Speculation on the PPD in F@H of the new 3080 and 3090 Ampere cards ?

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

VirtualLarry

No Lifer
Aug 25, 2001
56,227
9,990
126
If its like last time. the miners bought up virtually all of the video cards, and due to supply and demand, the prices skyrocketed and availability was almost non-existant. Do you blame the MFG'ers for not predicting this ? and taking 3 months to up production, since it was not in their plan ? They are supposed to have a crystal ball, and know everything ?
Well, it's the law of supply AND demand. There was increased demand from miners, but that's only HALF of the equation. The other half, I place squarely on the mfgs. If their lead time on mfg of cards, is such that a "burst" of demand, can wipe out supplies, and raise retail prices to 2x-3x of what they should be, then that's not the miner's fault. Honestly, too, it was the mfg's that held back production ON PURPOSE, and saw that people were willing to pay higher prices for the cards, so they RAISED PRICES for future generations of GPUs. Again, that's ALL on the mfg's, pretty-much, and not the miners.

AMD was almost in the same place with Zen. They had no idea how quickly it would catch on, and it took them a while to get supply up to demand. But on a whole nother level. They were a little short, and prices did not skyrocket.
I'll give you credit for that one, although, due to COVID, AM4 mobos were in short supply, and prices did rise to 2x-3x MSRP on popular or harder-to-find mobos. But you are correct, CPU prices themselves didn't rise much over MSRP, if they did at all. (*)

(*) Edit: I just checked Newegg, and for the first time in a long time, the Ryzen R5 3600 6C/12T Zen2 CPU is @ $209.99 right now, from Newegg themselves, which I believe IS over MSRP. (That's actually kind of distressing, I wanted another one for one of my own rigs, I should have ponied up the money when they were $154.99 or somesuch over at Amazon.)
 
Last edited:

StefanR5R

Elite Member
Dec 10, 2016
5,459
7,718
136
Furthermore, not the fault of miners are:
Dirt*-cheap coal power generation (*literally) which is heavily subsidized not just by current but more so by future society.
The ongoing crisis of the financial industry.
Negative interest policies.
Demand for the service of letting money disappear and reappear.

But back to the topic: Manufacturers generally cannot increase production on short notice easily. On top of that, they won't increase production even if they could if there is indication that a certain demand is not going to develop into sustained business.

AFAIK a new wave of GPU mining in volume could be (if at all) a thing mostly in Asia, with cheap coal power generation around, and where quantities of equipment can be had without the delay of transoceanic shipping, and overhead incurred by distributors.
 

Kiska

Golden Member
Apr 4, 2012
1,007
290
136
But back to the topic: Manufacturers generally cannot increase production on short notice easily. On top of that, they won't increase production even if they could if there is indication that a certain demand is not going to develop into sustained business.

And to add to this point, AMD got really burned in the R9 2xx days where bitcoin mining was a thing, by the time they and TSMC ramped up production, the cards were no longer in demand
 

lane42

Diamond Member
Sep 3, 2000
5,721
624
126
"Tom's Hardware speculates that cryptomining could raise prices again"
2 underlined words mean everything. IMO, it could but it won't. :)
 
  • Like
Reactions: Ken g6

StefanR5R

Elite Member
Dec 10, 2016
5,459
7,718
136
No benchmarks yet; these are still under NDA, and the duration of that NDA is under NDA too.
This latter NDA was lifted. Review reports of 3080 are under NDA until September 16, 06:00 PDT, of 3090 until September 24, of 3070 until October 15.
 

Endgame124

Senior member
Feb 11, 2008
954
669
136
This latter NDA was lifted. Review reports of 3080 are under NDA until September 16, 06:00 PDT, of 3090 until September 24, of 3070 until October 15.
Interesting. So by the time you can find out how fast the 3090 is, the 3080s will be sold out. Probably closer than they liked to admit for the price difference
 

StefanR5R

Elite Member
Dec 10, 2016
5,459
7,718
136
Here are several meticulous electrical and thermal measurements, besides performance tests: igorslab.de
As far as I can tell, page 8 "Studio: Rendering" is indicative for FP32 computational workloads, and page 9 "Studio: Video & Picture Editing" represents integer workloads with various word widths, right?

Of course, the particular device tested here, 3080 FE, will apparently be quite irrelevant in the market due to limited production quantity.
 
  • Like
Reactions: biodoc

biodoc

Diamond Member
Dec 29, 2005
6,257
2,238
136
There's at least one 3080 folding as reported in the folding forum. There's a link there to this forums thread where there is a FAHControl image. The owner of the card obtained a passkey but the QRB hasn't kicked in yet. According to the FAHControl image, the card is working on project 16918 and completing work at a TPF of 57 seconds. I plugged those numbers into the FAH bonus point calculator. and it's making at just over 3.9 M PPD. The project is a relatively small number of total atoms so it may do better on projects with a larger number of atoms.

