VOGONS


First post, by Nemo1985

User metadata
Rank Oldbie
Rank
Oldbie

Let's collect information, experiences and thoughts about those cpus:

Winchip C6

Filename
c6.pdf
File size
595.84 KiB
Downloads
101 downloads
File comment
Idt Winchip C6 Datasheet
File license
CC-BY-4.0

Very slow cpu, like a 486 on steroids. It supports mmx extensions, supported voltage v3.52 or 3.3. It doesn't support fractional multipliers
Cpu ID: 000541

Winchip 2

Filename
w2.pdf
File size
616.92 KiB
Downloads
90 downloads
File comment
Idt Winchip 2 Datasheet
File license
CC-BY-4.0

Fpu performance like the k6-2 cpus, while on integer it's a bit faster than the first model. It supports mmx and 3dnow extensions as all the next versions, voltage v3.52. It doesn't support fractional multipliers
Cpu ID: 000585

Winchip 2A

Filename
w2a.pdf
File size
649.59 KiB
Downloads
94 downloads
File comment
Idt Winchip 2A Datasheet
File license
CC-BY-4.0

Quite obscure cpu, sold with an infamous evergreen adapter and some missing pin which do not allow the multiplier to be chosen, they can be soldered and the multiplier selection is restored. Voltage v3.52
Apparently it doesn't support the 2x multiplier but fractional are added (2,33 and 2,66 other than 2,50 and higher ones), this is the only cpu that support fractional multipliers.
Cpu ID: 000587

Winchip 2B

Filename
w2b.pdf
File size
717.03 KiB
Downloads
87 downloads
File comment
Idt Winchip 2B Datasheet
File license
CC-BY-4.0

This is also quite uncommon, it uses a organic package instead of the ceramic one, it uses a lower voltage than the older revisions, my cpu is rated for v2.8 but to make it work reliably I need to set to v2.9. This version, again doesn't support the fractional multipliers.
Cpu ID: 00058A

I did some test with cachecheck v7 and those are the results of L1 cache (according to the program the cpu has 32kb of cache, while the datasheets says it has 32kb for data and 32kb for instructions:
Winchip C6: 209,5 MB/s 5.0 ns/byte 3.3 clks
Winchip 2: 837,8 MB/s 1.3 ns/byte 0.9 clks
Winchip 2A: 837,8 MB/s 1.3 ns/byte 0.9 clks
Winchip 2B: 837,8 MB/s 1.3 ns/byte 0.9 clks

So apparently there are no differences between the 2 versions of the cpu, but I also did some other tests, which I don't totally trust so I will try them on a different configuration.
I also needed to make them at 66mhz of bus (they crippled the performance quite a lot on the later cpus) but since the 2a doesn't support the 2x multiplier there are no alternatives here are the results:

Filename
w2b.pdf
File size
717.03 KiB
Downloads
87 downloads
File comment
Idt Winchip 2B Datasheet
File license
CC-BY-4.0

Subtle differences but apparently the 2A cpu is the fastest, I confess I didn't expect that.

Edit: added the bios writer guide, removed the benchmarks

Attachments

Last edited by Nemo1985 on 2022-03-01, 11:38. Edited 2 times in total.

Reply 1 of 26, by Anonymous Coward

User metadata
Rank l33t
Rank
l33t

All the winchip2s look about the same to me, and those are only marginally faster than winchip1. The only exception being Quake. I guess they improved the FPU on Winchip2, or maybe used 3DNow.

I think the most pathetic aspect of the Winchip 2B is that the lower core voltage had almost no impact on the peak power output.

"Will the highways on the internets become more few?" -Gee Dubya
V'Ger XT|Upgraded AT|Ultimate 386|Super VL/EISA 486|SMP VL/EISA Pentium

Reply 2 of 26, by Nemo1985

User metadata
Rank Oldbie
Rank
Oldbie

Yes the difference in peak power was disappointing, I believe that the difference would be bigger if I could use the 100 mhz bus setting (in both performance and wattage), most was disappointing that the cpu was unstable at his rated voltage and I needed to add 0,1v. Unlucky until I find a solution to run the 2a with 2x multiplier it will be impossible to do it (maybe I could try the 2,5 setting).

