Why does VRAM die so often?...

[BladeVenom]

I've only had one go bad. I've mostly used ATI cards, and video cards have been one of the most reliable parts of a computer for me. I also usually don't overclock my video cards.

 

[Turbonium]

Originally posted by: BladeVenom
I've only had one go bad. I've mostly used ATI cards, and video cards have been one of the most reliable parts of a computer for me. I also usually don't overclock my video cards.

I don't overclock either - I run everything within spec. I've owned 2 video cards prior to this, and neither went bad (Radeon 7000, Radeon 9600 Pro), though neither ran nearly as hot as this 6800 GT.

 

[error8]

My 4870 vram is approaching death. I noticed that the memory doesn't sustain even 10 extra mhz over stock without artifacting. Now artifacts are all over the place in OCCT gpu test and they do disappear when I underclock Vram. I'm lucky that I don't have issues in games, but I noticed something weird one day in Batman. Oh, well, when it dies it dies.

In my opinion, like other have said, vram works at higher levels of stress then normal ram and it also heats up "better", thus greatly reducing it's lifespan.

 

[aka1nas]

Originally posted by: jiffylube1024
VRAM is pushed to the limits; system memory much less so. Just looks at the speeds they are released at - when Video memory was going past 1 GHz, system memory was between 200 and 400 MHz.
-----

And as for the 6800GT being released before the thermals were under control - it's been that way pretty much since Transform & Lighting was introduced, and particularly since the FX 5xxx series of video cards. It became an arms race between Nvidia and ATI for the performance crown, and the definition of "acceptable" thermals and lifespan for most modern graphics cards would make an Intel or AMD (CPU division) engineer cringe.

GPU makers have the added constraint that, unlike CPU's, their GPU's face upside down (in most tower cases), with much inferior cooling to CPU's, and not the most ideal location to dump heat (below the GPU or outside a tiny hole in the back of the graphics card).

Completely agree. Up until fairly recently, it was also unlikely that you would place a full load on our GPU for extended periods of time. Even while gaming, your GPU doesn't always heat up that much unless the scene is very complex. This is of course changing due to things like GPGPU and PhysX.

As GPGPU becomes more common, the vendors will have to validate their cooling solutions more carefully. They could probably get away previously with having it pass the equivalent of a 3dmark run and stamping a passed sticker on the card.

 

[Meghan54]

Originally posted by: Zap

Originally posted by: Turbonium
On the other hand, I've virtually never heard of system RAM (run at stock) going bad.

You've obviously never owned any Crucial Ballistix DDR2 from the past few years. (and bryanW1995 beat me to it)

I've also had some Corsair and A-Data RAM DDR RAM die on me.

Note that this doesn't mean I've stopped buying those brands.

Quite true.........Crucial Ballistix is the brand I've had the most failures with, but I've also had a set of Mushkin Redline fail, also some Buffalo Firestix, several sets of Kingston HyperX, as well as a few sticks of "generic" RAM by Corsair, Crucial, Kingston, G. Skill, and no-name.

On the other hand, the two most recent video card failures I've suffered both were from VRM overheats, not RAM failures. I have had other video cards fail, but most were things like capacitor failures and the like....don't know or remember if any ever had RAM failures on them.

 

[Turbonium]

Originally posted by: Meghan54
Quite true.........Crucial Ballistix is the brand I've had the most failures with, but I've also had a set of Mushkin Redline fail, also some Buffalo Firestix, several sets of Kingston HyperX, as well as a few sticks of "generic" RAM by Corsair, Crucial, Kingston, G. Skill, and no-name.

Overclocked, or within spec?

 

[reallyscrued]

Originally posted by: jiffylube1024
VRAM is pushed to the limits; system memory much less so. Just looks at the speeds they are released at - when Video memory was going past 1 GHz, system memory was between 200 and 400 MHz.

That's got nothing to do with it. VRAM is rated to run at a speed and that's the speed manufacturers build cards to run it at. What's all this pushing to the limits none-sense?

VRAm and run-of-the-mill system memory have very little in common to anyone who's wondering. You can't work out a clock for clock comparison.

 

[fffblackmage]

Originally posted by: Zap

Originally posted by: Turbonium
On the other hand, I've virtually never heard of system RAM (run at stock) going bad.

You've obviously never owned any Crucial Ballistix DDR2 from the past few years. (and bryanW1995 beat me to it)

Heh, I still have my Crucial Ballistix I bought more than two years ago and they still work fine. They ran on stock or underclocked settings 99.99999% of the time. Funny how people buy that particular ram for OCing and I underclocked them instead....

