raw performance data as prices may have completely changed a few years from now but the data will remain.

Great, thanks for taking these into account Bloodem!

bloodem wrote on 2022-02-19, 08:03:

I'm guessing you never tested a VIA C3. That's actually the idea: you DON'T need to disable the L1 cache to hit a gazillion speed points (especially since there are games which enable the L1 cache at startup). You can selectively disable the branch prediction, instruction cache, decrease the multiplier down to 3x and more importantly, on a motherboard like the GA-6BXC, you can also drop the FSB down to 50 MHz using software. I have more than 20 SS7 boards, none of them can do this (their FSB is adjusted either through DIP switches or jumpers). Maybe there are some that might at least offer a SoftFSB BIOS menu, but in 20 years I never came across one of these (so if there are, they are ultra-rare).

Yes, on a VIA chipset, Throttle works very well (in conjunction with many CPUs like the Athlon XP "Thoroughbred" or even the K6 series), at least I've never had problems with it (although, I saw reports on this forum that Throttle has issues with joysticks and other peripherals).
But you get the same or even more flexibility with a VIA C3 Ezra-T + GA-6BXC combo (without needing Throttle at all), with the added bonus that you have Pentium3-like gaming speed at the high end, plus all the undeniable advantages that come with a 440BX chipset.

1. You are right, and I never said so. I just replied to Warlord's claim that because Ezra has a 10x multiplier it is automatically more suitable for speed sensitive games. I think that the available 2x multi in case of K6 is more important than the 6.5x-10x multis of Ezra especially with L1 cache enabled. And I just offered a tie, never claimed the Ezra is worse 😀.
2. Just for completeness: On K6+ you can enable/disable write allocation, write ordering, disable only L2 cache, disable L1 but enable L2, decrease the multiplier down to 2x and on motherboards like Zida Tomato you can also drop the FSB down to 50 MHz using BIOS.
3. I do not know other regions, but in Eastern Europe Zida Tomato boards were the cockroaches of late Socket7 era, they could be found everywhere. And the whole series has Jumperless/Auto -Jumper function:

4. Yes, Throttle with a compatible VIA chipset means heaven and earth difference if you try to achieve a specific speed point at the low end. It can be considered as an 1x-16x CPU clock divider in case of VIA.
5. Yep, I have never thought of K6 as a PIII killer...

If in you're head you're thinking of K6 plain Jane 2s vs Coppermine PIIIs, there's quite a gap, but between a K6-III or 2+/3+ and a Katmai core PIII especially until it went to 133 FSB, there's a degree of performance overlap. This is due to Katmai not having on die cache at full speed. In the ~2ish years of games where sub ghz performance was still relevant, SSE wasn't very relevant yet. Then even the 133 Katmais get reeled back in when you do the same to a K6-III or later, increase FSB. But if you wanna be faster than that on a BX for cheap there's a good number of 667-933 coppermine PIIIs around still for $5.

bloodem wrote on 2022-02-16, 17:36:

Awesome! :-D Yeah, you'll need a CPU that's either natively running on 133 MHz, or one that can handle the overclock (I have som […]
Show full quote

Doornkaat wrote on 2022-02-16, 15:24:

Can do, will do!👍 (Please remind me if I forget.😅)
Since the program seems to do much more than just controlling the PLL I wasn't sure if you also wanted to know anything specific to the VIA C3 on this board.
I'll see if I have a CPU here that's likely to run at 133MHz FSB. The 1.1GHz will probably crash immediately.😄

Awesome! 😁
Yeah, you'll need a CPU that's either natively running on 133 MHz, or one that can handle the overclock (I have some 700 MHz Pentium 3 Coppermine CPUs that run just fine at 933 MHz, without additional voltage).
Or, you can just test other frequencies, that PLL handles quite a few: 66, 75, 83, 100, 103, 105, 110, 112, 115, 120, 124, 133, 140, 150.

So I tried it and it always claims original and new FSB was 124.0MHz, PCI was @ 41.33MHz and spread spectrum is 0.00%.
It also reads the original CPU frequency as 1100.90MHz and the new frequency as 1104.14MHz each time. This is all regardless to BIOS settings and values I try to set.