Reply 3 of 26, by Nemo1985

User metadata
Rank Oldbie
Rank
Oldbie

So I was redoing the tests (including the one on windows I didn't the first time) and the winchip2b is like... dead?
It hangs with doom (vanilla) and at soon windows loads, it's weird because it works fine all other tests.
with 2.8v (rated voltage) crash as soon doom loads and fast doom too.
with 2.9v fast doom works fine most of the times but doom still crash.
Any guess?

Anyway I can confirm winchip 2a is a bit faster than the other versions at the same clock speed, the results on dos benchmarks are quite repetible (probably because there are no other tasks in the background so there are no differences between a ran and another).

Reply 4 of 26, by Nemo1985

User metadata
Rank Oldbie
Rank
Oldbie

Edit, I can't add any other attachment to the first post, here are the results, finally complete.

photo_2022-03-06_15-46-54.jpg
Filename
photo_2022-03-06_15-46-54.jpg
File size
51.6 KiB
Views
2039 views
File license
CC-BY-4.0

There is not much to add.

The winchip2b is the very same exact copy of the winchip 2. According to the datasheet the 2a is almost the same, but I beg to differ, the results is always higher compared with 2 and 2b (which are the same).
Dos benchmarks are quite rock stable in their values, the biggest difference I noticed is Quake with a 0,1 difference between a run and another.
My winchip2b is somewhat broken it can't run stable at his frequency (200 mhz), but apparently it runs fine with lower settings, overvoltage is no solution, since with 3v the system fail to boot.

The wattage peak is not so different between the v3.52 and v2.8 because probably IDT did on purpose kept the higher voltage to have a better compatiblity with older motherboard.
The winchip 1 is a 486 on steroids, while the 2 is much better, from performance and also wattage point of view.

Reply 5 of 26, by feipoa

User metadata
Rank l33t++
Rank
l33t++

Are you able to include some photos of the CPUs you used? I don't remember my Winchip2 being inside any kind of interposer.

Looking at the 133 MHz charts in the Ultimate 686 Benchmarks Comparison, with results at 2x66 (133 MHz), it looks like the Winchip C6 (48.6) and Winchip2 (56.7) demonstrated the interger performance between an IBM 5x86c (43.3) and a Pentium P54C (61.5). Values in () are normalised integer test results.

For floating point performance, the Winchip C6 (30.8) fell behind a MediaGX (31.7) and an IBM 5x86c (36.2), while the Winchip2 (55.4) demonstrated results similar to a K6 (56.9).

In DOS Quake, the Winchip C6 (7.9) was between an Pentium Pro (8.9) and a MediaGX (6.5), while the Winchip2 (10.8) fell between an AMD K5 (11.2) and a K6-2 (10.1). A Pentium P55C scored 13.6.

In Quake 2 OpenGL, the Winchip C6 (9.3) had about the performance of an IBM 6x86 (9.2), while the Winchip2 (14.8) fell between a K6-2 (14.3) and a K6-2+ (15.3).

3DMark99Max scored the Winchip C6 (452) similar to that of a Cyrix MediaGX (419), while the Winchip2 (1687) scored remarkably well, falling between a K6-2 (2213) and Pentium Pro (1416).

In DOOM, the Winchip C6 (47.1) and Winchip2 (52.8) were the slowest of all chips tested, with the next closest being a MediaGX (53.3).

Plan your life wisely, you'll be dead before you know it.

Reply 6 of 26, by Nemo1985

User metadata
Rank Oldbie
Rank
Oldbie

Hey Feipoa, great to see you there.

The cpu with the infamous evergreen adapter is the Winchip2A, i'm talking about this thing (picture courtesy of Tiido): http://www.tmeeco.eu/BitShit/PCschit/NemoStuf … PU/IMG_7842.jpg
Here are the restored pins: http://www.tmeeco.eu/BitShit/PCschit/NemoStuf … PU/IMG_7841.jpg

As for the cpus here is a picture:

IMG_20220225_040025.jpg
Filename
IMG_20220225_040025.jpg
File size
1.55 MiB
Views
1946 views
File license
Public domain

As for performance, my round up is more limited than yours, but yes I agree, on integers the winchip was the slowest one, while the winchip2 was a bit faster, they didn't do much progress anyway.
On my tests in quake 2, amd k6 and k6-2 had 0,8 fps difference, while the k6-2 cxt increased the value of 2,4 fps compared to the k6-2 (my tests were done at 200 mhz 66x3, since a direct comparison at 100 was not possible). Any winchip2 cpu was anyway slower than vanilla k6.

