Conclusions:

- The FX 5200 really does not like slow CPUs. On both the C3 and the Celeron there was noticeable stuttering in UT99 and 3DMark 01. In 3dmark the framerate was all over the place, more so on the Tualeron 1000 - probably because it allowed for higher max FPS. Not a video card I'd recommend for one of these builds. Out of curiosity, I took the 5200 out and installed it in a Socket A machine I had on hand (Athlon XP 2400+, MSI VIA KT333, 512MB of DDR333, Windows XP SP2) and not only was the experience a lot smoother, but 3DMark 01 score jumped to well over 7000 pts. I used the same Forceware version as with the other tests. I also tried a Leadtek Geforce 4 Ti4200, same version of forceware, with both the C3 and the Tualeron, and there was no stuttering whatsoever. Maybe the 694T chipset just doesn't like the FX series cards...

- There's about a 23% performance difference between the C3 1200A and the Tualatin Celeron 1000. Not great, not terrible. I suppose the C3 1200A performs close to an 833MHz Pentium 3 Coppermine. My sample was stable at 1400MHz, closing the gap with the Tualeron even more - but I believe it's not clock speed that brings the Nemiah core down (it is running 200MHz faster then the tualeron), but a combination of L2 cache size and FPU architecture. While faster then the Samuel's FPU, it still can't keep up with it's intel counterparts. I'm placing the blame more on the FPU then the cache subsystem because the AMD Duron has the same amounts of cache (64KB data + 64KB instruction + 64KB L2) but can very well match the performance of a Coppermine and even Tualatin Pentium 3.

- The VIA C3 can clock down to 266 MHZ (66x4) on the Epox motherboard. This can be done either by using SETMUL or directly from BIOS. Great for DOS games. So far everything I tried on the build has worked without any issues. Jazz Jackrabbit, Descent, Doom, Shadow Knights, Desert Strike, Commander Keen 6, Preshistorik, Stargunner are some titles I've spent a few minutes playing. The only game that I could not get to run is dyna blaster (character won't move in game). Setting up sound in dos is really convenient when you have an ISA slot. I just added the SET BLASTER argument and everything just worked. No TSR's, no SB-LINK to mess with.

- 66MHz operation is unfortunately impossible to get from BIOS / jumperless on all VIA boards I own (and I have quite a few). FSB is either configured automatically (MSI 694T PRO) or by jumpers (EPOX EP-3VSA2). BIOS will only let you step up the FSB (for example if it's set to 133, the EPOX will let you go as high as 200, but not lower then 133). Luckily the EPOX only has 2 jumpers for FPS selection, routing those to a set of switches mounted on the back of the case is an easy task. It's a real shame FSB cannot be selected in BIOS like on most intel i815 motherboards, but I consider this a minor inconvenience.

- the Pentium 3-S 1400 is a MONSTER. In some tests is managed to overtake the 1000MHz tualeron by 20-33%. UT99 was only fully enjoyable on the Pentium 3-S at 1280x960, at least with the 5200. In quake 2 the GPU bottleneck is evident, as the P3-S only managed to score an extra 7 FPS over the Tualeron. It would have been interesting to see how it performs on a DDR motherboard as I believe the extra memory bandwidth would be very beneficial.

- I might re-do all tests with a better video card, possibly the aforementioned Ti4200 as it provided a fully smooth experience in UT99 with both the Tualeron and C3 1200A while the FX 5200 was a stuttery mess.

3DMark printscreens:

I've found the Nehemiah to behave like a Coppermine clocked at half the frequency when it comes to 3D games. I have my C3 1200 running at 1466MHz and it performs about as well as a PIII 733, sometimes a little better even. ALU performance is much better. I really like that CPU, it provides enough performance for 1999-2000 games and can be slowed down with many steps in between. It's also quite cool, power consumption is low on these chips.

If you want to do a proper test, I suggest tweaking BIOS settings a little bit (CL2 over CL3 timings is an easy one to spot) and using a different GPU with earlier drivers to remove CPU overhead and also run lower resolutions to ensure GPU is not the bottleneck in any given situation. Overclock that Nehemiah a little bit as well, I've seen people run theirs at 1533 and even 1600 MHz, but I guess I didn't win the silicon lottery!

Socket3 wrote on 2023-09-21, 19:26:

Conclusions:
- the Pentium 3-S 1400 is a MONSTER. In some tests is managed to overtake the 1000MHz tualeron by 20-33%.

