ISA slot motherboards from AliExpress

Discussion about old PC hardware.

Re: ISA slot motherboards from AliExpress

Postby RayeR » 2017-9-12 @ 23:55

LSS10999 wrote:Some useful information for ISA chipsets were compiled in this page.

I see, but it's related to old MBs and chipsets, not mention any modern PCI-E based MBs...

LSS10999 wrote:This whitepaper

This looks interesting. But I don't understand this:
The DMA controllers can initiate transfers
between the LPC and host bus. This provides the option of
adding an LPT or fl oppy controller to the LPC port with DMA
support. But the chipset DMA controller cannot initiate
PCI to host bus transactions. To enable ISA-conform DMA,
the ITE 8888 PCI to ISA bridge integrated on the Kontron
MICROSPACE® MSM-eO(-N) features two additional 8237
DMA controllers.


They admit LPC can do ISA DMA - that is what we need for soundcards but then they write "chipset DMA controller cannot initiate
PCI to host bus transactions" - is it needed? Maybe only for this AMD platform but for Portwell Ruby is the LPC2ISA bridge all that is needed and proved it works (at least on G41+ICH7)

LSS10999 wrote:EDIT: Should comment that IMBA-G412ISA is also not quite flexible. It doesn't have any option to tweak CPU settings (multiplier, Vcore, etc.), and it couldn't handle 771-modded Xeons properly. I tried a few Xeons (E5450/E0, X5460/C0, X5470/E0) and while they do boot, the system becomes unstable with program crashes (sometimes the CPU could even miscalculate CRC) and sooner or later BSoD with a random error code.

Yes, industrial MBs don't include overclocking options. Do you have a related microcode in BIOS for your Xeon CPU?

BTW I just found this PC/104 http://c1170156.r56.cf3.rackcdn.com/UK_ADL_MM-SIO-R-01_2DS.pdf expansion board by Adlink using the same Fintek LPC2ISA so if some wiring would be done to LPC header on MB and from PCI/104 pin connector to an ISA slot it might work. But I couldn't find any e-shop offering this board and no idea about price. I would expect a nasty $$$ for industrial HW...
Gigabyte GA-P67-DS3-B3, Core i7-2600K @4,5GHz, 4GB DDR3, 128GB SSD, GF7900GT, SB Audigy + YMF724F + DreamBlaster combo
User avatar
RayeR
Member
 
Posts: 116
Joined: 2007-8-11 @ 13:26
Location: CZ

Re: ISA slot motherboards from AliExpress

Postby LSS10999 » 2017-9-14 @ 01:13

RayeR wrote:I see, but it's related to old MBs and chipsets, not mention any modern PCI-E based MBs...


It did reference the Intel document about the removal of PC/PCI in recent ICH chipsets, stating it's game overfor ISA sound cards unless you only use the synth, which is the case. Most of the page focused on boards and chipsets that are reported working well with ISA sound cards.

RayeR wrote:This looks interesting. But I don't understand this:
The DMA controllers can initiate transfers
between the LPC and host bus. This provides the option of
adding an LPT or fl oppy controller to the LPC port with DMA
support. But the chipset DMA controller cannot initiate
PCI to host bus transactions. To enable ISA-conform DMA,
the ITE 8888 PCI to ISA bridge integrated on the Kontron
MICROSPACE® MSM-eO(-N) features two additional 8237
DMA controllers.


They admit LPC can do ISA DMA - that is what we need for soundcards but then they write "chipset DMA controller cannot initiate
PCI to host bus transactions" - is it needed? Maybe only for this AMD platform but for Portwell Ruby is the LPC2ISA bridge all that is needed and proved it works (at least on G41+ICH7)


LPC DMA remained in order to avoid breaking support for FDC and ECP parallel ports provided by SuperIO chips, which might be required for industrial purposes. If LPC DMA were removed, modern board would NEVER be able to support floppy drives and parallel ports anymore.

Guess that the "chipset DMA controller cannot initiate PCI to host bus transactions" is what breaks most of the legacy compatibility in ICH6 or later. It not only breaks PCI-ISA, it also breaks whatever sound cards used to support legacy DOS audio (except Aureal Vortex series). In modern AMD chipsets (like 700/800/900 series) the situation is much worse that it's behaving the same way as the Intel chipsets without native PCI (some 5 to 7 series PCHs and all PCHs from 8 series onwards), the legacy functions is completely inaccessible even after loading TSRs (and probably under Windows and Linux as well). It's that fact that makes the AMD/AMI's "ISA DMA virtualization" thing interesting.