Reply 7 of 26, by Nexxen

User metadata
Rank Oldbie
Rank
Oldbie
feipoa wrote on 2022-03-11, 08:53:
Are you able to include some photos of the CPUs you used? I don't remember my Winchip2 being inside any kind of interposer. […]
Show full quote

Are you able to include some photos of the CPUs you used? I don't remember my Winchip2 being inside any kind of interposer.

Looking at the 133 MHz charts in the Ultimate 686 Benchmarks Comparison, with results at 2x66 (133 MHz), it looks like the Winchip C6 (48.6) and Winchip2 (56.7) demonstrated the interger performance between an IBM 5x86c (43.3) and a Pentium P54C (61.5). Values in () are normalised integer test results.

For floating point performance, the Winchip C6 (30.😎 fell behind a MediaGX (31.7) and an IBM 5x86c (36.2), while the Winchip2 (55.4) demonstrated results similar to a K6 (56.9).

In DOS Quake, the Winchip C6 (7.9) was between an Pentium Pro (8.9) and a MediaGX (6.5), while the Winchip2 (10.😎 fell between an AMD K5 (11.2) and a K6-2 (10.1). A Pentium P55C scored 13.6.

In Quake 2 OpenGL, the Winchip C6 (9.3) had about the performance of an IBM 6x86 (9.2), while the Winchip2 (14.😎 fell between a K6-2 (14.3) and a K6-2+ (15.3).

3DMark99Max scored the Winchip C6 (452) similar to that of a Cyrix MediaGX (419), while the Winchip2 (1687) scored remarkably well, falling between a K6-2 (2213) and Pentium Pro (1416).

In DOOM, the Winchip C6 (47.1) and Winchip2 (52.😎 were the slowest of all chips tested, with the next closest being a MediaGX (53.3).

Reading this was a good coffee break.
It was jungle back then and I'm finally making sense out of all those S7 cpus.
If there was a "what could have been made to correct bad performance in core designs" article I'd be done for life, but there would be too much to write I think 😀

The few times I used a Winchip I was always disappointed by the low performance. I don't remember if it was "low priced" enough to be an interesting solution; not for multimedia (being MMX + 3DNow capable I doubt it was offered for office only; not mouth watering I guess), low performance there too.

Last edited by Nexxen on 2022-03-11, 09:55. Edited 1 time in total.

PC#1 Pentium 233 MMX - 98SE
PC#2 PIII-1Ghz - 98SE/W2K

Reply 8 of 26, by feipoa

User metadata
Rank l33t++
Rank
l33t++

Interesting. I have a Winchip 2A which looks like yours, except my pins were perfectly intact. Do we know the purpose of this hack job? Was it just to limit multiplier options?

Plan your life wisely, you'll be dead before you know it.

Reply 9 of 26, by Nemo1985

User metadata
Rank Oldbie
Rank
Oldbie
feipoa wrote on 2022-03-11, 09:55:

Interesting. I have a Winchip 2A which looks like yours, except my pins were perfectly intact. Do we know the purpose of this hack job? Was it just to limit multiplier options?

I suspected it was done to lock the cpu multiplier, but i'm not so sure anymore.
The other 3 missing pin were the: AJ23 AK22 AL21, if i'm not wrong, I identified them just watching the pictures so the eye could trick me.

Last edited by Nemo1985 on 2022-03-11, 10:52. Edited 1 time in total.

Reply 10 of 26, by Nemo1985

User metadata
Rank Oldbie
Rank
Oldbie
Nexxen wrote on 2022-03-11, 09:52:
Reading this was a good coffee break. It was jungle back then and I'm finally making sense out of all those S7 cpus. If there wa […]
Show full quote

Reading this was a good coffee break.
It was jungle back then and I'm finally making sense out of all those S7 cpus.
If there was a "what could have been made to correct bad performance in core designs" article I'd be done for life, but there would be too much to write I think 😀

The few times I used a Winchip I was always disappointed by the low performance. I don't remember if it was "low priced" enough to be an interesting solution; not for multimedia (being MMX + 3DNow capable I doubt it was offered for office only; not mouth watering I guess), low performance there too.

