VOGONS


Reply 40 of 52, by red-ray

User metadata
Rank Oldbie
Rank
Oldbie
Hanamichi wrote on 2020-10-18, 17:06:

Do you have a spreadsheet of CPU benchmark results or know the ballpark figures for a 386DX and 486 33/40?

Thank you for checking out SIV V5.53 BIOS-01, all now looks OK to me. I don't have a spreadsheet, but looking in save files I have. I don't have an SMP 386 or 486 🙁

  Intel 386 Model 0 39MHz

Include Integer Time Floating Time

[Y]CPU-0 4 2.37 3 36.53

Elapsed 39.1 Total 4 + 3 Average 4 + 3 Minimum 4 + 3 Maximum 4 + 3 CPU Time 38.90

Intel 486 DX2 (P24) 65MHz

Include Integer Time Floating Time

[Y]CPU-0 34 2.13 13 7.89

Elapsed 10.1 Total 34 + 13 Average 34 + 13 Minimum 34 + 13 Maximum 34 + 13 CPU Time 10.03

Reply 41 of 52, by Hanamichi

User metadata
Rank Member
Rank
Member
red-ray wrote on 2020-10-18, 18:10:
Intel 486 DX2 (P24) 65MHz […]
Show full quote

Intel 486 DX2 (P24) 65MHz

Include Integer Time Floating Time

[Y]CPU-0 34 2.13 13 7.89

Elapsed 10.1 Total 34 + 13 Average 34 + 13 Minimum 34 + 13 Maximum 34 + 13 CPU Time 10.03[/code]

Happy to help. Thanks for the two benchmarks, this might be great news!

My slow benchmarks were taken with "Internal Cache" disabled in the BIOS.
Normally on a Slot 1 P3 with SetMul if you disable L1 or L2 both get disabled and you get around 386 performance from memory.

But this looks like I'm getting good 486 DX2+ performance and it feels like it in Windows 2K. Made my day!
Is there a way to see if L2 only is disabled with your tool?

386 and 486 SMP that's living the dream 😀

Reply 42 of 52, by red-ray

User metadata
Rank Oldbie
Rank
Oldbie
Hanamichi wrote on 2020-10-18, 18:45:

Is there a way to see if L2 only is disabled with your tool?

I suspect [ Cache-<n> Latency ] (Menu->Hardware->Cache Latency->Cache-<n>) will indicate this, do they?

I just remembered Re: Windows 2000 Disablded the CPU-1 Cache, does anyone know how to fix this please? and it should

Reply 43 of 52, by Hanamichi

User metadata
Rank Member
Rank
Member

Thanks ray, I was thinking that might show the cache behaviour.

Strange that your system and others had that issue with W2K. I think I have the same MPS options in the bios left at Auto.

Awaiting a new PSU, I'll post results when I get it.
(Borrowed my daily driver Nipron one for these tests that I need back. I no longer trust the Enermax PSUs I bought for my retro systems sadly.)

Reply 44 of 52, by red-ray

User metadata
Rank Oldbie
Rank
Oldbie
Hanamichi wrote on 2020-10-19, 14:22:

Awaiting a new PSU, I'll post results when I get it.
(Borrowed my daily driver Nipron one for these tests that I need back. I no longer trust the Enermax PSUs I bought for my retro systems sadly.)

OK, which PSU are you getting? The only Enermax PSU I have is a DigiFanless which seems OK to me, is it?

I to am none too keen of using elderly PSUs and am slowly replacing them with modern ones. I have a Thermaltake SPG-0650D-G in my Dual Intel Pentium IIIE (Coppermine) 1.00GHz and similar in many of my other systems which means I can see how much power they actually use 😀

file.php?id=83130

Reply 45 of 52, by Hanamichi

User metadata
Rank Member
Rank
Member
red-ray wrote on 2020-10-19, 14:54:

OK, which PSU are you getting? The only Enermax PSU I have is a DigiFanless which seems OK to me, is it?

Actually that is one of the best Enermax units they have ever made I think. The Gigabyte looks well featured for a 650W one too, especially the monitoring. The only thing I would say is with a dual PIII, adding a FX5900 Ultra and several drives might push the 3.3v/5v rails. A dual K7 system definately would.

I was referring to the older Enermax units around 1998-2005 which you can sometimes pick up new. Opened mine up and yeah no name caps as well as this issue:
Re: Yet Another PSU Went Pop Thread (NOS edition)

I had a good look at PSUs for my current gen and retro machines. My requirements are a bit tough: (Not made in China, Japanese caps, more then 20a on 3.3/5v for the retro machines, reliability over cost).

Anyway, only 2 brands fit the bill - Nipron and Zippy/Emacs. So I run a big Nipron PSU in my main machine and bought some NOS/used ones for my lower power retro machines. Also a couple of used Zippy/Emacs ones for their strong 3.3v/5v rails. You also have to replace/control the fan as they are intended for the server market.

Both brands are mostly sold B2B so obtaining them new is difficult and expensive. However, they seem to be built like tanks so I am confident using NOS/used ones.
I just didn't feel like ordering a 800W premium PSU (to get something quality) on overclockers only to risk it being defective with a months RMA during these covid times.

Reply 46 of 52, by red-ray

User metadata
Rank Oldbie
Rank
Oldbie
Hanamichi wrote on 2020-10-20, 15:19:

The Gigabyte looks well featured for a 650W one too, especially the monitoring. The only thing I would say is with a dual PIII, adding a FX5900 Ultra and several drives might push the 3.3v/5v rails. A dual K7 system definately would.