From the IT8888 datasheet the PCI-ISA bridge itself does feature 8237 DMA controllers of its own, most likely those are what actually mediate the transactions between the ISA devices and the chipset PC/PCI, though I'm not entirely sure. The PCI-ISA bridges' DMA related pins might have been connected to somewhere even in modern (ICH6+ based) motherboards, though it's not easy to find that out on a PCB that complex.

RayeR wrote:Yes, industrial MBs don't include overclocking options. Do you have a related microcode in BIOS for your Xeon CPU?


I did insert microcodes into the BIOS. It's just that the board probably has some underlying BIOS or wiring issues (or maybe power management related) that makes things incompatible with LGA771 Xeons. The board recognizes the Xeon correctly, and it can stay running under DOS and other system diagnostic programs for quite a while (I could run MemTest+ with two or more passes without issues), just it would crash once in a while with a different BSoD under Windows. Sometimes it can even miscompute things, causing CRC and data errors that goes away the next time the program runs, and after crash, keyboards would sometimes stop responding the next boot, requiring a hard reset to get the keyboard back running.

It seems most insudstrial motherboards aren't meant to be used to run out-of-spec stuffs. I also tried modding the BSEL of a Pentium D 965 from 266 (LLL) down to 200 (LHL) for a 865-based motherboard (IMB200) in hope it can boot properly, but still not successful. No matter the mod, the board would either not post or give it an incorrect, non-bootable bus clock (used to be 33MHz, this time I managed to make it 40MHz).

The only purpose for those boards is to provide additional I/O ports that consumer motherboards hardly provide (RS232/485/422, ECP Parallel, GPIO, etc.), yet still very useful in industrial purposes. They only support whatever hardware the document mentions, and that'd be the end of the line.

RayeR wrote:BTW I just found this PC/104 http://c1170156.r56.cf3.rackcdn.com/UK_ADL_MM-SIO-R-01_2DS.pdf expansion board by Adlink using the same Fintek LPC2ISA so if some wiring would be done to LPC header on MB and from PCI/104 pin connector to an ISA slot it might work. But I couldn't find any e-shop offering this board and no idea about price. I would expect a nasty $$$ for industrial HW...


I'm not sure about the possibility given the board itself might be rare. For desktop motherboard, the TPM header, which is often left unused, is connected to the MB using LPC, and might be able to be used for other purposes. It's just how the wiring would turn out...
LSS10999
Member
 
Posts: 119
Joined: 2009-10-28 @ 14:32

Re: ISA slot motherboards from AliExpress

Postby RayeR » 2017-9-14 @ 16:23

LSS10999 wrote:It did reference the Intel document about the removal of PC/PCI in recent ICH chipsets, stating it's game overfor ISA sound cards unless you only use the synth, which is the case. Most of the page focused on boards and chipsets that are reported working well with ISA sound cards.


Yes, I know and refer this document also on my site. But we know that Portwell Ruby is at least one MB based on newer PCI-E chipset with ICH7 that is proved to work fine with Soundblaster (TmEE). So I belive the same approach with LPC2ISA bridge can be replicated on all newer MBs (without use of PCI2ISA bridge). But I didn't found any newer MB with this fintek LPC2ISA bridge... BTW AMD was good with sound HW virtualization long time ago on their's Geode MBs where they emulated SB on AC'97 in SMM via VSA (part of proprietary BIOS binary blob). I have one such thin client Compaq Evo T20: http://rayer.g6.cz/hardware/evo-t20.htm and you can see a short part of the Boost demo running on this PC with sound: http://rayer.g6.cz/hardware/evo-t20/boostdmo.avi
The major problem is that this thin client doesn't have PS/2 KB (it's missing a superIO chip at all) but only USB and I didn't find a working USB driver for DOS. It has crappy proprietary BIOS without SETUP that's missing USB KB legacy emu. It's quite weird it contains SB emu but no KB emu. But it was designed to run Windows NT (embedded) or CE not DOS...

LSS10999 wrote:I did insert microcodes into the BIOS. It's just that the board probably has some underlying BIOS or wiring issues (or maybe power management related) that makes things incompatible with LGA771 Xeons.