Winchip 1 was a 486 basically. The meaning of it is not clear to me, why don't stick with a 486 on steroids then? Probably when it was for sale the 486 was eol already and IDT offered it as a cheap solution (still you had to change the motherboard).
All the details points to backward compatibility. They kept the v3.52 voltage but they were able to increase the performance and lower the TDP from winchip 1 to winchip 2. I also think that the winchip2 was somewhat a copy of the k6-2, they probably had an agreement with amd that not only allowed IDT to use the 3dpowernow extesions but also to use the fpu side of their cpus? Just my guess... Other than that, using the 3dnow was probably a marketing choice, but they gave a nice boost to the programs when used.

Reply 11 of 26, by Nexxen

User metadata
Rank Oldbie
Rank
Oldbie
Nemo1985 wrote on 2022-03-11, 10:51:
Nexxen wrote on 2022-03-11, 09:52:
Reading this was a good coffee break. It was jungle back then and I'm finally making sense out of all those S7 cpus. If there wa […]
Show full quote

Reading this was a good coffee break.
It was jungle back then and I'm finally making sense out of all those S7 cpus.
If there was a "what could have been made to correct bad performance in core designs" article I'd be done for life, but there would be too much to write I think 😀

The few times I used a Winchip I was always disappointed by the low performance. I don't remember if it was "low priced" enough to be an interesting solution; not for multimedia (being MMX + 3DNow capable I doubt it was offered for office only; not mouth watering I guess), low performance there too.

Winchip 1 was a 486 basically. The meaning of it is not clear to me, why don't stick with a 486 on steroids then? Probably when it was for sale the 486 was eol already and IDT offered it as a cheap solution (still you had to change the motherboard).
All the details points to backward compatibility. They kept the v3.52 voltage but they were able to increase the performance and lower the TDP from winchip 1 to winchip 2. I also think that the winchip2 was somewhat a copy of the k6-2, they probably had an agreement with amd that not only allowed IDT to use the 3dpowernow extesions but also to use the fpu side of their cpus? Just my guess... Other than that, using the 3dnow was probably a marketing choice, but they gave a nice boost to the programs when used.

A 486 was a marketing move to get some money in? Give a super cheap deal to get into S7? Probably there is an answer from the guys that made it.

The thing to do would be to get all those engineers - of the 486 to PPro era - and let them talk about where, why, how on every S7 chip 😀
I have to admit that I always loved W2 on ppga, and Rise's chips too. Looked like they could take on Intel's MMXs' 😀

All these benchmarks are gold, thanks for the effort! (and Feipoa too)

PC#1 Pentium 233 MMX - 98SE
PC#2 PIII-1Ghz - 98SE/W2K

Reply 12 of 26, by Nemo1985

User metadata
Rank Oldbie
Rank
Oldbie

I agree, the rise mp6 seemed so ahead of time thanks to his package.
It's also an interesting cpu, very low power consumption (at least the iDragon version I have) and it has the record of superpi 1mb time on my 200 mhz roundup and in doom too! While in other tests wasn't that good.
I'm still undecided if to post the new roundup I made, since I did the same topic some years ago (just with less cpups and less tests).

Reply 13 of 26, by Nexxen

User metadata
Rank Oldbie
Rank
Oldbie
Nemo1985 wrote on 2022-03-11, 12:11:

I agree, the rise mp6 seemed so ahead of time thanks to his package.
It's also an interesting cpu, very low power consumption (at least the iDragon version I have) and it has the record of superpi 1mb time on my 200 mhz roundup and in doom too! While in other tests wasn't that good.
I'm still undecided if to post the new roundup I made, since I did the same topic some years ago (just with less cpups and less tests).

Just post your data. Everything matters.
And if you have pics... that too!