Let me disagree.
But in terms of pure frequency, P3-S outperforms Tualeron by 40%.
It would be interesting to put the Tualeron on the 133 bus for comparison - it will be 1.33 GHz.

Such comparisons are always interesting. And time consuming to create, I suppose 😀

Socket3 wrote on 2023-09-21, 19:26:

There's about a 23% performance difference between the C3 1200A and the Tualatin Celeron 1000. Not great, not terrible. I suppose the C3 1200A performs close to an 833MHz Pentium 3 Coppermine.

If you mean the actual 'raw' computing power then I don't think your data backs that up. For such a comparison you need to eliminate the influence of the graphics card as much as possible. At least run the benchmarks at a very low resolution. Both FPU tests indicate the Celeron to be 60% to 90% faster even at the lower clock speed. But those two test will probably emphasise the FPU which also doesn't tell the whole story.

BTW. what is '4 bank interleave'? Sounds like some kind of dual channel to me but as far as I know this generation doesn't support anything like that.

VT82C694T does Interleave 4-Way, with everything tuned to the max its almost as fast as stock 440BX. C3 looks really good in those tests, from my experience its much closer to Celeron 300@450.

shevalier wrote on 2023-09-22, 08:39:

Socket3 wrote on 2023-09-21, 19:26:

Conclusions:
- the Pentium 3-S 1400 is a MONSTER. In some tests is managed to overtake the 1000MHz tualeron by 20-33%.

Let me disagree.
But in terms of pure frequency, P3-S outperforms Tualeron by 40%.

The frequency difference is obvious, and by that metric the C3 is 20% faster then the tualeron, something that does not reflect in benchmark scores and as such not relevant. Here's how I calculated the differences in performance:
- 3dmark 01 - tualeron is 17% faster
- UT99 - tualeron is 23% faster
- Quake 2 - tualeron is 31% faster

Then I averaged the results -> (17+23+31)/3= 23.6% performance difference between the tualeron and the C3.

I did not bother applying the same methodology for the P3-S since the data is invalidated because of a clear GPU bottleneck (at least in quake 2) but here we go:

- 3dmark 01 - p3S is 20% faster
- UT99 - P3S is 21% faster
- Quake 2 - P3S is 3% faster

Averaging this out, the P3S winds up 14.6% faster according to my data.

shevalier wrote on 2023-09-22, 08:39:

It would be interesting to put the Tualeron on the 133 bus for comparison - it will be 1.33 GHz.

That's a good idea. I could try overclocking the tualeron 1000, but I haven't tested this chip extensively yet. Maybe I'll give it a go at one point.

HanSolo wrote on 2023-09-22, 11:01:

Such comparisons are always interesting. And time consuming to create, I suppose 😀

Socket3 wrote on 2023-09-21, 19:26:

There's about a 23% performance difference between the C3 1200A and the Tualatin Celeron 1000. Not great, not terrible. I suppose the C3 1200A performs close to an 833MHz Pentium 3 Coppermine.

If you mean the actual 'raw' computing power then I don't think your data backs that up. For such a comparison you need to eliminate the influence of the graphics card as much as possible. At least run the benchmarks at a very low resolution.

I agree. I did mention the GPU bottleneck and the fact that I want to redo all benchmarks with a Geforce 4 Ti4200. The P3S is clearly bottlenecked by the 5200, but the tualeron not so much.

HanSolo wrote on 2023-09-22, 11:01:

Both FPU tests indicate the Celeron to be 60% to 90% faster even at the lower clock speed. But those two test will probably emphasise the FPU which also doesn't tell the whole story.

That is also true, but as you can see, the synthetic FPU benchmarks don't have much bearing on actual gaming performance... in games, both direct3d and opengl the difference is much lower. In fact the results encouraged me to try out newer 2001 games like ground control and serious sam the first encounter - and I was very plesantly surprised to discover that both games are playable on the C3 Nemiah. I say surprised because I tried them a few years ago on a Samuel C3 and they were completely unplayable.

Garrett W wrote on 2023-09-22, 08:31:

I've found the Nehemiah to behave like a Coppermine clocked at half the frequency when it comes to 3D games.

