VOGONS


First post, by blueclouds8666

User metadata
Rank Newbie
Rank
Newbie

Hi, I've been recently doing some testing to my Pentium III Coppermine machine with XP Professional SP3. I have been experiencing for quite a while a problem in regard to performance consistency so i decided to investigate what was the cause behind it. It seems the more USB devices i plug into the machine, the slower it runs, and the impact is specially notable while playing games. To demonstrate this issue, I've set up different hardware configurations and run PrBoom OpenGL intro scene as a way to benchmark the performance. The followed procedure has been the same for each hw configuration: run PrBoom at 1024x768 windowed, and set a timed 30s benchmark with Fraps as soon as the intro scene started. This are the results:

1. GeForce card + Ensoniq audio PCI card, no USB devices connected: the fastest i could get it to run with audio
MIN: 59, MAX: 93, AVG: 77.367

2. GeForce card + Sound Blaster Live USB audio card: slightly slower
MIN: 53, MAX: 84, AVG: 71.200

3. GeForce card + Sound Blaster Live USB audio card + tp-link 802.11b/g/n USB adapter: worse
MIN: 44, MAX: 78, AVG: 61.600

4. The same as 3, but now the tp-link is connected to a VIA USB 2.0 PCI port instead of the onboard main panel: even worse
MIN: 27, MAX: 52, AVG: 39.200

5. The same as 4, but now two usb flash drives are as well connected to the VIA PCI: unplayable
MIN: 17, MAX: 39, AVG: 25.000

I've used PrBoom GL in this case, but the performance impact is as noticeable with any other game. My motherboard comes with two USB 1.1 onboard ports, which cause less peroformance impact in contrast with the VIA PCI, but yet it's slow enough to make some games unplayable (in tests 2 and 3, onboard USB ports were used. VIA card is only connected in tests 4 and 5).

I'm unsure if this performance impact is normal, being an old P3 at 800 MHz, or if it's either a hardware conflict, BIOS bug or some bad misconfiguration... That's why i'm asking here, as i'm still intrigued why is this happening. Thank you very much in advance for your help.

Reply 1 of 13, by emosun

User metadata
Rank Oldbie
Rank
Oldbie

I think you might be just hitting an operating system issue.

I use WFFLP on my older configs that can run it and don't experience this issue. The way something like xp SP3 uses the usb ports might have changed significantly vs 2001 or vs 2006 when WFFLP was made.

If I had to make a pure guess it's just expecting a way bigger processor or way higher bus speeds somewhere that a p3 system cannot provide. Maybe it's an instruction set the cpu is missing. One thing I would try in your position , is disabling the onboard usb and trying a usb 1 card (yes those exist). Maybe the card simply being an older standard on a different bus will effect something.

But honestly I bet this will fix itself if you ran xp sp1 , sp2 or WFFLP

Reply 2 of 13, by Schule04

User metadata
Rank Member
Rank
Member

Which mainboard and Geforce card are you using? Did you update to the newest BIOS and drivers? That definitely does not sound normal.

You would get better performance with Win98Se or WinME. WinXP has a lot of overhead on Pentium 3 and older

Reply 3 of 13, by cyclone3d

User metadata
Rank l33t++
Rank
l33t++

The old.. USB sucks performance was a known issue back when it was new-ish.

What IRQ is USB using? What other devices are sharing that same IRQ?

What is the polling rate set to for USB? Default is 500Hz. If you lower the polling rate, it should reduce the performance hit of USB but also reduce the responsiveness of USB peripherals.

To top it off, when real MIDI ports went away and all the mfgs moved to USB MIDI adapters, there were major issues with synchronization and latency due to how USB works. One thing that helped was to increase the USB polling rate.
The other was to disable CPU core parking via a registry setting.

You are also running XP on a system that is more suitable for Windows 9x.

The constant USB polling and interrupt requests are going to have a performance hit on a system like that.

