980GTX MSI vs Reference

[epixoip]

So If I have a single GPU/double GPU PC, is it big problem to have a OEM card? Cause It usualy doesn't suffer from over-heating. And secondary, would there be a Hash per second drop?

Go back and re-read what I wrote. Overheating is just one of the issues you might face with an OEM design card. The much bigger issue is components failing prematurely, causing the card to die. 99.9% of the time when you hear of someone's GPU dying, it's an OEM design card. Reference design cards are damn-near bulletproof.

For most algorithms you likely will not see a drop in hashrate since the GPU chip itself is the same. However, for certain algorithms (such as those that are memory-bound) it is entirely likely you will see a drop in hashrate due to cheaper RAM, cheaper memory controller, etc.

[blackstarek]

So by what you just said, this should be reference card.

http://www.czc.cz/evga-geforce-gtx-980-4...43/produkt

[epixoip]

That is indeed a reference design card!

[dingrite]

I never really understood the entire reference thing. You say that they cost more than OEM cards yet I see the same price or even cheaper wherever I look.


Theoretically speaking and I don't know anything about the topic, if reference cards are mandated and regulated by the GPU manufacturer they should be ideal, but the prices I see don't support that assertion.

What happens if OEM's decide to cut down costs by meddling with reference component quality, can they face any penalties?

[epixoip]

I never said reference designs cost more -- I said they cost more to manufacture. And the chipset manufacturer dictates the price, therefore the profit margins are very slim. This is why several OEMs (such as Gigabyte and MSI) only ever run one batch of reference design cards.

The fact that OEM cards often are just as expensive, or even more expensive, than reference design cards tells you what kind of markup OEMs apply to their designs. They are a fraction of the cost to produce, yet they command just as much (if not more) money than reference design cards due to their flashy (and largely superficial) fan and shroud designs and fancy marketing names. Because of this high markup, we can get some seriously absurd discounts on OEM design cards. Yet we almost always have to pay nearly full price for reference design cards, even when ordering straight from the manufacturer (usually a $20-25 discount is all we can squeeze out of them.)

If an OEM deviates from the reference design, then it's not a reference design card. It's an OEM design card. They would not be allowed to use the reference design shroud & branding on the card, they'd have to design their own shroud. Otherwise yes, it would be a breach of contract.

[dingrite]

I never said reference designs cost more -- I said they cost more to manufacture. And the chipset manufacturer dictates the price, therefore the profit margins are very slim. This is why several OEMs (such as Gigabyte and MSI) only ever run one batch of reference design cards.

The fact that OEM cards often are just as expensive, or even more expensive, than reference design cards tells you what kind of markup OEMs apply to their designs. They are a fraction of the cost to produce, yet they command just as much (if not more) money than reference design cards due to their flashy (and largely superficial) fan and shroud designs and fancy marketing names. Because of this high markup, we can get some seriously absurd discounts on OEM design cards. Yet we almost always have to pay nearly full price for reference design cards, even when ordering straight from the manufacturer (usually a $20-25 discount is all we can squeeze out of them.)

If an OEM deviates from the reference design, then it's not a reference design card. It's an OEM design card. They would not be allowed to use the reference design shroud & branding on the card, they'd have to design their own shroud. Otherwise yes, it would be a breach of contract.

I see, that is interesting.
What about the superclocked cards? They are usually reference design, but what was done to their clocks? Seems like w/e was done would require messing with a component somewhere.

[epixoip]

There's nothing special about superclocked or factory overclocked cards. They just have a bios that comes set at higher clocks, which you can easily do yourself.

[dingrite]

There's nothing special about superclocked or factory overclocked cards. They just have a bios that comes set at higher clocks, which you can easily do yourself.

Do you have any evidence that reference designs are actually superior in build/component quality?
I have been trying to find a source for it and failed. Or are you just speaking from experience? If so, what sample size did you deal with?

[epixoip]

(You probably will not find a source for most of the things I say. If you do manage to find a source, you will likely see me quoted as that source's source.)

My source is first-hand experience and specific knowledge. If you know who I am and what I do, then it should be no surprise that we have direct relationships with several OEMs, and we have purchased literally thousands of GPUs over the past five years. So we have quite a bit of experience with this stuff, more so than anyone else.

I can count the number of reference cards that we have had fail on one hand. Either they're DOA / fail burn-in testing, or they last forever. OEM design cards are the exact opposite. I can count the number of OEM design cards that HAVEN'T failed on one hand. But this really isn't that big of a surprise to us, because that's literally the point of an OEM design card: "how can we cut costs on this design and maximize profits."

One of the other things we commonly see are OEMs shipping new GPUs with old PCB designs (most of the GTX 900 OEM designs are based on previous GTX 600 and 700 designs), but we also saw last year that some OEMs were shipping new GPUs with previous-gen firmware too. When I wrote od6config last year, I started getting lots of emails from people who bought a 290X, but od6config would tell them that it wasn't an Overdrive6 device. That's because several OEMs were shipping cards with 290X chips, but with old Overdrive5 firmware. This was done intentionally because they didn't want to add the extra circuitry required by Overdrive6. Yet more cost savings.

Anyway, I have dozens of OEM design card horror stories, but it would take me hours/days to recall them all. Just take my word for it: do not buy an OEM design GPU unless you really know what you're doing. There are some OEM design cards worth buying, but you have to know a lot about GPUs to make that determination. Play it safe and just buy a reference design card.

[Difermo]

since subject is reference and non-reference cards, ill ask here

I read that reference cards are not for 24/7 use and that they have heating problems. Bigger then OEM cards that are for 24/7 use.
Is this true?