PC#1 Pentium 233 MMX - 98SE
PC#2 PIII-1Ghz - 98SE/W2K

Reply 14 of 26, by BloodyCactus

User metadata
Rank Oldbie
Rank
Oldbie

I have I think a 2A, its still on the interpose which all it does is route power + ground from a far away pin, its labled "Evegreen MXPro", and the missing pin. I've never been able to get the fan off the top off it to check. I was going to throw it in the trash.

--/\-[ Stu : Bloody Cactus :: [ https://bloodycactus.com :: http://kråketær.com ]-/\--

Reply 15 of 26, by swaaye

User metadata
Rank l33t++
Rank
l33t++

Winchip isn't a 486. It's a relatively simplistic in-order RISC design. Maybe Byte did an article on it.

The original is Winchip C6. Winchip 2 improves MMX (dual issue), has 3DNow in the "Winchip 2 3D" version, and improved FPU performance. It's supposed to be about 10% faster per clock in general. Winchip 2+ (A / B?) has double the L1 cache of Winchip 2 / Winchip 2 3D and that should have added another nice speed boost overall.

There is also an essentially unreleased Winchip 3. It's superpipelined and L1 is up to 128K. I think they jumped to Cyrix III instead.

https://books.google.com/books?id=U9n74ngfDU8 … PA144&lpg=PA144
https://books.google.com/books?id=0TsEAAAAMBA … epage&q&f=false

Screenshot_20220311-125259_Firefox_1.jpg
Filename
Screenshot_20220311-125259_Firefox_1.jpg
File size
532.27 KiB
Views
1830 views
File license
Fair use/fair dealing exception

Reply 16 of 26, by Nemo1985

User metadata
Rank Oldbie
Rank
Oldbie

Thank you for the corrections.

I think the Winchip 2-3d was just the initial name of Winchip 2, there isn't any winchip 2 without 3dnow, despite the lost "3d".

According to my tests, that commercial advertising is quite exagerate.
At 200 mhz (66x3), the wattage peak on dos is visible on the fifth post. Compared to others cpus I tested:
Pentium was 36.9 watt, while mmx was 37.9. So quite similar to C6.
K6 was indeed higher to 46,9 and the Cyrix MX (v2.9) was 53,2 watts.
That was the full system calculations since I don't have the right tool to measure the cpu power draw, but since all the others components were the same.

About the cache, there is no difference between the various winchip 2 models.
It can be seen from the test I did and even better on the official documents I posted on the first post (page 11 of every document).
They all have 32kb for instructions and 32kb for data.

Immagine.png
Filename
Immagine.png
File size
80.48 KiB
Views
1804 views
File license
Public domain

That also explain why people think the performance difference from 2 to 2a I got was inside the margin of error, there is no proof according to the datasheets of any architectural difference between 2 and 2a or 2b (other than the lowered voltage of 2b, probably intended to be used mainly on ss7 machines that may not like the old 3.52 voltage).

I agree according to the datasheets they indeed doubled the l1 cache in winchip3, but since it never made to the public, unlucky we will have no chance to test it.

BloodyCactus wrote on 2022-03-11, 12:35:

I have I think a 2A, its still on the interpose which all it does is route power + ground from a far away pin, its labled "Evegreen MXPro", and the missing pin. I've never been able to get the fan off the top off it to check. I was going to throw it in the trash.

Are you willing to post some pictures?

Thanks!

Last edited by Nemo1985 on 2022-03-11, 23:23. Edited 1 time in total.

Reply 17 of 26, by swaaye

User metadata
Rank l33t++
Rank
l33t++

Perhaps that means the Winchip 2+ was actually also abandoned. Maybe in favor of getting to Winchip 3 ASAP.

2B could just be the die shrink mentioned in that PCMag table.

Reply 18 of 26, by Nemo1985

User metadata
Rank Oldbie
Rank
Oldbie
swaaye wrote on 2022-03-11, 23:20:

Perhaps that means the Winchip 2+ was actually also abandoned. Maybe in favor of getting to Winchip 3 ASAP.

2B could just be the die shrink mentioned in that PCMag table.

Yes, it could be like that. 2b was indeed produced at 0.25 µm, instead of 0.35.

I agree with Anonymous Coward, that despite they lowered the voltage the power consumption was almost the same of 2\2a.

Unlucky my 2b is probably faulty and I can't source another, if anyone has one and is willing to share some tests it would be nice.