Yamaha YMF modified setupds and drivers
Yamaha XG resource repository - updated November 27, 2018
Yamaha YMF7x4 Guide
AW744L II - YMF744 - AOpen Cobra Sound Card - Install SB-Link Header

Reply 4 of 13, by auron

User metadata
Rank Oldbie
Rank
Oldbie

ditch the usb live! card for an actual pci live! card and ditch the usb 2.0 card altogether - its performance impact should speak for itself and there's numerous other threads about this, so it's not just you. i'd also suggest going with an actual ethernet card with SMB or FTP for file sharing. an 11n usb adapter is much newer than pentium iii so it's only expected that the drivers are built with faster CPUs in mind. finally sp3 of xp is said to be much heavier than vanilla or sp1, so overall you're really dealing with a suboptimal configuration for your hardware.

Reply 5 of 13, by Horun

User metadata
Rank l33t
Rank
l33t

Agree with cyclone3d and auron ! USB 1.1 was horrid for cpu/system resource usage. Not a problem if you were just printing to a USB printer while the Office app was sitting idle but actually trying to game with anything USB also going on is a bad thing.
Running XP on a P3 is really pushing limits.

Hate posting a reply and then have to edit it because it made no sense 😁 First computer was an IBM 3270 workstation with CGA monitor.

Reply 6 of 13, by auron

User metadata
Rank Oldbie
Rank
Oldbie

i actually didn't have any perceivable issues with a usb mouse even on a k6/233 setup (HX chipset usb, of course no silly USB cards), but you could be right and there is a performance difference. just not sure about how to go and benchmark this exactly...

Reply 8 of 13, by The Serpent Rider

User metadata
Rank l33t
Rank
l33t

Some driver issues which hog CPU time too much, most likely USB Sound Blaster. OS itself has nothing to do with it. Pentium 3 itself also has nothing to do with it. P3 is more than capable to run Win2k/XP.

I must be some kind of standard: the anonymous gangbanger of the 21st century.

Reply 9 of 13, by blueclouds8666

User metadata
Rank Newbie
Rank
Newbie

Thank you for all your suggestions! i didn't expect such quick response

I think you might be just hitting an operating system issue. I use WFFLP on my older configs that can run it and don't experience this issue. The way something like xp SP3 uses the usb ports might have changed significantly

I also had that thought, but now that you tell me, i'll give a try to XP SP1 and/or WinFLP to see if there's a difference.

Which mainboard and Geforce card are you using? Did you update to the newest BIOS and drivers?

Tried with two different AGP cards, a GeForce 4 series and an FX series, but both present the same issue. I'd prefer not risking on BIOS updates, specially since this motherboard model seems a bit uncommon and finding replacement P3 motherboards nowadays is quite hard and often expensive.

What IRQ is USB using? What other devices are sharing that same IRQ?

I remember checking there was no IRQ conflicts when using the PCI slot 2, but yet the performance issue was still present. that was time ago so i should probably give that another check. I'll also try switching the polling rate to see if that helps.

Some boards also have obviously buggy "legacy usb" implementations but I don't think its relevant here, op did not mention using a mouse or keyboard with USB

Yes i'm using both PS/2 mouse and keyboard.

Some driver issues which hog CPU time too much, most likely USB Sound Blaster.

The thing is, using the ensoniq PCI sound card + any other USB device connected, be it flash drive or wifi adadpter presents a performance impact as well. also the driver used for the Sound Blaster USB is the one that comes with the operating system (it's a standard sound card), not the official sound blaster one which i'm aware is not very good.

I'll tune you up once i try your suggestions.
Regards

Reply 10 of 13, by hyoenmadan

User metadata
Rank Member
Rank
Member

VIA cards on retro configurations are absolutely HORRID for anything but non intensive communications like mouse/kb (and sometimes even such usage causes problems). For its legacy USB1.2 implementation it uses UHCI design, very sensible to PCI chipset implementations, and which does all the controller hardwork in the host CPU (these controllerless crap you know very well with winmodems and winprinters). Its USB2 EHCI implementation requires MSI/MSIX style interrupts to even work, or work properly.

