I'm wondering if something is wrong with my setup or expansion card. With my onboard USB ports I have about 30 MByte/sec, but over the PCI card it's only 20MB/s. Internal drives (connected to PATA and SATA1, it's an older machine) run fine at about 95 MB/sec.

If anyone has an external HDD connected to USB 2.0 on a PCI expansion card (not PCIe!), please post the transfer speed you get (roughly, no exact benchmark necessary) with that setup. A way to quickly measure is (as root, run two times because some drives are slow the first time due to APM):
Results for an external drive connected to USB 3.0 or eSATA on a PCI card (again, not PCIe!) would be interesting as well.

Thanks in advance.

With usb, you're on the best at around 30mb/S. Windows boxes struggle in the main to get past 12mb/s as someone who has copied large amounts that way on usb-2. A usb-2 pen drive here does 20mb/s

20mb/s is poor for an internal hd. Your chipset driver choice is poor. When I see this, I recommend compiling in the chipset driver for your specific box and not including pci generic support, because that driver is tried first if it exists. \It will usually spit a message like

<Something> not 100% native, will probe irqs later . . . .

and the proper driver won't load.

True. At least theoretically, there were some chipsets in the past that had problems to reach that speed.

Plain wrong. Windows boxes are as fast as Linux boxes with copying from USB. If you get low speed when copying from or to Flash drives this is usually due to cheap and crappy Flash drives, not because you use Windows. If you aren't limited by the drive (usually when you use an external harddisk or high quality Flash drives) you get the same speed on both OSes.

The OP clearly states that this is an external disk and that the internal disks work at the expected transfer rate, so the rest of your post is not relevant to this question.

@instag: The slower transfer speed can be caused by several factors, for example a crappy controller on the expansion card or other PCI devices on the same bus (either expansion cards or devices built into the mainboard, which is quite common with older machines).

That's right, I'm wondering about the HDD throughput connected over USB to PCI.

I tested with two different PCI cards (one with a NEC, the other a VIA chipset), nothing else in the PCI slots. Admittedly, those were very cheap (around 6 EUR each). lspci says "NEC Corporation USB 2.0 (rev 04)" and "VIA Technologies, Inc. USB 2.0 (rev 51)".
Concerning onboard devices (VGA, serial, USB etc.), I deactivated them in the BIOS, still the same speed. Could they interfere even when not in use?

I'm also suspecting that the controllers on the PCI cards are crappy, that's why I'm trying to find out if someone has a faster card. Damn vendors only state the max. theoretical bus speeds for USB and SATA, completely useless.

A little known cause of issues on pci is voltage.
One 5V card limits the bus to 33Mhz; All 3.3V cards allows higher(66Mhz) but I think lower voltage again would probably be required to go beyond that.

That said, my usb-2.0 disk does 29MB/s in a usb3 port, and 28MB/S in a usb-2.0 port. It does seem to use xhci-hcd, and not ehci-hcd. USB-1.0 modules are rarely loaded.

Did you try to install the card to another PCI slot?

I once installed a 4-port USB2 card (NEC chip) on PCI slot on an old motherboard (Asus P4B), because the on board ports were usb1.1. So, I can't compare on board and expansion ports. I can check transfer speed on this card.

You are absolutely right. I never had a real speed more that the 30% of this theoretical USB-2 speed. I asked for this to computer shops, I got no answer.
I finally (some years ago) found something in the web page of the usb consortium: they recommended to the device developpers to limit the speed of their devices to the 30% or so, in order to leave bandwith for other devices on the USB bus.
Of course, there are devices such sound cards that can't be limited to such values, otherwise they couldn't work correctly.

I recently looked for some usb stick with a write speed higher that the 7-8 MB/s I usually experienced. I bought a Kingston DT R3.0 (usb3), in rubber enclosure, that advertises write speed up to 25GB/s on usb2 and 30GB/s on usb3 (according "internal tests"), advertised read speeds are around the double of these. I get real write speed lower than 20 GB/s, read speed 20-25GB/s on usb2 port.

I use the Corsair Voyager GT and get the advertised speeds. The speed you get is not only dependent on the flash drive, but also the controller. AMD's SB700 southbridge was famous for getting crappy speeds on USB.

By the way, you won't get GB/s in any case, I think you mean MB/s.

Thanks. Yes, of course. Advertised speeds are in MB/s, my speeds are in MiB/s.
I'll check it on other computers, too.

USB cards without an internal power connection are all 5V because this is what the USB specs require on one of the wires. But the PCI bus on my old motherboard is 33MHz anyway. With a theoretical speed of that PCI bus of around 130 MB/s I still should reach the speed of onboard USB 2.0 at around 30 MB/s, for a single connected HDD. But I don't, and anyone else it seems.

I guess these are values for HDDs connected to onboard USB ports?

Yes, but no difference.

I'd say they were referring to USB 1.x devices.

I think those hdparm speeds are a bit theoretical.
I quoted 29.x MB/S for my usb port setup to hard disk

time (8.3G copy) gave me 15m05s real time. Even if the dual core cpu doubled the time, (i.e. 7m32s) it still doesn't compute to the ~1.75G/min that you would expect at 29MB/S.

I have one of those amd chipsets here (sb600) and never got 'small willy syndrome' over it's usb speeds :-). There are other bottlenecks.