Fun with a graphics card

[Chromatix]

Normally I'm pretty good at sorting out my computers, but this one has me very puzzled.

I got myself a Asus RadeonHD 5750 card, and tried putting it in my main games PC to replace an 8800GT - and it didn't work. The machine didn't even beep at me, just started flashing the HD light. (It does produce a beep pattern if I give it no graphics at all.) It hasn't got far enough for driver conflicts to be a possibility.

Now, I know the 5750 isn't much faster than an 8800GT in most of the benchmarks, but I think it'll be worthwhile for some specific things that I run. I can use the 8800GT elsewhere.

The bizarre thing is that we have an AMD CPU, an ATI Xpress 3200 chipset, a m/board made by Asus, a card made by Asus, and a GPU made by ATI/AMD/whatever - and they refuse to work together. The PSU works (I measured it, including the PCIe dongle), and the power consumption of the 5750 should be *less* than the 8800GT anyway.

I put the 5750 in a different machine - with a slightly older m/board with an AMD CPU, an ATI Xpress 200 chipset, a m/board made by Asus, and a smaller PSU (which I had to use the PCIe power adapter with, since it didn't have it's own). It works. But that machine runs Linux, which isn't what I got the card for.

A much smaller RadeonHD 4350 - not from Asus this time - works in the same slot on the Windows machine, and there are no driver conflicts. This is the card I got for the Linux machine I used for testing.

There is a second PCIe slot, but none of the three cards will fit in it, because there's a capacitor in precisely the wrong place on the m/board, which hits the DVI socket on each card and prevents them from seating. I think I have only one card that will fit in the second slot, an "old" X1950 Pro which I already know doesn't have the performance I'm looking for.

I already took one of the 5750s back to the shop, because I couldn't believe that this kind of conflict - between components made and assembled by the same companies - existed in this century. I dug deeper when the replacement acted precisely the same way.

What I might have to do is leave the 5750 in the older m/board, and change the CPUs and HDs over so that it runs Windows while the newer board runs Linux. This would have *some* advantages, mostly since the newer board is also in a huge case which I could use to make a fileserver, but I'd prefer to have the choice.

Any ideas from the peanut gallery?

[Midorikawa]

I know you checked with a PSU dongle, but it still sounds like insufficient power. Try swapping PSUs out with the one that does work and see what happens.

[Broomjockey]

Yeah, my first thought is insufficient power being delivered. If it won't even go to POST, and it's not the card itself, there's not a lot of other options.

Odd though. I know when I once put a card in and nothing was getting enough power, when I tried to turn it on, my computer emitted the howl of the damned.

[Midorikawa]

Odd though. I know when I once put a card in and nothing was getting enough power, when I tried to turn it on, my computer emitted the howl of the damned.

I must say, I've never seen that result. I have, however, seen machines act fine, but never output video of any kind, or fail with all manner of weird BIOS codes, RAM, CPU, etc. I've learned to always just swap the PSU first, rather than test, since PSUs can fail in weird ways that only one certain piece of hardware can bring out.

[Broomjockey]

I must say, I've never seen that result.

You know that high-pitched beep that cheap smoke detectors put out? Imagine that as a solid tone and then up it an octave. That is the noise it made. Stopped when I put the old card back in.

[Chromatix]

Well let's see. The PSU in the machine that works is 350W, and is technically too small, which is probably why it doesn't come with PCIe cables.

The PSU in the one that *doesn't* work is 500W, and comes with a PCIe cable built in, plus a modular cable for a second one. I've tried it with both of those. Each of the 12V rails is rated for 18A.

I've just measured the voltage on the live, powered-up connector under load, and it's about 14V. That seems a bit high, but it would get regulated down on board the card.

If I leave the card in the socket but remove the PCIe power from it, the machine behaves as though it doesn't exist. I can then boot into Windows using the X1950 in the second slot (which also requires PCIe power). Yes, I've tried it with both cards powered too, it has the same problem.

I suppose I can try stuffing the enormous and noisy kilowatt-rated Enermax Galaxy back into this case...

