The issue looks like a typical memory issue. Since those traces that are cut are memory related, they could explain your artefacts. You can try to fix them, regarding the second layer I'm not sure, it might be GND anyways.

I'd give it a shot.

alright, i was thinking a moment ago, an approach which would be easier on my back would be to try and flash the card's BIOS to 64mb instead of 256mb. hopefully in that case, the cut memory traces would simply be inactive. 64mb for this card in even a fast P4 computer will likely be way faster than what's needed.

i removed the solder mask and used a multimeter to confirm there are 5 cut traces. the 6th looks bad but still works. the plan is to put 5 pieces of copper wire down, solder them in place and if it works, cover it with hot glue. some of the traces have vias which i'll use as the solder points.

If you disable memory, you'll also cut the bandwidth in half, meaning you'll have a 64MB/64Bit Radeon X700. What a weird thing. I think It'd be performing around Radeon 9600 level then.
However I'm not sure if you can "control" which of the memory modules are disabled, by just setting 64MB as the memory target.

As you mentioned, the testpads around those traces are already very inviting as a mountpoint for one side of the repair wires.

I think there is a very good chance to get this going again.

tehsiggi wrote on 2025-07-23, 05:30:

If you disable memory, you'll also cut the bandwidth in half, meaning you'll have a 64MB/64Bit Radeon X700. What a weird thing. […]
Show full quote

If you disable memory, you'll also cut the bandwidth in half, meaning you'll have a 64MB/64Bit Radeon X700. What a weird thing. I think It'd be performing around Radeon 9600 level then.
However I'm not sure if you can "control" which of the memory modules are disabled, by just setting 64MB as the memory target.

As you mentioned, the testpads around those traces are already very inviting as a mountpoint for one side of the repair wires.

I think there is a very good chance to get this going again.

Probably less than a 9600... even those are usually 128MB or 256.

(I know the Mac version exists, but I'm specifically talking about retail cards.)

PcBytes wrote on 2025-07-23, 10:47:

tehsiggi wrote on 2025-07-23, 05:30:

If you disable memory, you'll also cut the bandwidth in half, meaning you'll have a 64MB/64Bit Radeon X700. What a weird thing. […]
Show full quote

If you disable memory, you'll also cut the bandwidth in half, meaning you'll have a 64MB/64Bit Radeon X700. What a weird thing. I think It'd be performing around Radeon 9600 level then.
However I'm not sure if you can "control" which of the memory modules are disabled, by just setting 64MB as the memory target.

As you mentioned, the testpads around those traces are already very inviting as a mountpoint for one side of the repair wires.

I think there is a very good chance to get this going again.

Probably less than a 9600... even those are usually 128MB or 256.

(I know the Mac version exists, but I'm specifically talking about retail cards.)

Yeah it's the question how the bandwidth starving will influence the performance. The GPU itself is basically double a 9600 in every aspect:

9600:

• Pixel Shaders 4
• Vertex Shaders 2
• TMUs 4
• ROPs 4

X700:

• Pixel Shaders 8
• Vertex Shaders 6
• TMUs 8
• ROPs 8

A curious experiment.. If I only had time this week 😁

Hi, any update on this topic?
Currently tackling an x740 (medion take on the x700) with artifacts too, although mines in way worse shape.... Underscores moving around the screen and other funky stuff.

These samsung memory chips are known to be unreliable. 3/3 cards I owned either had artifacts or were unstable after stress tests (artifacts, PC lockups with artifacts). I do not know why but pretty much all of these cards which have either Samsung or elixir DDR memory in TSOP or BGA are artifacting. Hynix ones seem to last longer.

Medion has GDDR3 memory and those Samsung chips are perfectly fine. Samsung had major quality issues with GDDR and GDDR2 memory.

Just don't bake the card, it usually destroys caps which do not like high temperatures.

The Serpent Rider wrote on 2025-07-26, 21:58:

Medion has GDDR3 memory and those Samsung chips are perfectly fine. Samsung had major quality issues with GDDR and GDDR2 memory.

Huh, so the GDDR3 is guaranteed to be artifact free?
Don't wanna hijack this thread more so I will create a new thread.

kotel wrote on 2025-07-27, 08:58:

Huh, so the GDDR3 is guaranteed to be artifact free?

Not in 20 years.
I think the strategy of choice is simple.
"Not tested", but the price is two times lower than the market average - we try our luck.
If we don't get lucky - we repeat.
Or we look for lots with Furmark screenshots.

PS. I don't know about other local flea markets.
But this is a great answer: "card not used for mining" for FX5700Le...

Testing GPUs using furmark is unnecessary. I usually do 1-2h of Sanctuary benchmark (unigine) followed by 1-2h video memory test along with various 3d mark versions. If it looks good I clean it with wet cloth and repaste it.

Even if the GPU is advertized as tested don't think that the seller tested it thoroughly. Usually it means just that it boots and perhaps video driver doesn't crash on initialization.

If you don't want to take the risk you need to ask for 1-2 days warranty.

Honestly i much prefer to buy cards which were not cleaned/repasted and with no furmark screenshots. If 3d works at all, like 3dmark or any game runs, it is sufficient and there is adsolutely no reason to torture old hardware with furmark or even long runs of any benchmarks.

