(rumor) NVidia GTX 780

A

AndyE

Limp Gawd
Joined
May 13, 2013
Messages
276
As of now just in the rumor department.
http://wccftech.com/nvidia-geforce-...etail-images-leaked-specifications-confirmed/

But if true, the potential announcment of a GTX 780 with these specifications will likely further the GPU "agenda" in folding.

A quick back of the enevelope calc:
1) performance: With 13 instead of 14 SMX of the GTX Titan, performance could easily be 13/14 of the Titan. With current 7663 WU this would mean 130k PPDF
2) system cost. With a rumored 500 Euro MSRP (prices in US might be in the 5-600$ range. Add for a 2 card setup 300$ for the other components, 1500$ will get you a 260k PPD system
3) Energy consumption. If the GTX Titan is an indication, the above mentioned 2 card setup will likely be under 400watt on the wall outlet. Delivering then 650 PPD/watt.

interesting times are coming,
Andy
 
Last edited:
I've heard some scary prices on the cards, hopefully they turn out not to be true. Would hate to see re-branded GTX 670/680's as 760/770's with almost no price drop and the $150-200 increase on the GTX 780 people are talking about. Still, without doing the math it could work itself out to be a good bang for the buck PPD wise when you compare them to the $1K cards.

Can't wait for Maxwell though. I just want the 7xx Series to just be old news now. Anyone else expecting great Folding news with Maxwell?
 
The spec leaks* for the new AMD gpus would have then twice as powerfull as the curent 7000 series but on a 20nm die so they would use less electrisity and produce less heat.

Both AMD and Nvidia are saying that the next generation of GPUs are going to be good and that they will bring back the gpu wars.

Fun times ahead.


*not verifed other then Toms Hardware saying they seam legit.
 
I think that nvidia is releasing these cards because Maxwel is possibly delayed
 
Solution for a problem that doesn't exist at $600+.

7970's are under $400 market, so it needs to seriously outperform the AMD.
 
Over the years there have been different "agenda's" with FAH. First it was GPU1, then GPU2 on NVidia, then SMP, then Bigadv, then bigadv got nerfed due to the perceived points disparity between those WU and the rest. Then there was the big A4 push last year and now we are back at GPU due to the QRB being implemented. Next year who knows.....

I stay out of the turf war's now, I fold with 2p machines and that's it .I may moan occasionally but I can't be arsed any more
 
isnt this a repost os someone else post under a different title? either which way... haha, i'd for sure take the 780 off the bat, if it were 100$ cheaper :/ lol always a 100$ off my price range.
 
My 7970 atm is making any where from 100-120k ppd on core 17. With the prices that have been floating around for those GPU's, I dont think I'll be upgrading any time soon...
 
Ordered one. Can pick it up on monday. Curious about its folding perf vs. Titan.
Wouldn't be suprised, if on par or even better (due to better OC-ability).
 
I love you AndyE haha, if something sounds OK you just go buy it so the rest of us can see from your experiments.

Keep us updated... I was bouncing back and forward between buying a Titan today but if the 780 is just as good might as well grab that.
 
Ordered one. Can pick it up on monday. Curious about its folding perf vs. Titan.
Wouldn't be suprised, if on par or even better (due to better OC-ability).
Click to expand...

Titan OCs just fine, it's just a matter of cooling (since gpu boost 2.0 keys everything off temperature).
 
Quisarious said:
Titan OCs just fine, it's just a matter of cooling (since gpu boost 2.0 keys everything off temperature).
Click to expand...
I know, I did some OC with mine.
But first infos in [XS] referred to 1300/7300 MHz with the GTX 780 which is beyond the capabilities of the Titan (at least mine). I have Asus GTX Titan's.

Another pointer to this indication is Anandtech's 780 review with FAHBench.

The performance difference 780/Titan seems to be quite small
http://anandtech.com/show/6973/nvidia-geforce-gtx-780-review/18

Andy
 
AndyE said:
I know, I did some OC with mine.
But first infos in [XS] referred to 1300/7300 MHz with the GTX 780 which is beyond the capabilities of the Titan (at least mine). I have Asus GTX Titan's.

Another pointer to this indication is Anandtech's 780 review with FAHBench.

The performance difference 780/Titan seems to be quite small
http://anandtech.com/show/6973/nvidia-geforce-gtx-780-review/18

Andy
Click to expand...

If you read the thread, they comment that it was boosting to a max of 1202. That's the same that most Titans will hit. Clock speed is limited by temperature, and a hard limit on power draw. Since the 780 shares the same PCB, and has the same limits, as Titan, I'm guessing without a modified power delivery system 1202-1214 will be the practical limit for both (that's where titan runs in to power-limit draw). I can set my titan to 1350 no problem, but it won't boost any higher than 1214. At 1214 it's over the power limit, so it intermittantly drops back into the 1100's, and net result is slower than if I set at 1202, which is right at the power limit and therefore runs constant at that speed. I suppose with less shaders it should be able to go a little higher, but that thread (as early as it is in the process), would seem to indicate otherwise.
 