Yes, as the Thermaltake SPG-0650D-G is only good for 20A @ +5, but the system already has a pair of SCSI disks and I have no need to add more.

I guess I am lucky as my Dual K7 system powers the CPUs from +12 😀.

file.php?id=78374

Reply 47 of 52, by Hanamichi

User metadata
Rank Member
Rank
Member
red-ray wrote on 2020-10-20, 15:43:

Yes, as the Thermaltake SPG-0650D-G is only good for 20A @ +5, but the system already has a pair of SCSI disks and I have no need to add more.

I guess I am lucky as my Dual K7 system powers the CPUs from +12 😀.

That's true the MPX boards have EPS 12V on them, better then running a single Thunderbird for the rails.
You are about 85W max out of 110W on the 3.3/5V. Does the system have a thirsty AGP GPU and PCI-X cards?
Another SMP system I have in the works will be similar to your Tyan MPX.

Reply 48 of 52, by red-ray

User metadata
Rank Oldbie
Rank
Oldbie
Hanamichi wrote on 2020-10-20, 16:24:

Does the system have a thirsty AGP GPU and PCI-X cards?

The TYAN S2466 (Tiger MPX) is not in a case and I can't remember which GPUs I used back then, I suspect GPU-0 was a GeForce FX 5200 [NV34]. There were no PCI-X cards.

Hanamichi wrote on 2020-10-18, 14:43:

All hardware reading apps get confused when the multi is set low when reporting the FSB and RAM speeds.

Which other apps get it wrong? If you wish to get them fixed you need to tell their developers that it's only >= Stepping 3 Cascades + Coppermine CPUs that use 5-bits rather than 4-bits for the multiplier.

Reply 49 of 52, by Hanamichi

User metadata
Rank Member
Rank
Member
red-ray wrote on 2020-10-20, 17:06:
The TYAN S2466 (Tiger MPX) is not in a case and I can't remember which GPUs I used back then, I suspect GPU-0 was a GeForce FX 5 […]
Show full quote
Hanamichi wrote on 2020-10-20, 16:24:

Does the system have a thirsty AGP GPU and PCI-X cards?

The TYAN S2466 (Tiger MPX) is not in a case and I can't remember which GPUs I used back then, I suspect GPU-0 was a GeForce FX 5200 [NV34]. There were no PCI-X cards.

Hanamichi wrote on 2020-10-18, 14:43:

All hardware reading apps get confused when the multi is set low when reporting the FSB and RAM speeds.

Which other apps get it wrong? If you wish to get them fixed you need to tell their developers that it's only >= Stepping 3 Cascades + Coppermine CPUs that use 5-bits rather than 4-bits for the multiplier.

Ah I see, might be the SCSI drives. It would be great to know how much a Ti4600 and FX5900 Ultra pull from the motherboard respectively.

Hmm every single one I used: SoftFSB, CPUZ, CPUCool etc. Well now I place your tool higher then theirs 😀

I wonder if I could use SIV with an LCD from Matrix Orbital

Reply 50 of 52, by red-ray

User metadata
Rank Oldbie
Rank
Oldbie
Hanamichi wrote on 2020-10-21, 16:42:

Hmm every single one I used: SoftFSB, CPUZ, CPUCool etc. Well now I place your tool higher then theirs 😀

At a guess CPUZ get's it right for Coppermine, but not Cascades. You could try reporting the bug to the CPUZ thread and see if Franck fixes it.

Hanamichi wrote on 2020-10-21, 16:42:

I wonder if I could use SIV with an LCD from Matrix Orbital

IDK, can you plug it into a VGA card and extend the desktop to it? If so see Menu->Tools->Configure->LCD Panel Setup.

http://www.goverlay.com/ can read the SIV sensor data and display it, but I don't know if it works on XP

Reply 51 of 52, by red-ray

User metadata
Rank Oldbie
Rank
Oldbie
Hanamichi wrote on 2020-10-21, 16:42:

It would be great to know how much a Ti4600 and FX5900 Ultra pull from the motherboard respectively.

I booted the system with just then SIS 6326 PCI GPU then added the FX 5200 which looks to take about 2.5 Amps from +3.3

file.php?id=94664

Reply 52 of 52, by Hanamichi

User metadata
Rank Member
Rank
Member
red-ray wrote on 2020-10-22, 09:06:
Hanamichi wrote on 2020-10-21, 16:42:

It would be great to know how much a Ti4600 and FX5900 Ultra pull from the motherboard respectively.

I booted the system with just then SIS 6326 PCI GPU then added the FX 5200 which looks to take about 2.5 Amps from +3.3

Thanks for the stats, was that at idle? It looks like there is a big (edit I misread the screenshot) small difference like you say when you look at the maximum 3.3v rail readings.
I'll take a screenshot when I try a GF4/5 again.

I have dug up no more clues on changing the BSP CPU at will.

Official definition states:

The first CPU to respond becomes the BSP

Each processor broadcasts a BIPI to “all including self.” The first processor that
broadcasts a BIPI (and thus receives its own BIPI vector), selects itself as the BSP
and sets the BSP flag in its IA32_APIC_BASE MSR.

However CPU speed seems to have no impact on the selection. Rebooting multiple times results in the same one selected as well.