RTX 4090/4080 spec /price
 

Hi.
[URL]https://www.tomshardware.com/news/nvidia-geforce-rtx-4090-rtx-4080-price-release-date-specs-revealed[/URL]


basically 4090/3090 TI
Transistors 76b/28.3b
Streaming multiprocessor 128/84
GPU core 16384/10752
TFlops FP32 82.6/40

edit, found [url]https://www.techpowerup.com/gpu-specs/geforce-rtx-4090.c3889[/url]
FP64 is 1:64 FP32 speed, so that mean 1.290 TFlops against 0.556 TFlops fpr the 3090.

Note that the 12GB "4080" is really just a 4070.

:mike:

I have added entries for the 40[89]0 but of course no benchmarks yet.
[url]https://www.mersenne.ca/mfaktc.php?filter=rtx+4[/url]

I know it's just a number but not so long ago breaking the billion transistor barrier was impressive. Doubt we're ever going to hit a trillion transistors on a monolithic die, 100 billion may happen but not much beyond that as everyone transitions to MCM.



Rumour has it that the 40 series is going to cost a lot for a long time as there is a backlog of stock of 30 series. And mining should be crashing as Ethereum is now PoS so 2nd hand cards will keep them in stock for a while. And epic recession on the horizon etc etc.


4nm TSMC is tasty, it's a shame it's an nvidia card with questionable design choices like 1:64 and extreme power consumption. Hopefully low end 40 series underclocked and undervolted at reasonable prices end up existing and perform well.

[QUOTE=M344587487;613917]Rumour has it that the 40 series is going to cost a lot for a long time as there is a backlog of stock of 30 series. And mining should be crashing as Ethereum is now PoS so 2nd hand cards will keep them in stock for a while. And epic recession on the horizon etc etc.[/QUOTE]


[URL]https://arstechnica.com/gaming/2022/09/do-expensive-nvidia-graphics-cards-foretell-the-death-of-moores-law/[/URL]

"When asked about those price increases, Nvidia CEO Jensen Huang told the gathered press to, in effect, get used to it. "Moore's law is dead," Huang said during a Q&A, [URL="https://www.digitaltrends.com/computing/nvidia-says-falling-gpu-prices-are-over/"]as reported by Digital Trends[/URL]. "A 12-inch wafer is a lot more expensive today. The idea that the chip is going to go down in price is a story of the past."
...
"Generational price comparisons aside, Huang's blanket assertion that "Moore's law is dead" is a bit shocking for a company whose bread and butter has been releasing graphics cards that [URL="https://seekingalpha.com/article/4502479-nvidia-growth-beyond-moores-law"]roughly double in comparable processing power every year[/URL]. But the prediction is far from a new one, either for Huang—who [URL="https://www.cnet.com/tech/computing/moores-law-is-dead-nvidias-ceo-jensen-huang-says-at-ces-2019/"]said the same thing in 2019[/URL] and [URL="https://www.extremetech.com/computing/256558-nvidias-ceo-declares-moores-law-dead"]2017[/URL]—or for the wider industry—the International Technology Roadmap for Semiconductors [URL="https://arstechnica.com/information-technology/2016/02/moores-law-really-is-dead-this-time/"]formally announced it would stop chasing the benchmark[/URL] in its 2016 roadmap for chip development."


I have no problem switching to the red side for cards...

[QUOTE=LordJulius;613929]I have no problem switching to the red side for cards...[/QUOTE]

I've been watching Radeon VII prices on eBay since the Etherum PoS change. I've seen Buy-it-now prices around $400. One miner was offering 4 cards for $1000 today.

I should have my head examined for even considering another card. Though I do wonder if one could increase throughput noticeably running 5 cards at lower power vs. 4 cards at higher power (same total power consumed).

[QUOTE=Prime95;613932]I do wonder if one could increase throughput noticeably running 5 cards at lower power vs. 4 cards at higher power (same total power consumed).[/QUOTE]I run mine at -20% from nominal power on Windows which only costs ~-7% throughput. I've seen an occasional card go into an odd mode sometimes in Windows on a multi-RadeonVII system, where 1 card will produce ~half the speed, at ~one third the power and gpu clock rate, often preceded by gpu->host repeated error, and the AMD software can't change the GPU clock until the card is disabled and enabled again in device manager. So if one could force them to ~700MHz GPU clock, and leave ram clock at ~11xx MHz, more could be run on the same power budget and produce more aggregate throughput. That gets more appealing as the unit cost drops, and utility rates go up.

