Very old motherboard usb/sata options

[Kizoto]

Hi, I have KT7A-Raid with a 1GHz thunderbird CPU, this motherboard only has PCI+ISA (not pci-e), usb1 and 4 ide ports and I am wondering what options I have to improve i/o speeds. I am aware PCI maxes out at ~266mbps, but I still think there are benefits to be had if I can get some usb2 ports added along with some kind of SSD (for the random access, not the full throughput speed).

 

Ideally I could just get a PCI card with has both usb2 ports and sata (whatever the generation), but searching for old PCI cards is hard as 99% of the results are just pci-e cards.

 

Another option could be a pci->pci-e adapter along with an nvme ssd adapter card. I think I would be pushing my luck trying to boot from something like this though.

 

I have tried IDE->Sata adapters and found them not to be reliable.

 

Does anyone have some suggestions for what to look for?

[da na]

Nice board!
Pretty sure PCI NVME isn't a thing, the likelihood of that working is slim because of how differently PCI and PCIE are handled on the board. I'd recommend a USB 2.0 PCI card and a SATA PCI card, they are definitely out there. 

[Needfuldoer]

I haven't heard of combination USB 2.0 and SATA PCI cards, but each one definitely exists individually. If you have two PCI slots to use up, get a USB 2.0 card, a SATA card, and a used SATA SSD.

 

If the motherboard's IDE controller doesn't support 48-bit LBA, you'll be limited to 137 GB per physical drive. A 128 gig mSATA SSD in an mSATA to IDE bridge would be my choice for upgrading its storage without using up a slot.

[mariushm]

PCI maxes at 133 MB/s  ( 33 Mhz x 32 bit  = 33,333,000 x 32 bits = /8 bits in a byte = 132,000,000 bytes = 132 MB-ish .

The bandwidth is shared between all PCI slots. 

 

You would use that motherboard with Windows 7 at best, maybe Windows XP ... so you'll need to buy cards that still have support for that operating system. 

 

It looks like in at least 3 pci slots you can plug PCI-X cards which are normally used for servers. PCI-X is PCI with an extension to make the card running with 64 bit transfers instead of 32 bit (besides supporting 66 Mhz and higher clock speeds), but majority of cards will fall back to 32 bit in regular pci slots and still work fine. 

You can find lots of SATA 2 pci-x controller cards on eBay and other places, with good controller chips, better than cheapest via chipset cards on PCI slots. 

 

For example this $15 3ware 9550sxu - 3Ware 9550SXU-4LP PCI-X SATA II (3.0Gb/s) RAID Controller Card  - is supported in Windows 2003, and Windows 2008 and Windows 7 (if you have the latest firmware in the card) according to this page  : https://www.broadcom.com/support/knowledgebase/1211161499498/operating-systems-supported-for-the-lsi-3ware-and-megaraid-contr

 

I would risk wasting $15 (plus shipping) to see if it works in a regular PCI slot.

 

For USB ... can't say much, I'd say try to look for a card with a NEC or Intel controller chip, not VIA, VIA was usually more buggy.  But it will be tough, because most search results will give you pci-e usb adapters.

 

EDIT : apparently I'm incorrect ... the 3ware card above is keyed for  PCI 3.3v only slots ... you have 5v PCI slots so it won't work in your motherboard. 

But it's still worth looking for one that can be backwards compatible with PCI 5v  or PCI 3.3v / 5v 

 

PCI 3.3v  [----|---------]   PCI 5v  [---------|----]     PCI 3.3v/5v  [---|------|----]

 

The 3ware card is PCI-X with 3.3v only backwards compatibility  [----|---------][----]  

 

Sorry.

 

edit 2:   This $30  (new sealed box) Highpoint RocketRAID 8 port sata 2 uses the 3.3/5v pci-x, so it would fit in pci slots, but can't be 100% certain it will actually function in the pci slot. The chip (intel io80331 seems to support regular pci 32 bit)

Link : https://www.ebay.com/itm/284334496745

 

[Beerzerker]

Don't forget about the AGP slot for graphics, you can normally find a decent AGP card cheap enough.