1600525471343.png
 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,482
14,434
136
There's at least one 3080 folding as reported in the folding forum. There's a link there to this forums thread where there is a FAHControl image. The owner of the card obtained a passkey but the QRB hasn't kicked in yet. According to the FAHControl image, the card is working on project 16918 and completing work at a TPF of 57 seconds. I plugged those numbers into the FAH bonus point calculator. and it's making at just over 3.9 M PPD. The project is a relatively small number of total atoms so it may do better on projects with a larger number of atoms.

View attachment 30171
I just checked. I have a 2080TI with that unit doing it in 1 min 3 seconds for 3.3m ppd
 

lane42

Diamond Member
Sep 3, 2000
5,721
624
126
I 'am sure with driver updates the 30xxx cards will get much better.
When i first got my 2080ti some 1.5 years ago i was only getting
about 2 Mppd in F@H.
 

biodoc

Diamond Member
Dec 29, 2005
6,257
2,238
136
I just checked. I have a 2080TI with that unit doing it in 1 min 3 seconds for 3.3m ppd
I'm assuming you're on Linux. The 3080 owner is running Windows so it may be a bit better on Linux. There is a driver version for linux so that's encouraging.
Linux x64 (AMD64/EM64T) Display Driver WHQL 455.23.04 17.9.202

Odd that it's WHQL certified for Linux??
 
  • Haha
Reactions: TennesseeTony

StefanR5R

Elite Member
Dec 10, 2016
5,459
7,718
136
Sometimes their reviews come out a bit later than on other sites (some of which do everything to post their report at the very second when the embargo times out). On the other hand, I recall that some GPUs which were supposed to be reviewed never made it onto AnandTech.

Surely Nvidia sent samples only to a limited number of sites, but my guess is that AT was among these.
 

StefanR5R

Elite Member
Dec 10, 2016
5,459
7,718
136
Surely Nvidia sent samples only to a limited number of sites,
...and even fewer of the 3909FE, of course. Igor Wallossek said that German review sites apparently received 2 (two) samples of 3090FE, in total. AIB partner companies sent only a few samples too.
 

biodoc

Diamond Member
Dec 29, 2005
6,257
2,238
136
Igor's lab discovered the source of instability issues with some 3080 manufacturer's cards. High quality components (capacitors) are necessary for high boost clock stability. :)
 

Orange Kid

Elite Member
Oct 9, 1999
4,323
2,110
146
Well, I won't have to worry about owning a subpar 3000 series video card any time soon with the usual non existant inventory.
 

StefanR5R

Elite Member
Dec 10, 2016
5,459
7,718
136
Igor's lab discovered the source of instability issues with some 3080 manufacturer's cards. High quality components (capacitors) are necessary for high boost clock stability. :)
The board makers didn't find that out in time though, because they received a proper driver from NVidia far too late. (Or they went to market far too early after they received the first real driver; take your pick.)

They developed the boards with a pseudo driver which did not much more than allow for some basic thermal and power testing. Source: igorslab too, I don't recall whether this was in an article or in a forum post.

Another driver version was released now which reduces the microsecond long power draw spikes somewhat.
 

AshlayW

Member
Sep 6, 2018
32
97
91
www.eridonia-archives.com
RTX 3080's core is configured similarly to RTX 2080 Ti, in that it has 68 SMs. The difference is the Ampere SM has twice the FP32 rate per clock, so instead of 64x FMAC ops per clock you get 128. That essentially means in pure FP32 compute (I assume F@H is pretty much that) it could theoretically deliver up to 2X the performance of the RTX 2080 Ti in PPD. The Ampere cards also feature much faster GDDR6X and despite the narrower bus (320b vs 352b on the 2080 Ti) the RTX 3080 has 23% more memory bandwidth on tap. Imperfect scaling is likely to be encountered as Compute can be bandwidth gated so I would expect slightly under 2X the performance. My 2080 Ti crunches around 3-5M PPD depending on the WU, so I would say 6-8M is a safe bet for this card at 320W, would would indeed make it more efficient than RTX 2080 Ti at 250W.
 

Endgame124

Senior member
Feb 11, 2008
954
669
136
The 3080 and 3090 are so wide that a bunch of WUs won’t fully utilize the card, much like the 2080ti. I’m guessing the 3060 and 3070 will end up very close to the 3080 for probably 50% or more of the WUs.

I’m curious where the new AMD cards will land as they won’t have the advantage of CUDA, but they will have that 128mb of ultra fast ram available.