That was true for the samuel core witch came with a FPU clocked at half speed and a 12 stage pipeline. The Nemiah is considerably faster. My experience so far is that the Nemiah chips offer ~65-70% the gaming performance of an equivalent coppermine, even more if the game uses SSE - witch is another difference between Samuel and Nemiah cores. Samuel lacks SSE but has 3DNow!. VIA removed 3dnow support from the Nemiah chips and added SSE. You can see this in the unreal tournament benchmaks, a game that makes use of the SSE instruction set.

When I find a bit more time I'll bechmark the C3 1200A against an 800MHz coppermine and post the results.

Garrett W wrote on 2023-09-22, 08:31:

If you want to do a proper test, I suggest tweaking BIOS settings a little bit (CL2 over CL3 timings is an easy one to spot) and using a different GPU with earlier drivers to remove CPU overhead and also run lower resolutions to ensure GPU is not the bottleneck in any given situation. Overclock that Nehemiah a little bit as well, I've seen people run theirs at 1533 and even 1600 MHz, but I guess I didn't win the silicon lottery!

Unfortunately the test rig was only stable using CL 2-3-2-5 with the VIA C3. The tualeron would hang in UT99 and the P3-S would boot loop with those memory timings... The C3 did manage over 900MB/sec memory read in everest at those timings, witch is pretty impressive.

As for overclocing, the EPOX board does not have an option for voltage in bios, best I could do is 1400MHz. My abit ST6 could probably push it further with a voltage bump, but I don't feel the need to overclock the chip, as frankly the difference in performance from 1200 to 1400mhz is not really noticeable in demanding games like UT99. If I recall it did manage to score 50fps, so a 3 fps increase for 200Mhz.... not great. Quake 2 and 3DMark 01 responded pretty well to the extra 200Mhz. I can't remember exactly what FPS it managed in quake but I do remember 3DMark score jumped to over 4000 pts.

rasz_pl wrote on 2023-09-22, 11:33:

VT82C694T does Interleave 4-Way, with everything tuned to the max its almost as fast as stock 440BX.

The Epox performs similarly to my Abit BE6-II when it comes to memory speed and latency. Memory read and write are a bit over 1000 pts on both boards using a pentium 3 1100 SL5QW (fsb 100) + slotket on the abit and the Pentium 3-S 1400 on the Epox. This with PC133 ram running at 100MHz CL 2-2-2-5 on the Abit and PC133 ram running at CL3-3-3-5 with 4 bank interleave on the Epox.

rasz_pl wrote on 2023-09-22, 11:33:

C3 looks really good in those tests, from my experience its much closer to Celeron 300@450.

That would be true for older Samuel and Erza-T C3 CPUs, the Nemiah is much faster, as it runs it's FPU at full speed as opposed to half speed (on the Samuel and Erza), and it has SSE instructions. Nemiah is about 70% the speed of an equivalent P3, even faster if the game takes advantage of SSE.

[EDITED] - I forgot to mention something very important, it relates to your experience - the motherboard must fully support the C3 Nemiah core - otherwise it will perform very poorly (as per your experience) or the system will be unstable. My EPOX fully supports the Nemiah with the latest BIOS, as does my Abit ST6 - but my MSI 694T PRO with stock bios does not - and the 1200MHZ nemiah performs closer to a 350MHz pentium II on that motherboard.

Socket3 wrote on 2023-09-22, 20:32:

I agree. I did mention the GPU bottleneck and the fact that I want to redo all benchmarks with a Geforce 4 Ti4200. The P3S is clearly bottlenecked by the 5200, but the tualeron not so much.

The term 'bottleneck' is misleading because every influence of a graphics card distorts the result in such a comparison. And to make things worse, that influence is non-linear.
Let's say CPU A needs 5ms to calculate one frame, B need 10 ms and C need 15 ms. Additionally the graphics card needs 10 ms to draw the image.
That gives you framerates of 67 fps (15 ms), 50 fps (20 ms) and 40 fps (25 ms), although A is twice as fast as B and 3 times faster than C.

(Of course this is just a simplified example. The 10 ms for the card also include the graphics driver which will as well be faster on system A. But most work-time is spent by the GPU so in general this illustrates the problem when measuring performance of CPUs with graphics output included.)

But at the same time such a test is also very meaningful because it shows what difference a CPU has in real world applications. So I'm actually not criticizing 😀

Socket3 wrote on 2023-09-22, 20:32:

The frequency difference is obvious, and by that metric the C3 is 20% faster then the tualeron, something that does not reflect in benchmark scores and as such not relevant.