I am currently in China - many websites are blocked.

There seems to be 2 models available from EVGA - stock and superclocked. Can't access the EVGA website to check, If the 80 MHz higher base frequency is preserved on the top as well.
Local websites refer for the superclocked version a TDP of >250W.

The 780 has 1/7 less shaders. How much this will impact the Boost 2.0 algorithm with the FAH workload? Don't know yet, I'll look into this on monday when back home.

If there is interest, I can share results.

Andy
 
I was at my local PC shop buying some Blank DVDs and decided to buy a 780GTX for folding as well.

Currently have it boosted to 1175Mhz, beyond that it intermittently throttles down to stock clocks which slows things down more than it's worth.

Number of Frames Observed: 13

Min. Time / Frame : 00:01:07 - 150,483.9 PPD
Avg. Time / Frame : 00:01:09 - 143,988.8 PPD
Cur. Time / Frame : 00:01:07 - 149,037.8 PPD
R3F. Time / Frame : 00:01:08 - 146,035.9 PPD
All Time / Frame : 00:01:09 - 143,134.1 PPD
Eff. Time / Frame : 00:01:12 - 134,984.1 PPD
Click to expand...

I'll let it run for 12 hours and see if anything changes.
 
Last edited:
That's nice, little faster than I would of guessed. Need to cool that puppy somehow, see what it can really do =).
 
Thanassos said:
I was at my local PC shop buying some Blank DVDs and decided to buy a 780GTX for folding as well.
:D What a sales guy, good upselling skills....

Currently have it boosted to 1175Mhz, beyond that it intermittently throttles down to stock clocks which slows things down more than it's worth.
I'll let it run for 12 hours and see if anything changes.
Click to expand...
Thanks for sharing your data points. I just arrived back in Austria and will pick up mine tommorrow.

Did you buy a stock or OC version of the 780?
Click to expand...
Click to expand...
 
AndyE said:
Did you buy a stock or OC version of the 780?
Click to expand...

I got the only one in stock, a reference Gigabyte Model.

Zagen30 said:
Thanassos, any idea of what the power draw is at that frequency?
Click to expand...

Well, I'd say about 215W ~

Used to just have the 690 + 3930k going drawing 600watts ~ now I'm sitting at 808 +/- 15.

As a side note, left it going overnight but with SMP off, the 780 dropped down to 1:06 / frame @ 153kppd, even with SMP 4 the 3930k still seems to fail at "feeding" the 780/690 well enough, with SMP 4-10 on the lowest the GPU gets to is 1:09 effective a frame with SMP off altogether it drops down to 1:06/1:07. So in my system at least I've gained performance by turning off the SMP :/
 
Last edited:
So you're getting 150K on the 780 :eek:???

I thought the Titan was 120K and up to 140K max with good OC's or has Core_17 been tweaked even more beneath our noses?
 
140k Seems a little low... my 690s are getting 80kppd per core @ 1215 Mhz.

Pretty sure AndyE has gotten his TITANs to 160k ~ which would make sense, the only difference is 13/14 SMX Units unless I'm missing something.
 
I was the one reporting initially 120/140k for the Titan, while others like Quisarious reported 150k+ numbers. With some decent OC to keep temps under control, my Titans are now processing 7663 WUs with TPFs of 1m6s which is slightly over 150k.

BTW, the difference in SMX is 1/7, not 1/14. One SMX is 192 shaders.
Which means that this part of the GK110 chip will use proportionallly less power (with proper power gating), giving more headroom with temp and TDP. This was my initial assumption when the specs started to "materialize". From a cost perspective and the current numbers Thanassos report, the GTX 780 seems to be the better folding card - no doubt.
This might change, when FAH is moving from single precision to mixed/double precision calculation like AMBER is doing, but I haven't seen any indication that this will happen. Those with more insight - please jump in.

Andy
 
You're not wrong though AndyE especially in Australia the 780 cost me $830 where as he TITANS are selling for $1300.... Although I'm thinking that as OpenCL support gets better AMD might pull ahead quite considerably just like they are in all the other OpenCL things.

Here is the HFM readout after going overnight. Ignore the failed units they happened while toying around with kboost in precision. Top is 780, two below are the 690 Cores.