For more on that, see [url]https://mersenneforum.org/showpost.php?p=611082&postcount=1[/url]
[url]https://mersenneforum.org/showpost.php?p=613936&postcount=2783[/url]

[QUOTE=LordJulius;613929][URL]https://arstechnica.com/gaming/2022/09/do-expensive-nvidia-graphics-cards-foretell-the-death-of-moores-law/[/URL]
"When asked about those price increases, Nvidia CEO Jensen Huang told the gathered press to, in effect, get used to it. "Moore's law is dead," Huang said during a Q&A, [URL="https://www.digitaltrends.com/computing/nvidia-says-falling-gpu-prices-are-over/"]as reported by Digital Trends[/URL]. "A 12-inch wafer is a lot more expensive today. The idea that the chip is going to go down in price is a story of the past."
...[/QUOTE]
Nice political wording from Jensen, the chip is a small piece of the product and most people don't need the 76 billion transistor version (which I assume has horrible yields which also doesn't help the chip price). MCM is making more sense every day.

MCM is inevitable and has been for some time. They just needed the cost/reward to hit a critical point. We appear to be reaching that point.
It was dabbled with many years ago with Core 2 Quads which were dual duos in reality.

rdna3 high end should use at least two compute dies, seems similar to Zen 1's toe-dipping. In the next few years intel is going nuts splitting up their CPU designs into modular components which might make its way to GPU, it'll at least make it to iGPU and server farms. Nvidia might not go MCM until the 60 series, they tend not to use unproven designs and given their position in the market they can probably get away with it.

I'm excited to pick one of these cards up. Terrible for PRP from the looks of it but it should be great at trial factoring!

hmmm
[quote=tomshardware]
In a surprising move by Nvidia, the company officially announced (opens in new tab) it will be pulling the RTX 4080 12GB off the launch table, and will be deleting the model for good. This comes after serious backfire from the community about Nvidia's horrible naming scheme for the RTX 4080 12GB, which featured significantly lower core counts and memory specifications compared to the 4080 16GB variant.
[/quote]

[URL]https://www.tomshardware.com/news/nvidia-unlaunches-rtx-4080-12gb[/URL]

I got my first RTX 4090 benchmark (no, I don't have one, I wish...) so you can see at least approximate mfaktc performance now. Spoiler: 4090 dominates.
[url]https://www.mersenne.ca/mfaktc.php[/url]

[QUOTE=James Heinrich;615670][url]https://www.mersenne.ca/mfaktc.php[/url][/QUOTE]Blank page is blank.

It only displays "Loading data ..."

But looking at the source everything is set to "Display: None".

Why?

hhmm

[QUOTE=retina;615673]Blank page is blank.
It only displays "Loading data ..."
But looking at the source everything is set to "Display: None".
Why?[/QUOTE]Table is rendered invisible first, then rows are displayed according to the filter specified (all rows if the filter is blank).
If you disable Javascript, you will not be able to use the page.

[QUOTE=James Heinrich;615678]If you disable Javascript, you will not be able to use the page.[/QUOTE]Perhaps I miss something here, but wouldn't making the table visible by default, and then use JS to hide stuff later, make it more accessible?

[size=1]Also, no need for the "Loading data..." thing IMO. People would see the table immediately, and all the browsers have a spinner or other indication that stuff is still loading, so it is redundant.[/size]

[QUOTE=firejuggler;615659]hmmm
[URL]https://www.tomshardware.com/news/nvidia-unlaunches-rtx-4080-12gb[/URL][/QUOTE]
Edited your post and removed tracker link from the quote. The link you gave outside the quote was right, and enough. (edit: not your fault, but when you post a quote like that, better paste it in notepad first, or any other pure-text editor, where all the text formatting and links are removed, then copy/paste it from there to the forum - that's a lot safer).
The naked Nvidia link, without the tracker, is here:
[URL]https://www.nvidia.com/en-us/geforce/news/12gb-4080-unlaunch[/URL]
Whoever is interesting of what the original link was, can go to tomshardware page and hover the mouse over it. A clever browser should remove the tracker, but not all people use clever browsers :razz:

[QUOTE=James Heinrich;615670]I got my first RTX 4090 benchmark (no, I don't have one, I wish...) so you can see at least approximate mfaktc performance now. Spoiler: 4090 dominates.
[url]https://www.mersenne.ca/mfaktc.php[/url][/QUOTE]

I had the chance to pick up one from a website but I decided to wait and see if they stock the specific model I wanted.


