PARKE wrote:

Just downloaded the last version above here. Intel 486DX2-66 (SX750) is recognized as an AMD chip. Biostar MB-8433UUD-A with Windows 95.

DX2-66aa.jpg

Thanks for the test.
Was an AMD CPU installed before the Intel one ?

CuPid wrote:

Thanks for the test.
Was an AMD CPU installed before the Intel one ?

Yes, Win 95 was installed when there was still an AMD P75 in there.
Speedsys reports [CPUID not support].

Finally got my other computer setup with Windows 98 (My AthlonXP was too unstable).

Had to jump through alot of hoops, bad hardware, bios settings, KVM, keyboard, mouse issues etc etc but here it is from a stable C2Q Q6600 running Windows 98SE heh.

I know it doesn't match the topic but thought it would be interesting.

Missing memory timings on memory tab (Likely 9x issue since works fine in Windows 10)
No name and manfucturer for Geforce 4 MX 420 on Graphics tab

CPU-Z sees the GTX 780 but obviously that doesn't work in Windows 98 so don't worry about it.

DosFreak wrote:

Finally got my other computer setup with Windows 98 (My AthlonXP was too unstable). […]
Show full quote

Finally got my other computer setup with Windows 98 (My AthlonXP was too unstable).

Had to jump through alot of hoops, bad hardware, bios settings, KVM, keyboard, mouse issues etc etc but here it is from a stable C2Q Q6600 running Windows 98SE heh.

I know it doesn't match the topic but thought it would be interesting.

Missing memory timings on memory tab (Likely 9x issue since works fine in Windows 10)
No name and manfucturer for Geforce 4 MX 420 on Graphics tab

CPU-Z sees the GTX 780 but obviously that doesn't work in Windows 98 so don't worry about it.

Thanks !

Here are results from my NexGen Nx586-P90:

Bench with idoff:

Bench with idon:

Most notable is that cpu-z sees a FPU, whereas the CPU has no FPU. Also the bench shows clearly that x87 emulation is used instead, most likely through win87em.dll.
CPU Frequency is correct.
L2 cache of 256 kB is missing.
idoff shows a noticable speedup in the bench.

Saved text file in attachment:

Thanks elianda.
Yes the x87 emulation is detected as an FPU, I'm working on a fix

Here is a new version with several fixes :
- x87 emulation is not detected as an FPU anymore.
- UMC8881 support (for L2 cache).

I'm looking for testers with :
- UMC/OPTI 82C391/82C392, 82C481/82C482 and 82C596/82C597. The software may report erroneous information about chipset name & L2, but I'm interested in getting a TXT report.
- Ti SXL/SXL2 (again, I could not get mine working)

http://download.cpuid.com/betas/cpuz_w95.zip

Thanks !

Ok, the emulated FPU is gone.

Maybe for the instructions list 'NXOP' could be added?

CuPid wrote:

- UMC8881 support (for L2 cache).

I posted some info on UMC 888x a while back that may be helpful if you want to add more details Re: What retro activity did you get up to today?

I have tested CPU-Z W95 on the Soyo SY-019I (http://retronn.de/imports/hwgal/hw_386eteq.html) with a 386DX-40 + i387-33 (at 40 MHz)

With Turbo-Off:

Remarks: CPU benchmark shows a factor 2 lower performance than the profile data for 386DX-40. My system is performing ok.

Is the benchmark memory intense? Maybe it benches swapping instead of actual the CPU?

I also have two 486 systems using UMC chipset, though one has no Win9x installed.
The other has 8 MB and I wonder if CPU-Z will even run on Win95 with 8 MB RAM and the benchmark tab would give any usable result?

@bakemono : thanks, this can indeed be very useful.

@elianda : thank you for the test, I think there is a way to tell if a 386 is Intel or AMD, if this is the case I'll add a fix.
the benchmark is a simple dhrystone/whetstone. AFAI could see, the dhrystone score varies with the cache hierarchy, but not much with memory speed (maybe more on 386 w/o L1).