[Kizoto]

I have a geforce 2 gts, a 1gb network card and a sound card too. Unfortunately the board has become quite difficult to post, it probably needs a recap so it's put a bit of a delay on trying the stuff suggested. Thanks for the help so far though, the motherboard has space to fit 64bit length cards so ill look into that once I have it posting reliably again.

[Beerzerker]

9 hours ago, Kizoto said:

I have a geforce 2 gts, a 1gb network card and a sound card too. Unfortunately the board has become quite difficult to post, it probably needs a recap so it's put a bit of a delay on trying the stuff suggested. Thanks for the help so far though, the motherboard has space to fit 64bit length cards so ill look into that once I have it posting reliably again.

Length of the card isn't what makes it 64 bit capable - Just sayin.
In fact the entire system isn't 64 bit capable (32 bit capable max) so it's a moot point even if the card in use was capable of it.

Since you already have a GF2 GTS use that at first and see how it does once you get whatever work it needs done. 

[mariushm]

The reason I suggested PCI-X cards is because:

1. they're less searched for, so because of less demand there's also lower prices

2. they usually have better controllers and most often better drivers - generic PCI sata controllers will have the cheap VIA and JMicron or maybe AsMedia if you're lucky.  Also, on NEW adapters, you tend to get pci to pci-e bridge chips and then a cheapo  pci-e x1 2 port sata controller from AsMedia or whatever. The pci to pci-e bridge chips can often cause problems like not being able to boot.

 

The actual transfer speed is the least concern because whatever CPU you have with such motherboard, it's not powerful anyway to work with 100 MB/s+ read/write speeds.

 

They'll also have better support for large size drives - for example the HighPoint card I linked above supports drives bigger than 2 TB and supports NCQ. Some regular PCI adapters don't support both.

 

The 64 bit is not about the CPU being 64 bit, or the operating system being 64 bit, is about how many bits are transferred through the PCI / PCI-X bus in one cycle, one of those 33.3 Mhz 

The standard PCI is 33.3 Mhz x 32 bits per cycle = 133 MB/s.   PCI-X can do  [ 32 bits OR 64 bits ]  x [ 33.3  OR 66.6 OR 100 OR 133 ] Mhz  but  depending on controller and firmware on the card, there's nothing preventing the card from running in 32 bit mode.

 

So you can use a 32 bit only processor (if you manage to find one) with PCI-X running 64 bit wide, no problem.

 

 

 

[Beerzerker]

25 minutes ago, mariushm said:

The reason I suggested PCI-X cards is because:

1. they're less searched for, so because of less demand there's also lower prices

2. they usually have better controllers and most often better drivers - generic PCI sata controllers will have the cheap VIA and JMicron or maybe AsMedia if you're lucky.  Also, on NEW adapters, you tend to get pci to pci-e bridge chips and then a cheapo  pci-e x1 2 port sata controller from AsMedia or whatever. The pci to pci-e bridge chips can often cause problems like not being able to boot.

 

The actual transfer speed is the least concern because whatever CPU you have with such motherboard, it's not powerful anyway to work with 100 MB/s+ read/write speeds.

 

They'll also have better support for large size drives - for example the HighPoint card I linked above supports drives bigger than 2 TB and supports NCQ. Some regular PCI adapters don't support both.

 

The 64 bit is not about the CPU being 64 bit, or the operating system being 64 bit, is about how many bits are transferred through the PCI / PCI-X bus in one cycle, one of those 33.3 Mhz 

The standard PCI is 33.3 Mhz x 32 bits per cycle = 133 MB/s.   PCI-X can do  [ 32 bits OR 64 bits ]  x [ 33.3  OR 66.6 OR 100 OR 133 ] Mhz  but  depending on controller and firmware on the card, there's nothing preventing the card from running in 32 bit mode.

 

So you can use a 32 bit only processor (if you manage to find one) with PCI-X running 64 bit wide, no problem.

 

 

 

I'm aware of all that, it's just the way it was worded  "The motherboard has space to fit 64bit length cards so ill look into that once I have it posting reliably again" could have led to misinformation being spread so had to clarify that in terms of bit capability of the system itself.

In terms of bus width/capability based on 64 bit width for a card that's on point.

32bit CPU's shouldn't be too hard to find since all Socket A chips are 32 bit capable, even really old Duron Spitfire chips can do it for example.