Looking back at my thinking... I should have bought as many of the ones I could have and resold (scalped) them on ebay!

[QUOTE=Magellan3s;615820]resold (scalped) them on ebay![/QUOTE]It's better you didn't. That tends to strip the warranty, at least in my experience with XFX.

[QUOTE=LaurV;615685]Edited your post and removed tracker link from the quote. The link you gave outside the quote was right, and enough. (edit: not your fault, but when you post a quote like that, better paste it in notepad first, or any other pure-text editor, where all the text formatting and links are removed, then copy/paste it from there to the forum - that's a lot safer).
The naked Nvidia link, without the tracker, is here:
[URL]https://www.nvidia.com/en-us/geforce/news/12gb-4080-unlaunch[/URL]
Whoever is interesting of what the original link was, can go to tomshardware page and hover the mouse over it. A clever browser should remove the tracker, but not all people use clever browsers :razz:[/QUOTE]

[QUOTE]https:// click .linksynergy. com /deeplink?id=kXQk6%2AivFEQ&mid=44270&u1=tomshardware-row-5933681771989526000&murl=https%3A%2F%2Fwww.nvidia.com%2Fen-us%2Fgeforce%2Fnews%2F12gb-4080-unlaunch%2F%3Fncid%3Dafm-chs-44270%26ranMID%3D44270%26ranEAID%3DTnL5HPStwNw%26ranSiteID%3DTnL5HPStwNw-Rt_VeFaEpKL5.5q7g71XvA

(spaces added to prevent automatic URL tags)[/QUOTE]

It's ugly as heck.

[QUOTE=Viliam Furik;615886]It's ugly as heck.[/QUOTE]
Yeah, well, yes... tomshardware has to make money too, they have one link like that in [U]every[/U] article (one only, kind of random, the rest of the links in each article are "normal"). His bad luck was that it happened to be exactly in the fragment he quoted, and he copy/pasted with link, is easy to miss, especially if your browser doesn't color/mark/underline, etc, the links. No harm done.

[QUOTE=James Heinrich;615670]I got my first RTX 4090 benchmark (no, I don't have one, I wish...) so you can see at least approximate mfaktc performance now. Spoiler: 4090 dominates.
[url]https://www.mersenne.ca/mfaktc.php[/url][/QUOTE]

Makes it almost pointless to run my 1070s.

[QUOTE=Mark Rose;618019]Makes it almost pointless to run my 1070s.[/QUOTE]They're still [url=https://www.mersenne.ca/mfaktc.php?filter=GP104-200|RX+480|4090]37% faster[/url] than my RX 480 at the same TDP.

[QUOTE=Mark Rose;618019]Makes it almost pointless to run my 1070s.[/QUOTE]

I thought you were just bringing the kit back online to heat your space using electricity during the winter?

Multiple options... Crypto (mostly a fool's game). Primes. Factors. Etc et al. Your kit; your time; your electrons. Your choice.

Work with what you are given. Please forgive me if I'm telling you how to chew gum... 9^)

Thanks for any and all cycles. Where ever you decide to send them. 8-)

[QUOTE=chalsall;618022]I thought you were just bringing the kit back online to heat your space using electricity during the winter?

Multiple options... Crypto (mostly a fool's game). Primes. Factors. Etc et al. Your kit; your time; your electrons. Your choice.

Work with what you are given. Please forgive me if I'm telling you how to chew gum... 9^)

Thanks for any and all cycles. Where ever you decide to send them. 8-)[/QUOTE]

Crypto paid for them. Broke even. I should have mined more. Only four running at the moment. I should have more hardware coming online soon.

I've got cheap electricity rates locked in for the next 22 months. Same with gas, though heating with computers is more fun. Going to see if I can break 2 MWh/month for giggles.