You're right that the clock speed doesn't say anything when comparing different CPUs, but both Tualatins are basically the same core with the Pentium running at 33% higher FSB which usually results in 33% faster CPU speed. A slight difference here is the 0.5 higher muliplier and the larger cache on the P-S

HanSolo wrote on 2023-09-23, 12:51:

You're right that the clock speed doesn't say anything when comparing different CPUs, but both Tualatins are basically the same core with the Pentium running at 33% higher FSB which usually results in 33% faster CPU speed. A slight difference here is the 0.5 higher muliplier and the larger cache on the P-S

That's not what we used to observe. We'd typically see an EB variant outperform an E variant by like 8-10%.

Garrett W wrote on 2023-09-22, 08:31:

I've found the Nehemiah to behave like a Coppermine clocked at half the frequency when it comes to 3D games. I have my C3 1200 running at 1466MHz and it performs about as well as a PIII 733, sometimes a little better even.

Ok so I found a 750Mhz Pentium 3 coppermine at the flea market today. It came together with an HP Pavilion 700 motherboard (Asus TUSL-L). The motherboard is dead, won't even power on - not even the green LED next to the ram slots will light up, but I decided to test the CPU in the same rig I used for this benchmark series, just to compare it to the C3 1200A. Here are the results:

Pentium 3 750MHz Coppermine 256k FSB100 cl 3-3-3-5, 133MHz ram, 4 bank interleave

Everest mem read: 783 mb/sec
Everest mem write 781 mb/sec
Everest mem latency 133.0 ns
FPU Julia 189 pts
FPU Mandel 100 pts

Unreal Tournament 1280x960 high - intro - 49.63 fps
Unreal Tournament 1280x960 high - fractal map, 1 bot (campaign) - n/a fps
Quake 2 3.24 1024x768 - timedemo 1 - 152.4 fps
3d mark 01 - 4072 pts

Here are some charts shocasing performance difference:

In 3DMark01 the performance gap is 2%. In Unreal Tournament there is a 5.5% difference in performance in favor of the P3 750, and in quake 2 the gap widens to 10%.

I'd say performance-wise the C3 1200A Nemiah is very close to the P3 750 - maybe someware between a P3 700 and the P3 750. Quake 2 is known to favor intel chips, witch might explain the 10% performance gap - or the gap may simply be present because Quake 2 does not leverage SSE instructions and the Coppermine has a better FPU.

In any case, if anyone was wondering what the 1200A's intel equivalent was - there we go 😀

smtkr wrote on 2023-09-23, 13:44:

HanSolo wrote on 2023-09-23, 12:51:

You're right that the clock speed doesn't say anything when comparing different CPUs, but both Tualatins are basically the same core with the Pentium running at 33% higher FSB which usually results in 33% faster CPU speed. A slight difference here is the 0.5 higher muliplier and the larger cache on the P-S

That's not what we used to observe. We'd typically see an EB variant outperform an E variant by like 8-10%.

I assume you mean where one runs at a FSB of 100, the other at 133 but both at the same effective clock of e.g. 600 Mhz (due to different a mulitplier)? That's not what I meant. Maybe my formulation was mistakeable.
I was talking about running the same core with the same multiplier at a higher FSB. This is pretty much the case with this Celeron-T 1000 and the Pentium-T 1400.

Here are some not-real world application, but actual same-clock benchmarks of these processors. Using 3DMark2001SE.
Re: Tualatin Celeron vs Williamette Celeron

gerwin wrote on 2023-09-24, 17:40:

Here are some not-real world application, but actual same-clock benchmarks of these processors. Using 3DMark2001SE.
Re: Tualatin Celeron vs Williamette Celeron

3DMark01 score with the C3 clocked at 1200 seems a bit low for a ti4200.... probably because the motherboard does not officially support the Nemiah.

Let me elaborate on that: - i've noticed that motherboards with no official support for Nemiah - be it Intel or VIA chipset - perform quite poorly. For example, on my MSI 694T PRO, the 1200A is not correctly detected (Post screen reads VIA C3 1197MHz). Memory performance is horrible - a little over 500mb/sec, game and benchmark performance as well. In 3DMark01 it managed to score a little over 3500 pts and UT99 is completly unplayable. Same results on my Soyo SY-7VEM.

I'm going to pull the Ti4200 out and run 3dmark01 with the nemiah to compare results.