Maybe it's also poor VRM design that cause some excessive Vdrops or Vripple on Vcore during heavy load transients. AFAIK those LGA771 Xeons have TDP 125W that is more than desktop C2D or C2Q (65/95W). Memtest SW gerenates quite constant Vcore loading so if it's ballancing on the edge the system may behave stable but as you run real-work apps with highly variable CPU load it may fail. If you have some DSO you can try to monitor Vcore on some pins and compare it with a less TDP CPU.
Gigabyte GA-P67-DS3-B3, Core i7-2600K @4,5GHz, 4GB DDR3, 128GB SSD, GF7900GT, SB Audigy + YMF724F + DreamBlaster combo
User avatar
RayeR
Member
 
Posts: 116
Joined: 2007-8-11 @ 13:26
Location: CZ

Re: ISA slot motherboards from AliExpress

Postby LSS10999 » 2017-9-15 @ 03:24

RayeR wrote:Yes, I know and refer this document also on my site. But we know that Portwell Ruby is at least one MB based on newer PCI-E chipset with ICH7 that is proved to work fine with Soundblaster (TmEE). So I belive the same approach with LPC2ISA bridge can be replicated on all newer MBs (without use of PCI2ISA bridge). But I didn't found any newer MB with this fintek LPC2ISA bridge... BTW AMD was good with sound HW virtualization long time ago on their's Geode MBs where they emulated SB on AC'97 in SMM via VSA (part of proprietary BIOS binary blob). I have one such thin client Compaq Evo T20: http://rayer.g6.cz/hardware/evo-t20.htm and you can see a short part of the Boost demo running on this PC with sound: http://rayer.g6.cz/hardware/evo-t20/boostdmo.avi
The major problem is that this thin client doesn't have PS/2 KB (it's missing a superIO chip at all) but only USB and I didn't find a working USB driver for DOS. It has crappy proprietary BIOS without SETUP that's missing USB KB legacy emu. It's quite weird it contains SB emu but no KB emu. But it was designed to run Windows NT (embedded) or CE not DOS...


That's impressive, even though the VSA might be very chipset-specific. Dunno if VSA includes support for FM OPL though, I couldn't find anything from the geode-vsa source code, only about the audio (XpressAUDIO), probably there isn't. From the demo the SB16 emulation sounds excellent.

RayeR wrote:Maybe it's also poor VRM design that cause some excessive Vdrops or Vripple on Vcore during heavy load transients. AFAIK those LGA771 Xeons have TDP 125W that is more than desktop C2D or C2Q (65/95W). Memtest SW gerenates quite constant Vcore loading so if it's ballancing on the edge the system may behave stable but as you run real-work apps with highly variable CPU load it may fail. If you have some DSO you can try to monitor Vcore on some pins and compare it with a less TDP CPU.


Apparently it's not just the 120W ones. The 80W E5450 also fails after a while. It's possible that there really are some VRM issues with the board causing problems with those modded LGA771 CPUs. LGA775 ones (like Q9650, 95W) work fine and stable though. Replacing the PSU with a more reliable one eased the BSoDs to the point that it'd only crash while inside the OS and could live through the OS installation process.

The board also doesn't have features to disable SpeedStep, which might be the cause of the fluctuating readings of Vcore under CPUZ, and the possible cause of all those BSoDs, even when I manually set minimum performance to 100% in Power Options. The undervolting still happens even when the CPU clock is kept at maximum (probably due to Vdrops). When under real workload, the CPU would randomly miscalculate and corrupt things, such as detecting corruptions when there isn't (like driver installers reporting CRC errors when the file was correct, and the installer extracts successfully upon the next attempt), causing memory-related BSoDs, etc.

EDIT: Happened to find a document about ISA DMA Support for congatec conga-EAF (an ETX computer-on-module based on AMD G-series APU) which mentioned that AMD's ISA DMA emulation uses SMIs (system management interrupts). So it's possible the legacy of VSA is still there on some modern embedded solutions, and that it has always been possible to emulate such through the use of SMIs to bridge legacy compatibility, just that the nature of SMI may result it impractical for certain industrial uses that requires strict timing.
LSS10999
Member
 
Posts: 119
Joined: 2009-10-28 @ 14:32

Re: ISA slot motherboards from AliExpress

Postby RayeR » 2017-12-02 @ 20:15

Hi,
as I mentioned about LPC here some time ago I have some progress on my project of LPC bus SuperIO expander http://rayer.g6.cz/hardware/lpc_sio.htm
I'm waiting for 5 PCBs from China now.
The question is if and how various OS can work with floppy controller when SuperIO is set to nonDMA mode (I assume that I will be able to configure it for usual legacy IO address so only difference will be in DMA because TPM header doesn't have LDRQ# signal routed on a pin). Is there some kernel config or patches at least for Linux? Maybe it will require a TSR or BIOS ROM module extension for DOS.
I know that HW part is quite easy but FW/SW work will be more complicated. It's just a testing prototype and challenge for me to make floppy live again in a modern PC (no I don't like floppy on USB).
Gigabyte GA-P67-DS3-B3, Core i7-2600K @4,5GHz, 4GB DDR3, 128GB SSD, GF7900GT, SB Audigy + YMF724F + DreamBlaster combo
User avatar
RayeR
Member
 
Posts: 116
Joined: 2007-8-11 @ 13:26
Location: CZ

Re: ISA slot motherboards from AliExpress

Postby RayeR » 2017-12-02 @ 20:29

LSS10999 wrote:EDIT: Happened to find a document about ISA DMA Support for congatec conga-EAF (an ETX computer-on-module based on AMD G-series APU) which mentioned that AMD's ISA DMA emulation uses SMIs (system management interrupts). So it's possible the legacy of VSA is still there on some modern embedded solutions, and that it has always been possible to emulate such through the use of SMIs to bridge legacy compatibility, just that the nature of SMI may result it impractical for certain industrial uses that requires strict timing.