ETA: I don't want to run a dual-core CPU *and* a card that requires supplemental power *and* two hard drives from a 350W PSU. So that's not an option.

[Chromatix]

Well, I just tried it with the Galaxy - no dice. (Wired it up without trying to mount it in the case.) Voltage at the PCIe connector is a much more reasonable 12.1V this time.

Pulling the PCIe power from the 5750 lets me boot using the X1950 again.

Throughout all this, the fan on the graphic card works fine. The heatsink on it isn't getting hot (which it does when it works in the Linux machine, since Linux doesn't know how to use power-management on it).

[Chromatix]

Looking on the Asus support forums gives a clue... the card has 1GB VRAM, whereas all the others are 512MB or less. Apparently there is a problem with the memory mapping on the A8R series boards, and they tend not to work with 1GB VRAM cards or even with 4GB RAM.

[Krivak]

That sucks. I can say though, that the card you mentioned and the PSU should be good. I've got a 550W PSU with a 5770, on a Dual core 2.81 AMD processor, and a gigabyte ultra durable 3. No problems, except the card sometimes like to wriggle loose every once in awhile. I'd suggest gigabyte, as i use most of their stuff in my comp with no problems. Asus seems..untrustworthy to me at this point.

[Midorikawa]

Asus seems..untrustworthy to me at this point.

Has ASUS' gear gone down in quality recently? I have an eeePC 1000HE, but beyond that, I've been on the Mac side of things for years, and we use Intel * at work, so I haven't worked with AMD/ATI/ASUS in years. I do have an ASUS mobo in my old Barton core Athlon XP 2800 machine, and it's been stellar. Barring power failures that exhaust the battery backup, Gentoo will run nigh on indefinitely on it. Had 325 days of uptime on it before a 4 hour power outage once, and it was still running like the first day I booted it and started the IRCd.

[Chromatix]

Asus still have a generally good reputation, but I'm wondering whether I've come across one of their Friday jobs - not the graphics card, but the m/board design.

I got a reply back from their tech support, recommending I zap the CMOS. By "removing the battery for 10 mintes with the power unplugged". There's a perfectly good CLRTC jumper, and I told them so. (No, it didn't help.)

[Chromatix]

Asus' best response now is "your graphics card or m/board may be faulty, contact your dealer." I already swapped the graphics card once, and the m/board is sufficiently old as to be out of warranty (but not old enough to be unreliable).

However, I have discovered that the program I was having the most performance trouble with responds very well to a larger CPU cache. I thought the 2x 1MB caches on my Opteron were big enough, but apparently not - and I can't put anything bigger in that board.

So, I have now pre-ordered a lovely Gigabyte m/board, a Phenom II X2 555 Black Edition, and 4GB of fast DDR3 to switch the Windows machine to. The Phenom won't come into stock over here for about a week yet, but everything else will be held for me until then. I can re-use all the other components of the system, including the CPU heatsink, so I get a damn good upgrade for less than €350 (not counting the graphics card).

Bonus: according to Anandtech, there's a significant chance that I might be able to unlock the other two cores hidden on the 555's die!

I can also still reuse the Opteron by migrating one of my Linux machines to it - those are presently running on single-core CPUs. Avoiding the single application that was demanding a huge cache, the Opteron is still a pretty darn good CPU, and I have other graphics cards that will run well in it as well. Bonus: I have a lot of big HDs that I was trying to figure out how to use, and the Opteron is already in the one case that will hold them all, with the best PSU for the job.

Finally, the single-core Athlon64s will still find a use as experiment targets and spares. I can do a surprising amount of useful things with a LiveCD or a ninja install, and I have... projects.

[crashhelmet]

I had a similar problem with an AGP ATI HD3850. Turned out my PSU wasn't supplying the right Amperage. It wanted 30A on the 12V rail. Until I got a PSU that could do that, it acted as if the card didn't exist.

CH

[Chromatix]

Well the card works just fine in the new m/board, with only a 400W PSU. I think I should get a bigger PSU before I try unlocking the extra cores on the CPU though.