I'm looking at TFing the 518 exponents at 76 bits in 96M and 97M to 77 bits. It looks a bunch more above there should be TFed higher as well. Would be nice to eliminate more DC work to speed that project along.

[QUOTE=Mark Rose;618029]I'm looking at TFing the 518 exponents at 76 bits in 96M and 97M to 77 bits. It looks a bunch more above there should be TFed higher as well. Would be nice to eliminate more DC work to speed that project along.[/QUOTE]Sounds excessive to me; GTX 1070 are 1:32 SP/DP so 76 bits is enough.

[url]https://www.mersenne.ca/exponent/96095581[/url] shows a 4 bit differential, 71 for CPU, 75 bits for GPUs.

[QUOTE=Mark Rose;618029]I've got cheap electricity rates locked in for the next 22 months. Same with gas, though heating with computers is more fun. Going to see if I can break 2 MWh/month for giggles.[/QUOTE]

Giggles is a worthy goal. 9-)

And... Depending on where you are currently located, if you can heat your human space(s) using renewable electricity rather than burning gas, that's a net plus for the whole enthalpy/entropy thing... 8-)

[url]https://www.tomshardware.com/news/vendor-confirms-rtx-4070-ti-is-a-resurrected-rtx-4080-12gb[/url]

not surprisingly

[quote]
According to the specification sheet, the GeForce RTX 4070 Ti uses the AD104 (Ada Lovelace) silicon, the same die at the heart of the GeForce RTX 4080 12GB. The die, which measures 295 mm², houses up to 7,680 CUDA cores. Although Colorful didn't confirm the number of CUDA cores, we don't think Nvidia will cut the die down. So instead, the GeForce RTX 4070 Ti will likely feature the entire die, meaning that it'll also have 240 Tensor cores and 60 RT cores at its disposal in addition to the 7,680 CUDA cores.
<snip>

<snip>

Hardware leaker MEGAsizeGPU, who has a solid track record with Ada Lovelace leaks, claims that Nvidia will launch the GeForce RTX 4070 Ti on January 5. The pricing, however, remains a mystery. Nvidia was originally asking $899 for the GeForce RTX 4080 12GB. However, logic would say that the chipmaker will likely have to lower the price tag given that AMD's Radeon RX 7900 XTX and Radeon 7900 XT, which rival the GeForce RTX 4080, will soon hit the market for $999 and $899, respectively. In addition, many rumors have suggested that AMD is reportedly facing a supply issue; therefore, RDNA 3 stock may be limited at launch, and maintaining MSRP will be challenging.
[/quote]

Just posting my experimental data with RTX 4090 and mfaktc in terms of power consumption vs performance. Power level is easily adjusted on the fly with [c]nvidia-smi -pl <WATTS>[/c], so easy to experiment and get some numbers:[code]Watt GHd/day Gd/W
450 = 12526 : 27.8
400 = 12270 : 30.7
350 = 11789 : 33.7
300 = 11134 : 37.1
250 = 10366 : 41.5
200 = 8016 : 40.0
150 = 5416 : 36.1[/code]Efficiency goes up as you reduce the power cap, down to about 250W (83% performance for 55% power).

[QUOTE=James Heinrich;627227]Just posting my experimental data with RTX 4090 and mfaktc in terms of power consumption vs performance. Power level is easily adjusted on the fly with [c]nvidia-smi -pl <WATTS>[/c], so easy to experiment and get some numbers:[code]Watt GHd/day Gd/W
450 = 12526 : 27.8
400 = 12270 : 30.7
350 = 11789 : 33.7
300 = 11134 : 37.1
250 = 10366 : 41.5
200 = 8016 : 40.0
150 = 5416 : 36.1[/code]Efficiency goes up as you reduce the power cap, down to about 250W (83% performance for 55% power).[/QUOTE]

Hi, what are the key performance metrics that matter for TL performance?
Is it FP64, FP32, FP16 GFLOP count, is it CUDA core count, version?

I have 3060 Ti, undervolted and currently I am chugging ~3000 GHzD/Day with 160w power consumption.
I am considering an upgrade, just wanted to account for TL too, what are the metrics to check. I understand 4090 is the TL king (not looking at PRO cards), but why?

