RTX3080

[LaurV]

Originally Posted by VBCurtis

Silicon lasts a lonnnng time.

+1. We still have 580 cards (early Fermi architecture), 4 of them, all in perfect working conditions, of which we bought 2 new, and replaced the mammoth air coolers with EKWB (there is a post here around with this work we did, the original coolers were like 2.8 slots wide, and they actually took all the space inside the case, not 3 slots as you expected, but 4, for each card, because if you have more than one, you need a little space in between too! and without raisers, there was no way to use them, so they were "out" immediately, in spite of the fact that we paid for ekwb's same amount of money as for one new card, at the time), and the other two we bought second hand were already modified to water (sapphire blocks) by the seller, so they were used by him for I have no idea how long. All 4 are in perfect working condition (did I say that?) after about 10 years of service, in and out in my different rigs (production, and experimental) along the years...

[Mark Rose]

I recently ditched my 580s because I couldn't put up with their inefficiency compared to modern cards.

I might pick up a 3080 for TF once the prices drop, but I'm going to wait until after AMD releases their new wares.

[Xyzzy]

Just a few years ago we had four 580s running. The electrical use (1kW) and heat generated was tremendous. Now we can easily duplicate that setup with a single 175W 2060 Super.

[James Heinrich]

Originally Posted by Xyzzy

Just a few years ago we had four 580s running. The electrical use (1kW) and heat generated was tremendous. Now we can easily duplicate that setup with a single 175W 2060 Super.

Or (barring purchase cost) replicate the power usage with 3x RTX 3080 and achieve 700% throughput for the same 1kW.

[Xyzzy]

FWIW, we have our 2060 Super running at ~120W and it yields 1.9-2K GHz-d/d, so we could run eight cards in the same 1kW for 15-16K GHz-d/d output. Crazy!

[tServo]

I'm beginning to think the demand is so great for these that you won't see any before the New Year.

[James Heinrich]

Originally Posted by tServo

I'm beginning to think the demand is so great for these that you won't see any before the New Year.

Stock levels will become a lot more complicated over the next month, with not only the 3080 in huge demand, presumably only small demand for the 3090, but then next week the 3070 being announced and I expect even bigger demand for that than the 3080 due to it's lower price point. Then the revised version of the 3080 (and maybe 3070 too depend how they launch it) with more RAM to counter the AMD launch at the end of the month...

[LaurV]

Originally Posted by tServo

I'm beginning to think the demand is so great for these that you won't see any before the New Year.

The demand is not "so great". It is just normal. The fact is that the offer is so small, close to zero. They all had technical problems which were (or still are being) fixed. Those who already bought the cards may even expect a recall, or will need to RMA them in the future. I won't jump to buy until the "market stabilizes". In fact, as I said before, I won't jump to buy them at all. They don't seem "more performant" to me. They reduced the power consumption by "cutting the transistors in half" (haha, this sounds soooo stupid, no, they didn't reduce the number of transistors, neither sliced every transistor and put only half of it on the silicon die, and discard the other half, what I wanted to say is the fact they went from 14nm to 7nm technology), which allows them to put more transistors there in the same volume of silicon, and run them at higher clock for about the same power consumption. But think about it a little: if my actual cards would run at 2GHz (like the 3080s do) they would produce almost double performance of the actual (i.e. 8THzDays/Day instead of actual 4). So, the 3080, at 6THzD/D, is not a buff, it sounds more like a nerf, just to use the "gamer's lingo".

Yeah, I know a am a bitchy son of a gun, but what I want you to get from here, don't jump to buy it until the market stabilizes, probably somewhere next spring (European spring). At that time you may get better prices and more stable cards, and the production won't be much different meantime, if you order the card now and still run your actual 2080s till December, when the card will come, or you order the card in March and get it in a week...

We may also be seeing (no "inside info", just a wild ass guess here, but a lot of hopes too...) a "Ti", or "buffed" version of it till then, so you will feel sorry for the "early buy".

[mackerel]

Originally Posted by LaurV

The demand is not "so great". It is just normal. The fact is that the offer is so small, close to zero.

The demand is great, unprecedented. Stock levels at launch were comparable to the 20 series. Web traffic was stated as being bigger than black friday. Some will argue how much faster they are (at gaming) but that kinda misses the point they are a LOT faster than their top performing predecessor, and at a lower price. It is a perfect storm. Anyone from Pascal generation or older will see really massive gains and there are a LOT of people looking to get that uplift.

I see a recall as extremely unlikely. The stuff floating around the net is at mostly armchair commentary by those who don't understand electronics. I have worked as an engineer for over two decades in the industry and I can't say I understand all the nuances that go into the power delivery design. Even those with the ability to modify hardware has not show any more than a miniscule difference in maximum overclocks by changing caps. An updated driver has already been released which is reported to help in many situations, although we'll probably never get a clear picture of what are genuine hardware problems vs driver or other initial compatibility problems, with unofficial reports of actual RMA rates consistent with the product category. Misdiagnosis is too easy by the masses who hear of a problem then attribute it to that, regardless that the true cause may not be found.

Where does the cutting transistors in half even come from? They did not go from 14nm to 7nm, they went from TSMC 12nm to Samsung 8nm. You can't even compare the process numbers between different fabs since they don't measure in the same way. TSMC seem more aggressive on using small numbers, as for a comparable node to Intel they're offset about as much e.g. Intel 10nm is comparable to TSMC 7nm. I don't know where Samsung 8nm fits in to that but it is a given it is not as advanced as TSMC 7nm.

[retina]

Originally Posted by mackerel

I don't know where Samsung 8nm fits in ...

See here:

https://www.techcenturion.com/7nm-10nm-14nm-fabrication

Samsung 8nm is lower density than both TSMC 7nm and Intel 10nm.

[LaurV]

Originally Posted by mackerel

Where does the cutting transistors in half even come from? They did not go from 14nm to 7nm, they went from TSMC 12nm to Samsung 8nm.

Whoops! Here my info was outdated! Sometime ago I read on the web about them being 7nm, and later I assumed it's so, due to the 2G clock. But that is even worse, it means they can't really get a large frequency boost...