 

[jiffylube1024]

Originally posted by: reallyscrued

Originally posted by: jiffylube1024
VRAM is pushed to the limits; system memory much less so. Just looks at the speeds they are released at - when Video memory was going past 1 GHz, system memory was between 200 and 400 MHz.

That's got nothing to do with it. VRAM is rated to run at a speed and that's the speed manufacturers build cards to run it at. What's all this pushing to the limits none-sense?

VRAm and run-of-the-mill system memory have very little in common to anyone who's wondering. You can't work out a clock for clock comparison.

I would disagree with your statement that just because VRAM and system memory run at manufacturer rated speeds that that shouldn't affect product life. Have you seen the difference in power delivery systems from the motherboard to the DRAM vs. on video cards for VRAM? On video cards, due to the high clockspeeds, they're necessarily more complex.

System memory is built to be above all stable because the side effects can be disastrous: system crashes and data corruption. Video memory on high-end GPU's is by its very nature bleeding edge because GPU's crave memory bandwidth. Video card manufacturers usually underclock their memory a bit compared to the rated speed to deal with hot/cramped computer cases, but even then the heat builds up and usually leads to a much faster death than in system memory. There is a cost to having VRAM run at 1000+ MHz and that cost is that it's pushed closer to the limit than system memory.

The other bonus for video card manufacturers is that memory glitches in GPU memory can cause graphical artifacts or, at worst, a game crash or system crash. They're not going to cause system data corruption, so they can be a bit more aggressive with their timings.

----------

Also, what's with saying that VRAM and system RAM have very little in common? Compared to pretty much every other component in a computer, they have a whole lot in common. Current system DDR3 is based on the design of GDDR3 from a few years ago AFAIK. One of the only major differences aside from VRAM being faster/newer is that it's soldered onto the video card's PCB while system RAM is on a DIMM PCB.

 

[zephyrprime]

Originally posted by: Zap
You've obviously never owned any Crucial Ballistix DDR2 from the past few years. (and bryanW1995 beat me to it)

I've also had some Corsair and A-Data RAM DDR RAM die on me.

Note that this doesn't mean I've stopped buying those brands.

I had some of that. It did indeed die after only a few months but that was due to some sort of manufacturing defect rather than stress. (not oc'ed by the way).

 

[evolucion8]

I Used to have a 2GB kit of Ballistix Tracer DDR1 500Mhz and both sticks died on me running at 400MHz, one died 3 months later and got replaced and still working fine today, the other one died 6 months later and they didn't had any more of those on stock and wanted to give me cheap and slow DDR 400 with 3-4-4-7 RAM stick as a replacement and wanted me to pay the price difference plus shipping, I said thanks but no thanks, that's why I only use G.Skill

 

[jiffylube1024]

Originally posted by: evolucion8
I Used to have a 2GB kit of Ballistix Tracer DDR1 500Mhz and both sticks died on me running at 400MHz, one died 3 months later and got replaced and still working fine today, the other one died 6 months later and they didn't had any more of those on stock and wanted to give me cheap and slow DDR 400 with 3-4-4-7 RAM stick as a replacement and wanted me to pay the price difference plus shipping, I said thanks but no thanks, that's why I only use G.Skill

Crucial had issues with their Ballistix DDR1 memory - it ran very hot and if I remember correctly required a lot of voltage to run at its rated speed, which helped kill the RAM extra quickly.

 

[Turbonium]

Ignoring outliers like Crucial Ballistix memory, would it then be safe to say that VRAM does indeed die more often, on average, than system RAM (at least with more modern, faster video cards)? I don't think it's unreasonable to assume this, based on some of the reasoning posted above (i.e. VRAM being pushed closer to its limits, and all that it involves in terms of system architecture requirements, heat generation, etc.).

 

[Meghan54]

Originally posted by: Turbonium

Originally posted by: Meghan54
Quite true.........Crucial Ballistix is the brand I've had the most failures with, but I've also had a set of Mushkin Redline fail, also some Buffalo Firestix, several sets of Kingston HyperX, as well as a few sticks of "generic" RAM by Corsair, Crucial, Kingston, G. Skill, and no-name.

Overclocked, or within spec?

All run at stock speeds....never OC my RAM. CPU's, well that's another question, but RAM, video cards....all run at stock speeds.

But then I'm including probably a decade of time here if not more.