[Kizoto]

Isn't using a 64bit pci card in a 32bit slot similar to using a 16x pci-e card in an 8x slot (assuming same generation)?

 

I have upgraded a couple of IDE based laptops in the past with ide->sata m.2 adapters and they made a huge difference. It does however mean the system ends up at 100% cpu usage whenever the drive is busy, but it beats waiting on an old ide drive by a fair margin.

 

Seeing as whatever I choose to use will be limited by the 32bit pci bandwidth, I am also wondering if its worth not bothering with any storage in the system itself and instead network booting a remote os (if i can figure out how), the 1gb network card should be as fast as any sata in this system, maybe with a little extra latency. It would be interesting to try all the options and benchmark them too.

[mariushm]

You can buy 4 GB IDE SSDs for $10 .. example : https://www.ebay.com/itm/193738886321

 

You can buy ab 8 GB one for $15 or so, example: https://www.ebay.com/itm/224911234267?

 

Or 16 GB for $17 : https://www.ebay.com/itm/293131754251

 

You can install Windows XP for sure in that space, Windows 7 mayl also work if you do some custom isos... there's various tools out there which shrink the iso installation.

 

I don't recommend network boot. Would be extremely difficult to actually run Windows from network, some linux or BSD may work.

 

(edit: they're 44 pin ide, of course you use a 40/80 pin + power  <-> 44 pin ide adapter)

[Guest]

The best stability is running regular old IDE drives with windows XP. There's no point having a faster HDD/SSD with a system that has 1000mhz Cpu to work with. It won't make it faster really. You'll be waiting for the processor to load the programs/apps anyways. So it's a moot point to buy a PCI sata card, the bandwidth just isn't there to utilize it.

 

Good Luck though!

[Kizoto]

I have got the board working again, replaced the memory/pci capacitors but the cpu/vrm caps all showed no degradation so didn't replace those (desoldered, measured them and they matched the 2200uF stated on the cap so just reattached them again). First time I have recapped a board, clearing the solder from the through holes so the replacement caps could be fitted was a nightmare, I ended up using a small hand drill and a needle to break through the solder in the holes

 

I found a 2GHz barton cpu that the system also posts with, it shows up as an unknown cpu though.

 

I might try ordering all the storage options and benchmark them all to get some real numbers on what differences there are. I still have some IDE IBM "deathstar" drives kicking around but they will probably fail if I tried to run them for long.

[Guest]

4 hours ago, Kizoto said:

I have got the board working again, replaced the memory/pci capacitors but the cpu/vrm caps all showed no degradation so didn't replace those (desoldered, measured them and they matched the 2200uF stated on the cap so just reattached them again). First time I have recapped a board, clearing the solder from the through holes so the replacement caps could be fitted was a nightmare, I ended up using a small hand drill and a needle to break through the solder in the holes

 

I found a 2GHz barton cpu that the system also posts with, it shows up as an unknown cpu though.

 

I might try ordering all the storage options and benchmark them all to get some real numbers on what differences there are. I still have some IDE IBM "deathstar" drives kicking around but they will probably fail if I tried to run them for long.

You have to use CPU-Z vintage version. I wanna say it's version 1.05, But @Beerzerkercan confirm the proper cpu-z on socket A.

[Beerzerker]

The latest version known to run correctly out of the usual releases is V1.59 but the referred to vintage version works fine and is readily available, you can get that from the CPU-Z site.
CPU-Z | Softwares | CPUID

There is the current vintage version 1.03 and you can get it by using the link I provided.
There is also an older versions tab but it only goes back to V1.63, not as far as V1.59 you'd need.

[mariushm]

12 hours ago, Kizoto said:

I have got the board working again, replaced the memory/pci capacitors but the cpu/vrm caps all showed no degradation so didn't replace those (desoldered, measured them and they matched the 2200uF stated on the cap so just reattached them again).

Just want to make this observation : it's not enough to measure capacitance with a multimeter. A regular multimeter can not measure ESR and other parameters of the capacitor and some capacitors can degrade over time without having their actual capacitance change or even having their capacitance increase.

