Nvidia 3xxx possible specs.

Having problems installing that new stick of memory? Found some great software or having issues with something? Or maybe want to chat about your PlayStation, X-Box, Nintendo, Sega, even your old Spectrum 48k....! Or maybe something you want to sell or acquire (computing related of course!). Let us know here...
damienh
Active UBT Contributor 10+ yrs
Posts: 1586
Joined: Mon Aug 26, 2019 2:16 pm

Re: Nvidia 3xxx possible specs.

Post by damienh » Thu Sep 17, 2020 3:22 pm

The nvidia-direct cards do look to be better value. The cheapest partner card is over £700.

I haven’t read the reviews in depth to see what they say about thermals mind you. I believe this card is >300w.

Woodles
Active UBT Contributor 20+ yrs
Posts: 11561
Joined: Thu Dec 20, 2007 12:00 am
Location: Cambridgeshire

Re: Nvidia 3xxx possible specs.

Post by Woodles » Thu Sep 17, 2020 3:41 pm

And I've just this minute received an email from NVidia saying they were now available :lol:

ASUS and Gigabyte both had cards at £649.98 on Scan and EBuyer.

The RTX 3080 is a 320W card, temperatures seem to stabilise in the mid to high 70s when playing games according to one review I saw.

Woodles
Active UBT Contributor 20+ yrs
Posts: 11561
Joined: Thu Dec 20, 2007 12:00 am
Location: Cambridgeshire

Re: Nvidia 3xxx possible specs.

Post by Woodles » Sun Sep 20, 2020 12:04 pm

There MAY be variants with more memory coming - https://videocardz.com/newz/gigabyte-co ... -3080-20gb

I expect they'll be a while yet judging by the demand for the first ones, why produce different cards when you can sell all you make of the original ones?

Woodles
Active UBT Contributor 20+ yrs
Posts: 11561
Joined: Thu Dec 20, 2007 12:00 am
Location: Cambridgeshire

Re: Nvidia 3xxx possible specs.

Post by Woodles » Mon Sep 21, 2020 10:01 am

Some info from PrimeGrid - https://www.primegrid.com/forum_thread.php?id=9294
They claim 2x FP32 performance among other things, which sounds really good for now...
Yes, 2x FP32 since 2x FP32 compute units.
PG softwares use FP32 right?
No, all of them use INT32.
Each Pascal core is a single INT32+FP32 unit. Turing has two separate units: INT32 and FP32. Each RTX-30 Ampere core has two units: INT32+FP32 and FP32. Then RTX-30 can perform 2x FP32 but 1x INT32.
3080 has 8704 'CUDA cores' but that's counting the FP and FP/INT cores separately, really it's 4352 FP and 4352 FP/INT.
For PrimeGrid it's 4352 cores (3080) versus 4352 cores (2080ti)
So not necessarily any faster on PrimeGrid. Better original price than a 2080ti, possibly worse power (although with half the CUDA cores, all the RTX and all the Tensor cores not being used, it may work out less)

The cores are clocked slightly higher and use a different architecture so should show some improvement.
Actual figures (against a base 2080) - https://www.primegrid.com/forum_thread. ... rue#143381
AP27:
2080: 660-720
3080: 290

PPS Sieve: (2 tasks)
2080: 215
3080: 170

DYFL:
2080: 103000-110000
3080: estimate after 15 mins = ~72000
Probably not worth it for pure PrimeGrid.

Woodles
Active UBT Contributor 20+ yrs
Posts: 11561
Joined: Thu Dec 20, 2007 12:00 am
Location: Cambridgeshire

Re: Nvidia 3xxx possible specs.

Post by Woodles » Mon Sep 21, 2020 10:11 am

More figures - https://www.primegrid.com/forum_thread. ... rue#143492
My UPS shows 64% load with it running GFN and 25% load without. It's a Smart UPS-1000 which is supposed to have 700W maximum load which means the 3080 is using 210W.

Some actual runtimes to confirm the earlier benchmarks. The DYFL time is probably a bit slower than optimum as I had a few hours of doing some other tests without suspending it.

24s Genefer 15 v3.21
52s Genefer 16 v3.21
135s Genefer 17 Low v3.21
145s Genefer 17 Mega v3.21
455s Genefer 18 v3.21
1,020s Genefer 19 v3.21
4,262s Genefer 20 v3.21
65,973s Do You Feel Lucky? v3.21

Woodles
Active UBT Contributor 20+ yrs
Posts: 11561
Joined: Thu Dec 20, 2007 12:00 am
Location: Cambridgeshire

Re: Nvidia 3xxx possible specs.