As I read that document I think they only used SMI for emulating ISA DMA but not a complete VSA SB emulation. (VSA doesn't require real SB it's emulating on onboard AC'97 codec). I think that VSA is quite a complicated, large and system dependent piece of code that would require a lot of programmers work to maintaind and support new platforms. As customers are mostly not interested/not paying enough for such feature the manufacturer cut the cost of devel. and derecate the VSA. That's a pity they didn't released it as an opensource so it might be included into Coreboot project and volunteeers could continue developing it. But it's sad reality of current closed hw/sources world...
Gigabyte GA-P67-DS3-B3, Core i7-2600K @4,5GHz, 4GB DDR3, 128GB SSD, GF7900GT, SB Audigy + YMF724F + DreamBlaster combo
User avatar
RayeR
Member
 
Posts: 116
Joined: 2007-8-11 @ 13:26
Location: CZ

Re: ISA slot motherboards from AliExpress

Postby LSS10999 » 2017-12-04 @ 06:19

RayeR wrote:
LSS10999 wrote:EDIT: Happened to find a document about ISA DMA Support for congatec conga-EAF (an ETX computer-on-module based on AMD G-series APU) which mentioned that AMD's ISA DMA emulation uses SMIs (system management interrupts). So it's possible the legacy of VSA is still there on some modern embedded solutions, and that it has always been possible to emulate such through the use of SMIs to bridge legacy compatibility, just that the nature of SMI may result it impractical for certain industrial uses that requires strict timing.


As I read that document I think they only used SMI for emulating ISA DMA but not a complete VSA SB emulation. (VSA doesn't require real SB it's emulating on onboard AC'97 codec). I think that VSA is quite a complicated, large and system dependent piece of code that would require a lot of programmers work to maintaind and support new platforms. As customers are mostly not interested/not paying enough for such feature the manufacturer cut the cost of devel. and derecate the VSA. That's a pity they didn't released it as an opensource so it might be included into Coreboot project and volunteeers could continue developing it. But it's sad reality of current closed hw/sources world...


The VSA has been open-sourced in 2006 under LGPL, with coreboot having its own OpenVSA based on it. However, it appears to be far from complete and there were little intent to extend this feature to modern chipsets. I think it's merely for supporting VSA-capable Geode systems on coreboot.

There's little need for Sound Blaster and other stuffs on those embedded modules, but ISA DMA might be needed in some appliances (which is why Intel 8xx industrial boards are still available to some extent, for customers who specifically need ISA DMA for certain devices). So far only AMD/AMI has the ability to utilize SMI to replicate ISA DMA functionalities which eliminates DMA related limitations at the expense of system responsiveness.

I'm not sure about the inner details of the VSA, but from the code it definitely looks like being highly chipset-specific as given the processing power of the CPUs at that time, it might not be practical to create such in a higher-level, more universal manner.
LSS10999
Member
 
Posts: 119
Joined: 2009-10-28 @ 14:32

Re: ISA slot motherboards from AliExpress

Postby RayeR » 2017-12-04 @ 11:31

LSS10999 wrote:The VSA has been open-sourced in 2006 under LGPL, with coreboot having its own OpenVSA based on it. However, it appears to be far from complete and there were little intent to extend this feature to modern chipsets. I think it's merely for supporting VSA-capable Geode systems on coreboot.


Aha, I didn't know. OK, it seems to be too complex and complicated that nobody was able to port it to support intel HDA chips and intel ICH/PCH :(
Maybe it also requires some special HW feature of AMD Geode chipset that inte ICH/PCH is missing...
Gigabyte GA-P67-DS3-B3, Core i7-2600K @4,5GHz, 4GB DDR3, 128GB SSD, GF7900GT, SB Audigy + YMF724F + DreamBlaster combo
User avatar
RayeR
Member
 
Posts: 116
Joined: 2007-8-11 @ 13:26
Location: CZ

Previous

Return to General Old Hardware

Who is online

Users browsing this forum: Google [Bot], hard1k, reignofchaos and 7 guests