FP64 is what matter.
The *speed* is directly related to the core count and FP64. AFAIK, we have not encountered any bottleneck there. Halve the speed of FP64 and double the core count and your production will be the same.

[QUOTE=firejuggler;631468]FP64 is what matter.[/QUOTE]For TF it's FP[B]32[/B] ("Single Precision") GFLOPS that matters for TF.
For gpuowl etc I have low confidence in my ability to predict performance, either from FP32 or FP64 theoretical throughput, but TF always translates directly.

All the mfkatx performance numbers on [URL="https://www.mersenne.ca/mfaktc.php"]my chart[/URL] are derived from FP32 GFLOPS and a magic multiplier for architecture (generally corresponding to CUDA level for NVIDIA):[code]$TF_GFLOPS_per_GHzDayPerDay = array(
'N' => array(
10 => 0.00,
11 => 14.00,
12 => 14.00,
13 => 14.00,
20 => 3.65,
21 => 5.35,
30 => 10.75,
35 => 11.55,
37 => 11.05, // Tesla K80 -- single benchmark, note that K80 is dual-GPU model
50 => 9.00,
52 => 9.00,
60 => 9.70, // Tesla P100
61 => 7.90,
70 => 3.58, // Titan V100 -- only one benchmark so far
75 => 3.30, // RTX 20x0
80 => 2.90, // A100-SXM4
86 => 6.15, // RTX 30?0/A?000
89 => 6.35, // RTX 40x0
),
'A' => array(
1 => 11.3, // VLIW5
2 => 11.0, // VLIW4
10 => 9.3, // GCN 1.0
11 => 9.3, // GCN 1.1
12 => 9.3, // GCN 1.2
13 => 10.9, // GCN 1.3
14 => 10.9, // GCN 1.3
15 => 10.8, // GCN 1.5
20 => 13.0, // RDNA 1 (RX 5700)
30 => 11.0, // RDNA 2 (RX 6600/6700/6800/6900)
40 => 15.0, // RDNA 3 (RX 7x00)
),
'I' => array(
10 => 9.5, // Arc A380, A770
),
);[/code]

[QUOTE=James Heinrich;631470]For TF it's FP[B]32[/B] ("Single Precision") GFLOPS that matters for TF.
[/QUOTE]

Neat, thanks for information, FP32 it is!
Sorry, if I missed it before, but how I submit my 3060 Ti benchmark? Mine is averaging around 3000, while benchmark table shows 2100 average.

EDIT: Never mind, I found it and I uploaded my benchmark.

[QUOTE=Jurzal;631474]how I submit my 3060 Ti benchmark? Mine is averaging around 3000, while benchmark table shows 2100 average.[/QUOTE]That's pretty normal these days. The performance table is based on nominal "stock" clockspeeds, while in reality with decent cooling you're likely to see both Boost and any manufacturer overclock on top of that, oftentimes a significant difference. Your [URL="https://www.mersenne.ca/mfaktc.php?filter=3060+Ti"]3060 Ti[/URL], for example, has a stock clock of [URL="https://www.nvidia.com/en-us/geforce/graphics-cards/30-series/rtx-3060-3060ti/"]1410[/URL] and your submitted benchmark showed 1965, nearly 40% higher, so the theoretical 2230 GHzd/d * 1.394 = 3108 GHd/d which is in line with your reported performance.

[QUOTE=James Heinrich;631486]That's pretty normal these days. The performance table is based on nominal "stock" clockspeeds, while in reality with decent cooling you're likely to see both Boost and any manufacturer overclock on top of that, oftentimes a significant difference. Your [URL="https://www.mersenne.ca/mfaktc.php?filter=3060+Ti"]3060 Ti[/URL], for example, has a stock clock of [URL="https://www.nvidia.com/en-us/geforce/graphics-cards/30-series/rtx-3060-3060ti/"]1410[/URL] and your submitted benchmark showed 1965, nearly 40% higher, so the theoretical 2230 GHzd/d * 1.394 = 3108 GHd/d which is in line with your reported performance.[/QUOTE]

Thanks for confirming!
Nvidia cards respond very well to proper undervolting + overclocking. Can gain 40% higher performance with -20% power consumption reduction.
1965 MHz clock from base 1410 MHz is with 165W power consumption, instead of 200W default.

