RTX3080

[M344587487]

Originally Posted by xilman

Not sure it would work but worth trying at least once. Probably not more than once but who knows?


Smashed up fresh (or defrosted frozen) peas makes a very good replacement for avocado for an guacamole-like dish. All other ingredients remain the same as in the original.
...

Probably not, but a big factor is that I have frozen peas and they're probably a better substitute than the 20 year old lentils at the back of the cupboard.

Originally Posted by xilman

...
Is enough hardware available yet, or do we need to continue along these lines? Presumably these (arguably more interesting) posts can be moved to the recipes thread in due course.

Until there's a single relevant benchmark IMO it's open season. Sloe gin is normally very sweet because typical recipes say to wait at least 3 months and add too much sugar. Whatever recipe you follow use a little less sugar and crack it open after 2-3 weeks instead of 3 months. More more tart, much more drinkable.

[LaurV]

Originally Posted by axn

Unless they go for less money than an RTX 3080, you're better off buying that one, as it gives better performance.

They already go for less money from the first day of launching.

[jwnutter]

@xilman & @M344587487

We have an excellent pandemic sourdough starter if there's any interest.

[axn]

Originally Posted by LaurV

They already go for less money from the first day of launching.

They = ?

I mean, 3080 is better performance than 2080 Ti. So if the second hand 2080 Tis are not offered _substantially_ cheaper than brand new 3080, you're better off going with 3080. Since the brand new 2080 Tis were something like $1200, and 3080s are $700, I don't know how much cheaper the second hand 2080 Tis will be.

[jwnutter]

Does anyone have empirical knowledge on the mean time to failure for an RTX GPU that continuously runs flat out at factory clock? I'm guessing 3 years, maybe 4??? The 2080 came out around September 2018, so maybe not enough cycles as of today to really know. That said, this has to be an engineering design parameter for Nvidia.

[VBCurtis]

Originally Posted by jwnutter

Does anyone have empirical knowledge on the mean time to failure for an RTX GPU that continuously runs flat out at factory clock? I'm guessing 3 years, maybe 4??? The 2080 came out around September 2018, so maybe not enough cycles as of today to really know. That said, this has to be an engineering design parameter for Nvidia.

If a fan fails but is replaced, does that count as a failure for this time measurement?

It's common for GPUs to last until a fan dies- and the card's value at that time strongly influences whether the fan is fixed or considered to have killed the card.

Silicon lasts a lonnnng time.

[jwnutter]

Originally Posted by VBCurtis

If a fan fails but is replaced, does that count as a failure for this time measurement?

It's common for GPUs to last until a fan dies- and the card's value at that time strongly influences whether the fan is fixed or considered to have killed the card.

Silicon lasts a lonnnng time.

Yeah, I think that works. However, depending on age at time of failure, this could be classified as an infant mortality failure. I suppose my question was geared more towards wear-out failures. https://en.wikipedia.org/wiki/Bathtub_curve

[Xyzzy]

Originally Posted by jwnutter

Does anyone have empirical knowledge on the mean time to failure for an RTX GPU that continuously runs flat out at factory clock? I'm guessing 3 years, maybe 4??? The 2080 came out around September 2018, so maybe not enough cycles as of today to really know. That said, this has to be an engineering design parameter for Nvidia.

After enough time the thermal paste will dry out, so a user-serviceable design is something to consider.

Somewhat relevant video:

[jwnutter]

Originally Posted by Xyzzy

After enough time the thermal paste will dry out, so a user-serviceable design is something to consider.

Very interesting video, thanks for sharing. Seeing as my 2080 runs under full load at nearly all times (70C @ 60% fan output) I don't think thermal cycling will be much of an issue for my installation. Time will tell I guess.

[kriesel]

Originally Posted by VBCurtis

If a fan fails but is replaced, does that count as a failure for this time measurement?

It's common for GPUs to last until a fan dies- and the card's value at that time strongly influences whether the fan is fixed or considered to have killed the card.

Silicon lasts a lonnnng time.

It's also common for the gpu ram to go flaky, the PCB to short, or other failure mechanisms while the fans are still working. I had a GTX1070 short at 27 months (warranty was 24 months). It's too soon to tell on the RTX2080s.
If nothing else fails, eventually the throughput per utility cost and slot occupation becomes too low to justify continued operation. I have a functional NVS295 received with used system. 2. GhzD/day TF, 23 watts is a very bad deal these days.

[LaurV]

Originally Posted by axn

They = ?
I mean, 3080 is better performance than 2080 Ti. So if the second hand 2080 Tis are not offered _substantially_ cheaper than brand new 3080, you're better off going with 3080. Since the brand new 2080 Tis were something like $1200, and 3080s are $700, I don't know how much cheaper the second hand 2080 Tis will be.

"They" is 2080 Ti, we thought it clear from the text. It was the same "they" you used. Look to the market man, except for a handful of 3080 distributed at launching, there was no 3080 sold below $1000, few were sold as high as $6000. Currently, they go for $1500, every site that claims lower price is "out of stock". After Nvidia fixes the capacitors bug and comes out with a "stable" series (when the most people who already bought it and have it in their hands will go for RMA/replacement, be sure of that!), the prices will float around $1200. While for 2080 Ti, we paid $1300 for our "combined cooling" (water+air) toys long time ago, and the prices continued to drop ever since. Easy way, if you don't want to browse, go to James' site and click to the prices column, for each card, you can see a graphic with evolution of the prices (wherever there is a known one). Now you can buy a 2080Ti air cooled for as low as $750 already. After all the craziness will cool down, we expect that the 2080Ti will go as low as $500-$700, and the 3080 for $800-$1000, while the 3090 may stabilize around $1500. The issue is the fact that, even they ARE better, they are by far NOT so good as the people expected them to be. Some even talk about a "stunt" (as Nvidia did every time to sell their "intermediary experiments"), and a "Ti" version may come later for the 3xxx series, at not-much-higher price (when the 2080Ti's will go even lower, haha).