NEC cards sometimes work a bit better in certain situations, as these are based in the OHCI specification design for USB1.2, which mandates a microcontroller on chip to offload all the cpu intensive task from the CPU (a bit like Firewire OHCI). And its USB2 implementation supports legacy interrupts in older chip revisions. So this is the only card to use in P3 and donwards if you really need USB in your retro configuration. But even in this case, you still will get a bit slow.

Reply 11 of 13, by cyclone3d

User metadata
Rank l33t++
Rank
l33t++

IRQ sharing is not the same thing as IRQ conflicts. You will almost always have multiple devices using the same IRQ.

Watch the boot information screen where it shows that info. It will tell you exactly what devices are sharing IRQs

Yamaha YMF modified setupds and drivers
Yamaha XG resource repository - updated November 27, 2018
Yamaha YMF7x4 Guide
AW744L II - YMF744 - AOpen Cobra Sound Card - Install SB-Link Header

Reply 12 of 13, by blueclouds8666

User metadata
Rank Newbie
Rank
Newbie

I have installed Windows 2000 SP4 on a different hard drive to determine if its an XP SP3 issue. I later had to switch the wireless USB device to a different one because there are no W2000 drivers for the tp-link. This are the results:

1. GeForce card + Ensoniq audio PCI card, no USB devices connected (Win2000)
MIN: 68, MAX: 96, AVG: 82.533

2. GeForce card + (onboard USB: Sound Blaster Live USB audio) + (to VIA PCI USB: belkin N1 wireless USB + two USB flash drives) (Win2000)
MIN: 50, MAX: 70, AVG: 59.467

I noticed a significant difference with the results thrown by the base run under XP, likely because it was running on a different GPU driver version, different wireless device, and some other PrBoom settings i might have missmatched. For better accuracy, i've decided to re-run the XP benchmark using the exact same driver version (forceware 93.71), exact same wireless device and driver (belkin N1), exact same PrBoom settings and version (PrBoom Plus 2.5.1.4), and exact same Fraps version (3.0.3). Now the results make a bit more sense:

1. GeForce card + Ensoniq audio PCI card, no USB devices connected (WinXP)
MIN: 71, MAX: 104, AVG: 86.767

2. GeForce card + (onboard USB: Sound Blaster Live USB audio) + (to VIA PCI USB: belkin N1 wireless USB + two USB flash drives) (WinXP)
MIN: 47, MAX: 72, AVG: 57.033

The results pretty much tell me there is no considerable performance difference between both OSes, which leads me to think is not an OS problem, although further tests with Windows 2000 RTM/Windows 9x would be needed to confirm that. I've check the IRQs for each and every benchmark run, and no device had conflicts nor shared addresses with others. they all have an individual IRQ address for themselves.

an 11n usb adapter is much newer than pentium iii so it's only expected that the drivers are built with faster CPUs in mind.

i think that's the case, as it seems the belkin causes a much smaller performance impact in contrast with the newer tp-link, yet it's the device that slows down the most.

VIA cards on retro configurations are absolutely HORRID for anything but non intensive communications like mouse/kb (and sometimes even such usage causes problems).

i'm going to try getting a NEC USB card if i can find one at a reasonable price. if there's a big performance difference i'll let you know.

In regard to the USB polling rate, it seems it's only relevant for USB pointers such as mice, but all my input devices are PS/2 so switching the rate wouldn't make a difference (correct me if i'm wrong). Reading your replies, it seems this level of performance impact achieved with the new benchmark runs is normal with my current hardware setup, considering the unoptimized stuff i plugged in (VIA PCI USB card and the wireless usb adapters). I guess i'll have a second eye on which devices i have plugged before running resource intensive applications. Thanks again for all your help.

Reply 13 of 13, by auron

User metadata
Rank Oldbie
Rank
Oldbie
blueclouds8666 wrote on 2021-04-07, 23:57:

The results pretty much tell me there is no considerable performance difference between both OSes, which leads me to think is not an OS problem, although further tests with Windows 2000 RTM/Windows 9x would be needed to confirm that.

this article may be of interest: https://www.igorslab.de/en/games-performance- … indows-directx/

seems like OS performance is a bit of a wash - old stuff runs better on older OS and new stuff on newer.