[QUOTE=Jurzal;631518]Thanks for confirming!
Nvidia cards respond very well to proper undervolting + overclocking. Can gain 40% higher performance with -20% power consumption reduction.
1965 MHz clock from base 1410 MHz is with 165W power consumption, instead of 200W default.[/QUOTE]

Seems strange that they would release cards like that.

[QUOTE=Jurzal;631518]
Nvidia cards respond very well to proper undervolting + overclocking. Can gain 40% higher performance with -20% power consumption reduction.
1965 MHz clock from base 1410 MHz is with 165W power consumption, instead of 200W default.[/QUOTE]

BUT does it compute correctly at that overclock+undervolt?

In my experience, expecially with TF, it's very easy to overlook wrong compute. You simply don't find factors, and there is no other indication that the GPU is not working correctly.

So if your GPU undervolts+overclocks fantastically, you should spend a significant effort making sure it still works correctly before jumping into serious TF. To check you need to run known-factors TF and verify that the factors are all detected correctly without exception.

[QUOTE=preda;631570]To check you need to run known-factors TF and verify that the factors are all detected correctly without exception.[/QUOTE]The easiest way to do this is [c]mfaktc -st2[/c] which will test a large number of known factors of all sizes across multiple kernels and different exponent sizes, and give you confirmation at the end:[code]Selftest statistics
number of tests 26192
successfull tests 26192

kernel | success | fail
-------------------+---------+-------
UNKNOWN kernel | 0 | 0
71bit_mul24 | 2586 | 0
75bit_mul32 | 2682 | 0
95bit_mul32 | 2867 | 0
barrett76_mul32 | 1096 | 0
barrett77_mul32 | 1114 | 0
barrett79_mul32 | 1153 | 0
barrett87_mul32 | 1066 | 0
barrett88_mul32 | 1069 | 0
barrett92_mul32 | 1084 | 0
75bit_mul32_gs | 2420 | 0
95bit_mul32_gs | 2597 | 0
barrett76_mul32_gs | 1079 | 0
barrett77_mul32_gs | 1096 | 0
barrett79_mul32_gs | 1130 | 0
barrett87_mul32_gs | 1044 | 0
barrett88_mul32_gs | 1047 | 0
barrett92_mul32_gs | 1062 | 0

selftest PASSED![/code]
(there is also the less-extensive [c]-st[/c] test which does the same thing, just less of it, since -st2 can easily take several hours on a slower gpu)

[Code]
Selftest statistics
number of tests 26192
successfull tests 26192

kernel | success | fail
-------------------+---------+-------
UNKNOWN kernel | 0 | 0
71bit_mul24 | 2586 | 0
75bit_mul32 | 2682 | 0
95bit_mul32 | 2867 | 0
barrett76_mul32 | 1096 | 0
barrett77_mul32 | 1114 | 0
barrett79_mul32 | 1153 | 0
barrett87_mul32 | 1066 | 0
barrett88_mul32 | 1069 | 0
barrett92_mul32 | 1084 | 0
75bit_mul32_gs | 2420 | 0
95bit_mul32_gs | 2597 | 0
barrett76_mul32_gs | 1079 | 0
barrett77_mul32_gs | 1096 | 0
barrett79_mul32_gs | 1130 | 0
barrett87_mul32_gs | 1044 | 0
barrett88_mul32_gs | 1047 | 0
barrett92_mul32_gs | 1062 | 0

selftest PASSED![/Code]

I have completed 4001 assignments, found 52 factors.
Mostly at wavefront 75-77.

Added screenshots of GPU72 config.
I will run a recently tested known factor 168785003 at 76-77 bit, to see if I match it.

[QUOTE=preda;631570]BUT does it compute correctly at that overclock+undervolt?

In my experience, expecially with TF, it's very easy to overlook wrong compute. You simply don't find factors, and there is no other indication that the GPU is not working correctly.

So if your GPU undervolts+overclocks fantastically, you should spend a significant effort making sure it still works correctly before jumping into serious TF. To check you need to run known-factors TF and verify that the factors are all detected correctly without exception.[/QUOTE]


Test succesful, factor found.
This GPU goes up to 2175 MHz with 1.081v on core, 1965 MHz with 0.925v on core, and that is a safe margin, since it can do it on 0.900v too.

I know my overclocks. Cheers.