No luck with the Eteq, no information is read at all 😢

CuPid wrote:

@elianda : thank you for the test, I think there is a way to tell if a 386 is Intel or AMD, if this is the case I'll add a fix.
the benchmark is a simple dhrystone/whetstone. AFAI could see, the dhrystone score varies with the cache hierarchy, but not much with memory speed (maybe more on 386 w/o L1).

No luck with the Eteq, no information is read at all 😢

Ok, my 386DX-40 shown here uses 64 kB L1 cache on the mainboard. With what config the reference 386DX-40 score was measured? Even if one increase cache to 128 kB it does not result in doubled Dhrystones.

Currently for me the main limitation for my 386/486 systems is that CPU-Z requires Win9x which prevents me testing on more systems.

elianda wrote:

CuPid wrote:

@elianda : thank you for the test, I think there is a way to tell if a 386 is Intel or AMD, if this is the case I'll add a fix.
the benchmark is a simple dhrystone/whetstone. AFAI could see, the dhrystone score varies with the cache hierarchy, but not much with memory speed (maybe more on 386 w/o L1).

No luck with the Eteq, no information is read at all 😢

Ok, my 386DX-40 shown here uses 64 kB L1 cache on the mainboard. With what config the reference 386DX-40 score was measured? Even if one increase cache to 128 kB it does not result in doubled Dhrystones.

Currently for me the main limitation for my 386/486 systems is that CPU-Z requires Win9x which prevents me testing on more systems.

I think it's my 386. I have M321 board with 256kB of cache + write-back enabled. (One additional caching chip, so total 8+tag+writeback = 10 chips) I have tuned the bios settings as much as possible and even changed some register values in bios with special software that are not available in it's bios itself. (for example enabling slow memory refresh)

So take it as very well tuned 386. Yet, i would expect total gain over your setup to be maybe 10%, not 100%...

Yes this is GigAHerZ score (15), and it matches what I got on a 486DLC-33 (~19) and a 486DLC-40 (~21).

elianda, the problem could come from the ref timer used to compute the time in the benchmark, you can check with Tools -> Timers if any of the timer is not well calibrated. The bench uses QPC, for information.

The timers are good. The issue is that the Dhrystone routine you use is extremely sensitive to enabled/disabled cache. Of course disabled cache gives a performance hit, but not like 50% performance on typical applications.

So my 386 system has 32 MB RAM where the first 16 MB get cached due to chipset limitations. I did the benches again with limiting RAM in Win95 to 16 MB using HIMEM 2.77G instead of 3.09.

Cached:

Uncached:

I repeated the bench several times to see if it shows really stable.

So disabling cache drops the score from 14.8 to 6.3 (42.5%). Seems overestimated.

elianda
I spoke too quickly, it seems that no reliable method exists that can tell if a 386 is Intel or AMD.

Concerning the Nexgen, I found no way to read the L2 size. However, I added the reading of the config registers, may I ask you to send a TXT report with that version on the Nx586 ?
http://download.cpuid.com/betas/cpuz_w95.zip

Thanks !

There you go.

Thanks. Those are your config registers, for information.

1	
2	CFG0			0x831051FF
3	CFG1			0x1
4	Port[0x92]	0xFC
5	Port[0x61]	0x24

Yes thanks, look like good performance settings. Most interesting is probably the bit 12 in CFG0 that can be translated with the FSB clock to memory timings according to the table on page 104.

feipoa wrote:

For NexGen information, I suggest you contact the guy who compiled this website, https://www.memotech.franken.de/NexGen/
He goes by mtx500 on the CPU-World forum.

I got a reply from mtx500.

What you would need is the "NxPC System Controller Databook" == "NexGen NxPC System Logic Data Book" 170 pages, June 23, 1993.
Warthman Associates wrote it, see
https://www.warthman.com/projects-NexGen-NxPC … c-Data-Book.htm

According to him, only AMD can provide a copy of the book today. It is worth trying, anyway.