[EDITED] - here it is

Guess I was wrong and your score is in the ballpark. Don't know why I thought the ti4200 should score higher, but I guess that's what a cpu bottleneck will do.

Socket3 wrote on 2023-09-24, 18:37:

3DMark01 score with the C3 clocked at 1200 seems a bit low for a ti4200.... probably because the motherboard does not officially […]
Show full quote

3DMark01 score with the C3 clocked at 1200 seems a bit low for a ti4200.... probably because the motherboard does not officially support the Nemiah.
Let me elaborate on that: - i've noticed that motherboards with no official support for Nemiah - be it Intel or VIA chipset - perform quite poorly. For example, on my MSI 694T PRO, the 1200A is not correctly detected (Post screen reads VIA C3 1197MHz). Memory performance is horrible - a little over 500mb/sec, game and benchmark performance as well. In 3DMark01 it managed to score a little over 3500 pts and UT99 is completly unplayable. Same results on my Soyo SY-7VEM.
I'm going to pull the Ti4200 out and run 3dmark01 with the nemiah to compare results.
[EDITED] - here it is
Guess I was wrong and your score is in the ballpark. Don't know why I thought the ti4200 should score higher, but I guess that's what a cpu bottleneck will do.

Thanks for double checking.
It should be a CPU bottleneck in my numbers there, I obtained a Ti4200 just for that reason.
In general, maybe there are some peculiarities about C3, that could benefit from special MSR programming at bootup. There are quite a few VIA datasheets that should hold such information. I read them partially in the past, but don't remember anything of that kind. The other thing is, that VIA had not quite mastered the CPU business, because their earlier Samuel and Ezra cores are kinda wonky. With slow memory access that I assume they were still trying to solve. Nehemiah was a lot better already, but after that they quickly moved to their own platform and soldered chips. With that in mind, I would not expect the Nehemiah to come any closer to intel performance.

gerwin wrote on 2023-09-25, 17:19:

Thanks for double checking. It should be a CPU bottleneck in my numbers there, I obtained a Ti4200 just for that reason. In gene […]
Show full quote

Socket3 wrote on 2023-09-24, 18:37:

3DMark01 score with the C3 clocked at 1200 seems a bit low for a ti4200.... probably because the motherboard does not officially […]
Show full quote

3DMark01 score with the C3 clocked at 1200 seems a bit low for a ti4200.... probably because the motherboard does not officially support the Nemiah.
Let me elaborate on that: - i've noticed that motherboards with no official support for Nemiah - be it Intel or VIA chipset - perform quite poorly. For example, on my MSI 694T PRO, the 1200A is not correctly detected (Post screen reads VIA C3 1197MHz). Memory performance is horrible - a little over 500mb/sec, game and benchmark performance as well. In 3DMark01 it managed to score a little over 3500 pts and UT99 is completly unplayable. Same results on my Soyo SY-7VEM.
I'm going to pull the Ti4200 out and run 3dmark01 with the nemiah to compare results.
[EDITED] - here it is
Guess I was wrong and your score is in the ballpark. Don't know why I thought the ti4200 should score higher, but I guess that's what a cpu bottleneck will do.

Thanks for double checking.
It should be a CPU bottleneck in my numbers there, I obtained a Ti4200 just for that reason.
In general, maybe there are some peculiarities about C3, that could benefit from special MSR programming at bootup. There are quite a few VIA datasheets that should hold such information. I read them partially in the past, but don't remember anything of that kind. The other thing is, that VIA had not quite mastered the CPU business, because their earlier Samuel and Ezra cores are kinda wonky. With slow memory access that I assume they were still trying to solve. Nehemiah was a lot better already, but after that they quickly moved to their own platform and soldered chips. With that in mind, I would not expect the Nehemiah to come any closer to intel performance.

I know that clock per clock the 1200A nemiah is not competitive, but considering how cheap it was... I don't understand why VIA didn't release a beefier version. My chips is stable at stock voltage all the way pup to 1400Mhz on most motherboards, so the clock pretty well. Some users reported over 1500mhz. VIA could have binned some faster chips, clocked them to 1500-1600 @ 1.6-1.7v, added more L2 cache, worked on that FPU a bit more and they'd have had a killer cheap gaming CPU. Back then (at least in my country) it didn't need to perform as well as an intel chip IF it was cheap and it at least got within 20% of the competition's performance. I'm saying this because the K6 line sold like hot cakes. The C3 could have as well if it performed a bit better.... Shame.

