RTX3080

[James Heinrich]

Originally Posted by moebius

I think the rating of mersenne.ca is totally exaggerated

Hence the big red disclaimer and request for benchmarks.

[moebius]

Originally Posted by James Heinrich

Hence the big red disclaimer and request for benchmarks.

To be honest, this card is unlikely to outperform any RX Radeon VII, I am happy to do a 1: 1 benchmark with my Vega64, the differences should be marginal according to my calculations.

[xilman]

Originally Posted by M344587487

Don't you want to see how far off topic this thread gets before someone has a card to test? Call it a social experiment and we're golden. Personally I'm hoping for a decent bread recipe.

I don't know whether this counts as decent but some years ago I discovered that replacing roughly 20% of the regular wheat flour in a standard soft-white bread recipe with pea flour works well. To make the flour put dried green peas in a coffee grinder.

The texture turns out different, more like ciabatta, and the colour is a pale khaki but the flavour is superb. The bread is protein-enhanced too.

[pinhodecarlos]

This guy at SRBase has a RTX 3090 but all wus were aborted. Hope to see some figures from his runs (TF)

http://srbase.my-firewall.org/sr5/sh...?hostid=204975

[M344587487]

Originally Posted by xilman

I don't know whether this counts as decent but some years ago I discovered that replacing roughly 20% of the regular wheat flour in a standard soft-white bread recipe with pea flour works well. To make the flour put dried green peas in a coffee grinder.

The texture turns out different, more like ciabatta, and the colour is a pale khaki but the flavour is superb. The bread is protein-enhanced too.

Thanks for the tip. I'm tempted to try it with ground frozen peas and less added water, odds are not great that that would work well but it's worth a shot.

[James Heinrich]

Somebody submitted an RTX 3090 mfaktc benchmark last night. Unfortunately it lacked clockspeed information so is not particularly useful.

[aheeffer]

Originally Posted by James Heinrich

Somebody submitted an RTX 3090 mfaktc benchmark last night. Unfortunately it lacked clockspeed information so is not particularly useful.

Let us assume 1700 Mhz. That is the boost frequency of the Nvidia's FE card and factory overclocked cards from Gigabyte and others go only marginally higher (1725 Mhz). Where do we get then?

[moebius]

Originally Posted by James Heinrich

Somebody submitted an RTX 3090 mfaktc benchmark last night. Unfortunately it lacked clockspeed information so is not particularly useful.

Sounds like you're not satisfied with the values, may somebody finally unravel the Nvidia mystery. Nothing against you, but I would need concrete values ​​with specific software to make a purchase decision.

[James Heinrich]

Originally Posted by moebius

Sounds like you're not satisfied with the values, may somebody finally unravel the Nvidia mystery. Nothing against you, but I would need concrete values ​​with specific software to make a purchase decision.

I'm not satisfied with the values in terms of providing reliable benchmark data for predicting performance, no. If you get a certain performance number at 1400MHz or 1700MHz or 1900MHz then that's very different results.

Originally Posted by aheeffer

Let us assume 1700 Mhz. That is the boost frequency of the Nvidia's FE card and factory overclocked cards from Gigabyte and others go only marginally higher (1725 Mhz). Where do we get then?

According to data posted by Neutron3529 (who is likely the mystery person who submitted the benchmark to me) the throughput of the 3090 is about 4700 GHd/d at 82% GPU usage at 1695MHz (unconfirmed if that's during the run, that the mfaktc-reported speed on launch), so ~5730 GHd/d overall (at 1695MHz, 4730GHd/d at stock 1400MHz) so that's what I've calibrated the performance table to for now:
https://www.mersenne.ca/mfaktc.php?filter=RTX+30%3F0

[moebius]

Originally Posted by James Heinrich

I'm not satisfied with the values in terms of providing reliable benchmark data for predicting performance, no. If you get a certain performance number at 1400MHz or 1700MHz or 1900MHz then that's very different results.

I'm not referring to the TF performance, which is usually better with nvidia, but to the PRP performance which is ultimately decisive for me.

[jwnutter]

Originally Posted by James Heinrich

so ~5730 GHd/d overall

Wow. I think once inventory floods back into the market I might have to look at an upgrade.