Cheers,

BjLrMWq.jpg
 
Last edited:
Finally I had been able to pick up my GTX 780 (the stock version, not the OC).
To better compare the total system impact I took two :)

I have 2 identical systems with two AMD 7970 each. CPU is a Celeron G1610, sufficiently powerful to drive the 2 cards but it doesn't do any folding. With 2 cards at 1150/1600MHz, each system can produce 200k PPD of WU 7663. Power consumption is 380 watt at the plug.


I took one system and replaced the 2 AMD cards with 2 GTX 780 cards - the rest of the system was kept unchanged.

Currently the first WU is running:

With stock settings, TPF is 1min22sec (111k PPD per GPU)
Power consumption is 405 watt. The difference comes mostly from the Celeron CPU, as the NVidia driver keeps the CPU running at 100%. The ATI driver is much better (CPU is at 18% plus running at reduced clock)

Changing the settings to 1175 MHz (+180MHz for the shader), +160 MHZ for memory, and TDP to 104%.
TPFs seem to settle in the 1min9sec range with this WU. Two GTX 780 are then producing 288k PPD. Power consumption is 430 watt
With the fan set to 90% (There is a difference in the available settings to the Titan which is limited to 85%), temperature of GPU-0 is 58 degrees Celcius and GPU-1 is 50 degrees. A 200mm fan in the side panel is getting ambient air (currently 20 degrees) to the GPUs.

Comparing it with the GTX Titans (1090 MHz, +100 MHz memory, 104% TDP):
2 GPUs are constantly between 300 and 310k PPD (TPF between 1:05 and 1:06)

Andy
 
AndyE said:
Finally I had been able to pick up my GTX 780 (the stock version, not the OC).
To better compare the total system impact I took two :)

I have 2 identical systems with two AMD 7970 each. CPU is a Celeron G1610, sufficiently powerful to drive the 2 cards but it doesn't do any folding. With 2 cards at 1150/1600MHz, each system can produce 200k PPD of WU 7663. Power consumption is 380 watt at the plug.


I took one system and replaced the 2 AMD cards with 2 GTX 780 cards - the rest of the system was kept unchanged.

Currently the first WU is running:

With stock settings, TPF is 1min22sec (111k PPD per GPU)
Power consumption is 405 watt. The difference comes mostly from the Celeron CPU, as the NVidia driver keeps the CPU running at 100%. The ATI driver is much better (CPU is at 18% plus running at reduced clock)

Changing the settings to 1175 MHz (+180MHz for the shader), +160 MHZ for memory, and TDP to 104%.
TPFs seem to settle in the 1min9sec range with this WU. Two GTX 780 are then producing 288k PPD. Power consumption is 430 watt
With the fan set to 90% (There is a difference in the available settings to the Titan which is limited to 85%), temperature of GPU-0 is 58 degrees Celcius and GPU-1 is 50 degrees. A 200mm fan in the side panel is getting ambient air (currently 20 degrees) to the GPUs.

Comparing it with the GTX Titans (1090 MHz, +100 MHz memory, 104% TDP):
2 GPUs are constantly between 300 and 310k PPD (TPF between 1:05 and 1:06)

Andy
Click to expand...

Can you see how high it will go? Plenty of thermal headroom there. I'd also consider dropping the memory OC, doesn't help for folding, and adds heat/stability issues.
 
Quisarious said:
Can you see how high it will go? Plenty of thermal headroom there. I'd also consider dropping the memory OC, doesn't help for folding, and adds heat/stability issues.
Click to expand...
Didn't do too much testing.

Some observations.
(removed the memory boost)

The 1175 MHz with my cards are solid - no visible transient throtteling. Looks like "start and forget" mode. BTW, my cards finished the first WU with 280k PPD (together).

Moving to 1202 MHz showed initial and infrequent drops with one card (but the card recovered back to 1202 MHz). No issue with the second card.
Moving to 1300 MHz: One card could sustain for more than 10 secs, the second card did almost finish one timeframe (1 min). But eventually, both reported errors

What I haven't check yet is, if my cards have 4 or 5 GPC clusters and if this difference in the GTX 780s shipped makes any difference in this regard.
 
update:

Given the good experience with the GTX 780 earlier today, I've changed a few things with my Titan's.

1) Dared to change the shader speed in one existing system also to 1175 MHz for the 2 GPUs. TDP continues to be on 104%.
TPF is now down to 1min 1sec - 1min 2sec, resulting in >330k PPD
Temperature is now 60 and 56 degrees respectively

2) Took the 2x AMD 7970 in my second system out and replaced it with 2 Titan's. Same settings and results as above.


Currently, 4x Titans and 2x 780s are at 940k PPD in total - will run it in this config for a while.
Need to get 2 cases for the 4x AMD 7970 which are now unused.

Andy
 
AndyE are the Titans in a system with the Celeron?

