Top boost bins are stabilized with the better cooling. But yes usually power wall triggers first. You could always flash a custom bios with higher powef limit.
Nvidia BigK GK110 Kepler Speculation Thread
- Thread starter A1xLLcqAgt0qc2RyMz0y
- Start date
-
- Tags
- nvidia
Top boost bins are stabilized with the better cooling. But yes usually power wall triggers first. You could always flash a custom bios with higher powef limit.
gamervivek
Regular
techreport(skyrim) show it drawing more than 290x just like hexus(far cry 3), in both there is little difference between uber and quiet modes.
http://techreport.com/review/25611/nvidia-geforce-gtx-780-ti-graphics-card-reviewed/11
curiously the only game which shows a substantial difference in favor of 780Ti, besides guild wars 2, is battlefield 4 while other games are barely faster, which kinda inverts the normal review.
http://techreport.com/review/25611/nvidia-geforce-gtx-780-ti-graphics-card-reviewed/11
curiously the only game which shows a substantial difference in favor of 780Ti, besides guild wars 2, is battlefield 4 while other games are barely faster, which kinda inverts the normal review.
sheepdogexpress
Newcomer
Are you kidding? Non-reference designed gtx 780's probably had the best response and furthest headroom as far as partner cards go.
http://www.hardocp.com/article/2013/08/14/galaxy_geforce_gtx_780_hof_edition_review/3#.Unx5RuKG-Sc
The gtx lightning did even better and the classified reached similar heights.
On top of this there's a 20 percent difference between some models in terms of performance between reference and none reference.
AMD 290x hit the wall where they start needing water much earlier than gtx 780 ti. They scale up power consumption much quicker than the gtx 780 ti or gk110 in general.
A non reference gtx 780 ti that given a boost clock of 1100mhz is likely going to boost to 1200 in game, with overclocking room past 1300 and maybe near the 1400mhz region with the right bios.
I was referring to stock clocks. NVIDIA's partners might release 1100MHz cards, which would be tough on the reference cooler (too much throttling to be worth it).
No I wasn't kidding. I was looking at the wrong thing though. While max overclocks don't look much different from what's possible on the stock cooler actual performance is higher due to less thermal throttling.
http://www.techpowerup.com/mobile/reviews/EVGA/GTX_780_SC_ACX_Cooler/29.html
DuckThor Evil
Legend
I got myself two Asus DCII 780 cards for a decent price of 459€ a piece that I'll be putting under water soonish with a 4770k that I'm planning on delidding. The cards luckily perform quite equally to each other. Results so far with the second card.
3dmark Firestrike
Firestrike Extreme
Advertised base and boost clocks are 889/941Mhz, at stock settings the cards actually boost to 1032/1045. Lifting the voltage marginally with software lifts those boost clocks by 2 boost bins (26Mhz). Overclocked the slightly better card completes the new 3Dmark with 1293Mhz boost clock (1280Mhz in 3dMark11) The other one requires 13-26Mhz lower clocks. Memory clocks around or little past 7300Mhz with both cards.
I'm slightly tempted to get an Asus Extreme motherboard to be able to VGA hotwire them and increase the voltage, but I probably won't do that now. The overclocked performance is pretty good already imo.
3dmark Firestrike
Firestrike Extreme
Advertised base and boost clocks are 889/941Mhz, at stock settings the cards actually boost to 1032/1045. Lifting the voltage marginally with software lifts those boost clocks by 2 boost bins (26Mhz). Overclocked the slightly better card completes the new 3Dmark with 1293Mhz boost clock (1280Mhz in 3dMark11) The other one requires 13-26Mhz lower clocks. Memory clocks around or little past 7300Mhz with both cards.
I'm slightly tempted to get an Asus Extreme motherboard to be able to VGA hotwire them and increase the voltage, but I probably won't do that now. The overclocked performance is pretty good already imo.
I got myself two Asus DCII 780 cards for a decent price of 459€ a piece that I'll be putting under water soonish with a 4770k that I'm planning on delidding. The cards luckily perform quite equally to each other. Results so far with the second card.
3dmark Firestrike
Firestrike Extreme
Advertised base and boost clocks are 889/941Mhz, at stock settings the cards actually boost to 1032/1045. Lifting the voltage marginally with software lifts those boost clocks by 2 boost bins (26Mhz). Overclocked the slightly better card completes the new 3Dmark with 1293Mhz boost clock (1280Mhz in 3dMark11) The other one requires 13-26Mhz lower clocks. Memory clocks around or little past 7300Mhz with both cards.
I'm slightly tempted to get an Asus Extreme motherboard to be able to VGA hotwire them and increase the voltage, but I probably won't do that now. The overclocked performance is pretty good already imo.