Oh well, at least for our use (flexible retro gaming) these chips are golden (and literally as well).

Socket3 wrote on 2023-09-25, 18:33:

I know that clock per clock the 1200A nemiah is not competitive, but considering how cheap it was... I don't understand why VIA didn't release a beefier version. My chips is stable at stock voltage all the way pup to 1400Mhz on most motherboards, so the clock pretty well. Some users reported over 1500mhz. VIA could have binned some faster chips, clocked them to 1500-1600 @ 1.6-1.7v, added more L2 cache, worked on that FPU a bit more and they'd have had a killer cheap gaming CPU. Back then (at least in my country) it didn't need to perform as well as an intel chip IF it was cheap and it at least got within 20% of the competition's performance. I'm saying this because the K6 line sold like hot cakes. The C3 could have as well if it performed a bit better.... Shame.

Oh well, at least for our use (flexible retro gaming) these chips are golden (and literally as well).

I can agree with that. Though there were also licensing issues that surely affected how things went with the C3.
CPU shack wrote it up here:

https://www.cpushack.com/2022/01/07/the-many- … ts-of-via-cpus/

Intel, as was its custom, sued VIA in 2001 asserting patent infringement, which it is likely VIA was expecting. As with the cas […]
Show full quote

Intel, as was its custom, sued VIA in 2001 asserting patent infringement, which it is likely VIA was expecting. As with the case of Intel and Cyrix, VIA countersued,
...
The deal also added a small detail that leads to todays discussion, it granted VIA a 3 year grace period to continue making bus and pin compatible processors up through 2006.
....
Looking at the table above we can see VIA took many roads in the development of their CPUs, with many that went nowhere. Some may see this as a lack of direction or focus, but in a lot of ways VIA seemed to be trying to figure out the best CPU for the market at the same time they were trying to make the best CPU from an engineering standpoint. Where these two paths converged you had a marketable CPU that made it into mass production, and where they didn’t, or where legal road blocks arose, the design was canceled.

Socket3 wrote on 2023-09-25, 18:33:

I don't understand why VIA didn't release a beefier version.

From Wikipedia
- C3 1.0A Release Date January 22, 2003
- The first Athlon 64 FX was based on the first Opteron core, SledgeHammer. Both cores, produced on a 130 nanometer process, were first introduced on September 23, 2003
- On February 1, 2004, Intel introduced a new core codenamed "Prescott"

In 2003, who needed a Socket 370 CPU?

PS. VIA C3 has more collectible value than practical value. I think so.

shevalier wrote on 2023-09-26, 02:58:

PS. VIA C3 has more collectible value than practical value. I think so.

It's actually extremely practical. An Ezra-T (with a bit of very easy overclocking) has a top speed similar to a Pentium 3 Katmai @ 600 MHz, and a speed flexibility better than that of a Pentium MMX (you can basically hit most speed points: slow 386 / fast 386 / slow 486 / fast 486 / slow pentium / fast pentium, etc).
Furthermore, it has the advantage that it runs on the 440BX chipset (and based on my experience, without a BIOS update, it will still work on many boards just fine, even if it's not properly detected), so you also get the excellent stability & compatibility that the 440BX platform brings to the table.

This means that, for people who want a "do it all PC/time machine", that is suitable for Win98 and DOS gaming, such a platform with the Ezra-T would be better in every possible way compared to a Super Socket 7 platform, and also much better compared to a Pentium MMX/430TX platform.

Unfortunately Ezra CPUs are pretty much impossible to get. (At least that's my experience in the last few years)

HanSolo wrote on 2023-09-26, 08:52:

Unfortunately Ezra CPUs are pretty much impossible to get. (At least that's my experience in the last few years)

There was a very reliable seller from China who had a ton of 1 GHz Ezra-T CPUs up until 6 - 12 months ago, but the stock has since dried up. He still has normal Ezra CPUs @ 800 MHz, though, but I imagine these will be quite a bit slower (don't have any experience with them, so can't say for sure).

I should mention that the Nehemiah, which can still be found for a decent price, is a very good alternative. That one has a higher max performance for Win98 gaming (~ Pentium 3 / Athlon @ 800 MHz), but has less flexibility in DOS (it lacks 486DX-33 / DX2-66 / DX4-100 speed ranges, but it still has very good flexibility in the 286 / 386 and even Pentium area). Depending on the games one plays, this CPU might be even more desirable than the Ezra-T.