In my experience due to the CPU Intensive workload of Core17 on nVidia there's benefit in OCing your CPU a little higher. Even on my 3930k @ 3.2Ghz with the 780 @ 1175 (No Mem OC) it won't go below 1:09-1:10. With a bump up to 4.0 It drops to 1:06-1:07. Not a huge gain but that seems to reflect closer to your Titans than your 780s so maybe your Celeron is gimping your 780s?

Ultimately we want stability over pure performance otherwise it's bad for the project but I'm keen to push it for curiosity, does the Memory speed seem to have a meaningful impact at all in your toying around?
 
Both systems are identically built with Celeron G1610 CPUs. So the 1:01 TPF of the Titan is with a stock Celeron doing nothing else than feeding (and polling) the 2 GPUs.

In other workloads the memory speed had an impact, but here it seems that FAHCore is compute bound. Took the advice of Quisarious and moved the mem slider back to normal. No impact on TPF with FAHcore17
 
Alright great thanks for that. Slightly off-topic but given their current performance and my financial position at the moment I'd be looking at getting another 780 or 2 within the next week pending stock.

My question to you would be cost and power wise, it would be cheaper to build 2 systems with the Celeron and 2 GPUS instead of building 1 System with 4 GPUS correct? As to feed the GPUs I'd need to upscale to Dual Core + HT or Quad Core.

I'm only asking you because you seem to have a bunch of hardware to play around with mate. (I haven't considering nor do I even know what AMD has on offer.)
 
Thanassos,
I've tried 3 different system configs with 4 GPUs.

1) Dual Xeon system with Asus Z9PE-D8 motherboard (worked flawless)
2) Síngle LGA-2011 system (Asus X79WS) (worked flawless)
3) Single LGA-1155 system (Asus Z77WS) (does not work, too many errors to be useful)

Based on the experiences in the last few weeks, I am currently settling around 2 GPU systems. Much more stable, cheaper, more efficient, better temperature. While the Titan and 780 can be put into adjacent PCI slots, the card with the blower covered by the back of the second card is usually 10 degrees hotter. As my AMD 7970 collapsed when covered like that, I started to use motherboards with 2 PCI slots farther apart. For these systems, I use a simple, cheap Celeron G1610 just to drive the GPUs, no folding on the CPU.

When you use a dual PCI motherboard, take care how the second PCI slot is connected to the CPU - there are multiple options available.

I just ordered three MSI Z68MA-G45 (G3) motherboards. They are cheap (60 Euro each), and have all things I need.
CPU: As said, Celeron G1610 (40 Euro)
Mem: 2x 2GB, or 2x 4GB should be ok. (30 or 50 Euro)
PSU: I use "be quiet!", with 730 watt. With a load of between 4-500 watt, they are not stressed and stay silent. There are many PSU available, just pick one with support for 2 GPUs.
Case: any miditower with good ventilation will do it. I keep the sidepanel open and flip the case 90 degrees so that hot air can ascend. Mine were 40 Euro.
I found recently the Coolermaster HAF BX, got 3 cases today. I like the positioning and airflow (to directly feed cold air into the GPU). They are more expensive, but I think they are better than normal cases
Vents: Whenever space is available, I feed cold air directly into the GPUs (one or two smaller fans). One additional 200mm is placed over the GPUs to either blow cold air on the whole case (NVidia), or suck the hot air away (AMD 7970). It is just a matter of flipping the fan.
The Coolermaster HAF BX comes very close to an ideal case for this application, but it is 90 USD. I'll report back, when I have all my systems transitioned.

In the end, the system stability is what "drives" me more and more towards a set of 2 GPU systems. Originally, I wanted to get into the 4x GPU bandwagon, but the simplicity and low overhead of a 2GPU setup made me switch.

Andy
 
AndyE, I totally agree with your conclusions. I don't go over 3x GPU in a single box and that is on a MB that has extra spacing between GPU PCI-E slots so there is one slot between GPUs. I run with the side off and a 120mm fan screwed to the side panel rail extracting hot air. I have another 120 in the front feeding in cold air. This is with GTX 570s.

In the past I had run 2x 9800GX2 (covers removed) and a GTX250 in a single box. That box was always pretty hot but it ran like that a good long time. I would have problems if I let it *cool* too long before starting it back up. I guess once they got used to running hot, they didn't know any different.
 
Sorry to Nercro the thread but I think I have an issue with the 3rd 780 GTX I just picked up. Out of the box placed it on +200 Mhz like my other 2 and it worked great through out the night.

On the particular WU is is crunching at the moment it goes as normal with a TPF of 3:00~ every 10-15% though it takes nearly 20 minutes to complete a frame then back to normal. There's no Bad State Detected warnings, no drop in clocks or GPU power utilization?

Anyone else seen this behavior? Worth returning the card?
 
You must log in or register to reply here.
Back
Top