That doesn't seem very decent when you could have bought a pair of R9 290s for ~€350 each; especially if you're going to water-cool them anyway.
I believe the card's bios has a table of which voltages are required for which clocks and this is what drives the boost functionality. What I don't get is why manual overclocking squeezes so much more out of the card than boost. Is there a hard limit to the range of voltage bins that boost uses?
E.g. from AT's review it's clear that the max voltage doesn't increase with overclocking yet the max boost clock does. Why wouldn't it hit that max boost clock anyway at stock?
http://www.anandtech.com/show/7492/the-geforce-gtx-780-ti-review/16
DuckThor Evil
Legend
Few things... First of all I said decent, not great. In any case these were 200€ a piece more expensive just few days prior to me ordering them, 290 vanilla pricing wasn't announced yet, but they are more like 399€ in Finland currently with a reference board and I got 3 games per card with these. nVidia has some unique features either today or upcoming that I value and I believe these will hold their own against 290s. I also have a separate AMD rig with a 7970 Matrix Platinum and it is not really needing an upgrade at the moment imo, it's future is undecided for now.
I believe the card's bios has a table of which voltages are required for which clocks and this is what drives the boost functionality. What I don't get is why manual overclocking squeezes so much more out of the card than boost. Is there a hard limit to the range of voltage bins that boost uses?
E.g. from AT's review it's clear that the max voltage doesn't increase with overclocking yet the max boost clock does. Why wouldn't it hit that max boost clock anyway at stock?
I think it's mostly just leaving OC headroom, playing it safe and usually you also lift the power and temperature target when overclocking. As you can see from the Anandtech review the power consumption does go up quite a bit with the OC, even if the voltages don't, that's more noise too.
Same here. At equivalent price and performance nVidia gets the nod. Also depending on your personal financial situation price can become less of a factor.
That makes sense. Don't think I can live with a card that's not running some sort of 3rd party cooling. The noise/temp benefits are too great, even compared to nVidia's decent stock cooling.
DuckThor Evil
Legend
yeah I have to say I was quite impressed or even stunned yesterday when my friend installed a GTX 770 MSI gaming OC card with the Twin Frozr cooler. It performed very nicely and we tried 100% fan setting and the sound level was far lower than I could have imagined. At auto settings and the card running close to 1300Mhz it was barely audible (over case fans).
TDP limitation ( and temp limitation ).. when they overclock the card, they too push the power limit .. thats just the reason.... At the moment you lift up the enveloppe ( temp and tdp ), the max boost clock will go as high as it can beneath this enveloppe. Its why i was not really understand why hardware.fr have test the 780Ti this way calling it "uber".. because the max boost clock will increase, providing a "manual " overclock.
Its the case on the card i have seen... but its possible they have put a limitation.. The anand article explain really well the binning problem.. you could hit the higher level of clock, but on their chip, the higher level of binning is just not available.. hence why you have some difference on the max boost clock speed ( when the tdp and temp limit are increased ofc, i dont speak about stock settings. )
It look like there's some increment ( see it as level ), so if you dont meet 2 factors ( the next increment available + voltage needed ( defacto, maybe it is an possible increase in voltage and not a level on the voltage ), the turbo will not make kick the next increment on clock speed.. it could, and this can look stable, but i suspect they have bin the chips like that, for maintain a big range of stability across different binning. ( a little bit like what we have seen with 7970 by past ).
Last edited by a moderator:
That whole grey area between allowable boost "buckets" and how to attain them / how they relate to some internal, preconceived voltage and thermal limit is what turned me off from buying into the 600 series vs the 7000 series from AMD. For better or worse, I prefer the AMD method of setting my clock, setting my voltage, and setting my power and being accountable for it working or not.
The R9 series is, IMO, getting a lot closer to how NV does it -- and I'm not necessarily a fan. I think we all know where it will end, which is the same for CPU overclocking.
The R9 series is, IMO, getting a lot closer to how NV does it -- and I'm not necessarily a fan. I think we all know where it will end, which is the same for CPU overclocking.
Personally, its not a problem for me, IF, i can still use a third party software for overclock the card the same good old way. ( set vcore, set clockspeed as i want... )
iMacmatician
Regular
NVIDIA has announced the Tesla K40:
http://www.nvidia.com/object/tesla-servers.html
It's worth noting that the table says the Tesla K40 has a GK110B chip. Assuming that leaked slide from a month or so back is legitimate, that would mean GK110B = GK180 (which I think was already speculated).
http://www.nvidia.com/object/tesla-servers.html
It's worth noting that the table says the Tesla K40 has a GK110B chip. Assuming that leaked slide from a month or so back is legitimate, that would mean GK110B = GK180 (which I think was already speculated).
Similar threads