Thing is - there is always a degree of trust here. There are well known and unavoidable ways to scam. And 1-2 day warranty will not help. Like heat up the GPU to make it work with bump failure (by far the most common and pretty much the only completely irreparable fault) and sell it, it'll survive some benchmarks and will last for a while. Sometimes days, sometimes months.

And then there is random luck... I had a funny case once - i've given away a system to a friend for free back during 2020 mess "for kids to play some games", it had GTX760 which i bought new and never had issues with it. That card died in less than a week. And yeah, bump failure, "fixed" it a few times for him with some heat until he bought a replacement. Had i sold it the buyer would have considered me a scammer...

Archer57 wrote on 2025-07-29, 00:21:

it is sufficient and there is adsolutely no reason to torture old hardware with furmark or even long runs of any benchmarks.

If you use your graphics card specifically for gaming, and not for looking at your old time desktop, then stability is very important.
Many old RTS games don't have autosave.
And replaying a level for an hour and a half isn't very exciting.

I used Furmark as an example, the sellers simply don't know how to do anything else.
Personally, I prefer the cyclic test "Nature" + CPU test # 1 in 3DMark 2003.
For several hours.

shevalier wrote on 2025-07-29, 05:20:

Personally, I prefer the cyclic test "Nature" + CPU test # 1 in 3DMark 2003.
For several hours.

May I ask why do you also do CPU test #1 in 3dmark03? Is it just to test the CPU and GPU at the same time?

Old hardware can die unexpectedly which is why you should prefer to buy it cheap, without looking for a specific brand. I have not had a bumpgate era GPU die on me but I have a GeForce RTX 2080Ti OC that died after a few months of very light use. It was benchmarked and furmarked at the beginning and looked fine. Furmark wasn't able to fully utilize it, the driver probably has a protection for it. I'm not buying expensive used cards anymore, it is a waste of money.

kotel wrote on 2025-07-29, 06:59:

shevalier wrote on 2025-07-29, 05:20:

Personally, I prefer the cyclic test "Nature" + CPU test # 1 in 3DMark 2003.
For several hours.

May I ask why do you also do CPU test #1 in 3dmark03? Is it just to test the CPU and GPU at the same time?

Changing the context in 3DMark tests allows you to identify instability of the video card interface (AGP or PCI-e) due to cycles of intensive data transfer and their absence.
And cyclic heating/cooling.
The cyclic test "Nature" is simply a strong load. If the card is unstable, then sooner or later the test will fail.
But, in my opinion, alternately running "Nature" + "CPU test #1" leads to this much faster.
Let's say 20 minutes instead of 2 hours.
But this is my personal experience.

"Planes" - general instability of video memory, leading to flickering textures.
"Trolls´ Lair" - you need to pay attention to shadows when the light source moves. Extra light pixels begin to appear when several memory cells fail.
"Proxycon" - in my experience, shows nothing at all.

AlexZ wrote on 2025-07-29, 07:06:

Old hardware can die unexpectedly which is why you should prefer to buy it cheap, without looking for a specific brand. I have not had a bumpgate era GPU die on me but I have a GeForce RTX 2080Ti OC that died after a few months of very light use. It was benchmarked and furmarked at the beginning and looked fine. Furmark wasn't able to fully utilize it, the driver probably has a protection for it. I'm not buying expensive used cards anymore, it is a waste of money.

That's why furmark is bad it can outright fry cards on which power limit was not yet implemented because it exceeds design power, or it does nothing on cards which have power limit because it simply bumps into that and throttles.

And yeah, there is no way to be sure with used hardware. It could have just died, or you might have been scammed with GPU which already had bump failure and was heated up to make it work briefly.

Archer57 wrote on 2025-07-29, 11:11:

or it does nothing on cards which have power limit because it simply bumps into that and throttles.

I don't know how modern NVIDIA graphics cards work.
RX(and newer) Radeon graphics cards use a complex temperature control algorithm.
The graphics card under load will try to maintain a "target temp" until the "target RPM" exceeds a set value.
If the "target temperature" = 70 degrees, then the video card will heat up to exactly 70 degrees under almost any noticeable load. Naturally, with very different fan speeds.

After this point, the RPM rise to 100% almost instantly, and most likely the fans will not stop once the temperature drops to the "zero fan stop temperature"- the probability of a malfunction of the video card cooling system is almost 100%.
Poor quality thermal grease, too thick thermal pads, crooked installation of the cooler, etc.

Therefore, the modern video card needs to be warmed up with something for testing, and Furmark is suitable for this.

I don't test AGP or early PCI-e with Furmark because I don't see the point.
The "interface speed" test in GPU-Z in fullscreen mode is quite sufficient.

Do not use Furmark for pre-heating. It heats up the GPU too fast. I set up fan profile to heat up to about 70'C in Sanctuary benchmark which results in 62-65'C in games. My GeForce GTX 275 ran at 85'C for about 30 minutes before I noticed temperature problem, fortunately no damage done. Target temperature set in BIOS is too high. After repasting, cleaning it and configuring fan profile manually it runs fine.