Post by Woodles » Fri Sep 25, 2020 1:51 pm

Well after all the excitement over the RTX3080 launch, the RTX3090 launch passed pretty quietly, were any boards actually released?

scole of TSBT
The Scottish BOINC Team member
Posts: 43
Joined: Wed Dec 07, 2011 12:00 am

Re: Nvidia 3xxx possible specs.

Post by scole of TSBT » Sun Sep 27, 2020 1:04 am


Woodles
Active UBT Contributor 20+ yrs
Posts: 11561
Joined: Thu Dec 20, 2007 12:00 am
Location: Cambridgeshire

Re: Nvidia 3xxx possible specs.

Post by Woodles » Sun Sep 27, 2020 10:10 pm

Hi Steve, most members here can't access the TSBT forums, they only get to see the first post.

I assume this is the bit you meant?
#18 Re: 30 series
Post by scole of TSBT » Sat Sep 26, 2020 4:06 pm

https://videocardz.com/newz/manufacture ... top-issues

scole of TSBT
The Scottish BOINC Team member
Posts: 43
Joined: Wed Dec 07, 2011 12:00 am

Re: Nvidia 3xxx possible specs.

Post by scole of TSBT » Sun Sep 27, 2020 10:32 pm

Yes, that's the article. John P. Meyers of SETI.USA is engineer and critiques most new hardware. This is another of his comments...
Many of them that didn't do so well in the silicon lottery are crashing, artifacting, freezing up, blackscreening, etc. Because of that, GPU makers will start issuing new BIOS versions that downclock the GPUs so they'll work properly. Of course this means crunching perf will also drop. Not sure yet by how much.
RTX 3090 $1500 24GB VRAM 10,496 cores - ? TFLOPS - 350W
RTX 3080 $700 10GB VRAM 8,704 cores - ? TFLOPS - 320W
RTX 3070 $500 10GB VRAM 5,888 cores - ? TFLOPS - 220W

I've left the TFLOPS out because I've heard the CCs are now single issue instead of double. If true, the 3090 would be just under 15 TFLOPS, if not almost 30 TFLOPS. Need to look into it further
M'kay i believe i have it figured out. What I believe they did was added a second vector float ALU pipeline that shares the same register space and execution control flow. Yes, you can say they have doubled the core count but you should also consider that they are "less capable" cores. Similar to what AMD pulled off with bulldozer cpus. And got sued over.

Basically, each core can do FP+INT at the same time, just like the 2000 series. But it can also now do FP+FP instead. However, when it does FP+INT at the same time, the total TFLOPS are cut in half. *sigh* always with the shady gimmicks

Woodles
Active UBT Contributor 20+ yrs
Posts: 11561
Joined: Thu Dec 20, 2007 12:00 am
Location: Cambridgeshire

Re: Nvidia 3xxx possible specs.

Post by Woodles » Sun Sep 27, 2020 10:41 pm

Yeah, as I understand it each CUDA 'core' has two execution units in it, one does FP and the other does FP + INT.

Nvidia are claiming each unit and stating the 3080 has 8704 cores.

Add up the FP and you do indeed get 8704 compute units but for pure INT loads, half the chip is idle (4352 compute units active)

UBT - Timbo
UBT Forum Admin
Posts: 8814
Joined: Mon Mar 13, 2006 12:00 am
Location: NW Midlands
Contact:

Re: Nvidia 3xxx possible specs.

Post by UBT - Timbo » Mon Sep 28, 2020 1:01 am

Woodles wrote:
Fri Sep 25, 2020 1:51 pm
Well after all the excitement over the RTX3080 launch, the RTX3090 launch passed pretty quietly, were any boards actually released?
Hi Mark

This might interest you:

https://www.theregister.com/2020/09/24/ ... _shortage/

Seems NVidia got caught out on the 3080 launch and tried to prevent scalpers and bots getting hold of the 3090s that were made available.

It's not clear how well they succeeded, but there are reports of some 30x0 GPUs on sale on ebay at greatly increased prices...which perhaps are scammers or people wanting to make a fast buck.

regards
Tim

homefarm
Active UBT Contributor 5+ yrs
Posts: 255
Joined: Thu Dec 15, 2016 6:41 pm
Location: Sussex
Contact:

Re: Nvidia 3xxx possible specs.

Post by homefarm » Fri Oct 30, 2020 8:45 am

I think Nvidia must be reconsidering their PR status. No doubt they are paying someones mega bucks, basically to pi$$ people off. Nothing available to begin to satisfy demand.

I also wonder just how the speed increases will pan out in real world calculations in live projects. :?: Some figures seem spectacular, but not all tasks are optimised or able use all the features available. Even the compilers etc will need updating etc.

Post Reply