Especially in the CPU VRM area, ESR is the most critical parameter - they are constrained by height due to CPU coolers so they often install 2 or even 3 capacitors in parallel in order to lower the final ESR, not to get more capacitance.

Polymer/ Solid capacitors have much lower ESR than electrolytic capacitors, so you can often replace the electrolytic capacitors in the CPU VRM area with lower capacitance ones (as long as the total capacitance is enough, around 3000-4500 uF) - so for example, you could replace 6-8 2200uF 6.3v electrolytic capacitors with 1200-1500uF 4v (even 2.5v) solid capacitors (lower voltage is fine, the cpu voltage is usually under 2v)

 

12 hours ago, Kizoto said:

 

First time I have recapped a board, clearing the solder from the through holes so the replacement caps could be fitted was a nightmare, I ended up using a small hand drill and a needle to break through the solder in the holes

 

Using drills is very risky, because you can break internal layers in the circuit board.

You need decent soldering iron or preferable soldering station because in the VRM area there's a lot of copper in the motherboard which absorbs heat.

 

The easiest strategy is to actually cut the capacitor at the base to leave a bit of the leads, then add a couple blobs of solder on the bottom and while the blob is heated by your iron tip, you push the leads from the front side to clear the hole.

 

12 hours ago, Kizoto said:

I found a 2GHz barton cpu that the system also posts with, it shows up as an unknown cpu though.

 

I might try ordering all the storage options and benchmark them all to get some real numbers on what differences there are. I still have some IDE IBM "deathstar" drives kicking around but they will probably fail if I tried to run them for long.

 

[Kizoto]

I have made some progress with this, my results so far:

 

IDE mechanical drive ~40MB/s r/w sustained, 0.01 r/w random.

IDE SSD (disk on module) ~40MB/s r/w sustained, ~2MB/s r/w random

IDE -> SD card ~40MB/s r/w sustained, ~5MB/s r/w random

 

running any of these tests though a usb2 pci card gives the same results but caps the speeds at 25MB/s.

 

Sata SSD (either m.2 or normal 2.5") using a pci card gives best results with ~80MB/s r/w, 15MB/s r/w random

 

stuff which didn't work well:

IDE -> CF -> SD card (less than 2MB/s for everything)

NVME is seen by the OS, but the crucual/micron drive I have spare for testing has no 32bit drivers, I might try some other makes to see if it helps.

 

At the moment I am running windows xp from the IDE->SD card adapter, I can benchmark stuff attached to the SATA PCI card from in windows, the raid bios for the card also shows what is attached during post, but I am unable to boot from it.

 

Is there any advise on how I could get around this? I tried running plop boot manager from an sd card, but it just locks up when I choose to boot from the sata drive (it works for usb booting though), are there other boot managers I could try?

[jaslion]

19 hours ago, Kizoto said:

I have made some progress with this, my results so far:

 

IDE mechanical drive ~40MB/s r/w sustained, 0.01 r/w random.

IDE SSD (disk on module) ~40MB/s r/w sustained, ~2MB/s r/w random

IDE -> SD card ~40MB/s r/w sustained, ~5MB/s r/w random

 

running any of these tests though a usb2 pci card gives the same results but caps the speeds at 25MB/s.

 

Sata SSD (either m.2 or normal 2.5") using a pci card gives best results with ~80MB/s r/w, 15MB/s r/w random

 

stuff which didn't work well:

IDE -> CF -> SD card (less than 2MB/s for everything)

NVME is seen by the OS, but the crucual/micron drive I have spare for testing has no 32bit drivers, I might try some other makes to see if it helps.

 

At the moment I am running windows xp from the IDE->SD card adapter, I can benchmark stuff attached to the SATA PCI card from in windows, the raid bios for the card also shows what is attached during post, but I am unable to boot from it.

 

Is there any advise on how I could get around this? I tried running plop boot manager from an sd card, but it just locks up when I choose to boot from the sata drive (it works for usb booting though), are there other boot managers I could try?

Booting from a pcie sata card is one of those things that is often not possible unless its an active raid controller card. Even then its very hit or miss on these old platforms.

 

Speed is a lot better than expected I thought that thunderbird would reach 100% usage before it would much further than half the pci bus speed for the ssd.