GTX750Ti... wait a sec...

So why do some of these have 640 Cuda Cores and some have 960?
Different chip or something unlocked?
 
my rig is non overclocked 5870, 270x,270x i get 860kh from the 270xs and 404 from the 5870. power draw is 522 at the wall.

I see people claim to get 500kh out of a 270x but how stable is that? how often do you reboot or have downtime? One of my 270x refuses to even hash more then 2 minutes at 1500 memory. im sure 450-475 is feasible but id rather my stuff work 24x7 for weeks without intervention.

This has me very intrigued. If it was $149 i might order one to play with... but not $179+
 
I see people claim to get 500kh out of a 270x but how stable is that? how often do you reboot or have downtime? One of my 270x refuses to even hash more then 2 minutes at 1500 memory. im sure 450-475 is feasible but id rather my stuff work 24x7 for weeks without intervention.

14 x 270 running at 1100 core / 1250 mem, bios undervolted to 1.1v on core. Haven't touched it in weeks.
 
wait are we talking about 290 or 270xs? my 270xs need 1500mem to get over 470kh. I currently run mine at 1040/1450~ for 430kh (not saying im using the best optimized settings)

my 290s work great at 1250mem.
 
so is this card really gonna shine with maxcoin or whatever those new coins are that Nvidia is supposed to be fantastic on? I might just have to build a rig of these in hopes of those new algo's taking off. I know people are dieing to put Nvidia to actual use and its coming at some point. I just can't wait to have the superior Nvidia quality mining away. AMD has always been a PITA to deal with
 
wait are we talking about 290 or 270xs? my 270xs need 1500mem to get over 470kh. I currently run mine at 1040/1450~ for 430kh (not saying im using the best optimized settings)

my 290s work great at 1250mem.

270x are just over clocked 270. I don't have any of the 270x but my brother does. I have a feeler out to him regarding his settings and kh/s. I think he runs them similar to mine though.
 
270x are just over clocked 270. I don't have any of the 270x but my brother does. I have a feeler out to him regarding his settings and kh/s. I think he runs them similar to mine though.

I'm at 1150/1450 and only doing about 475. I think my TC is around 15k area. 475 is still really nice. I need to undervolt these cards but I haven't even had them a week yet.
 
I'm at 1150/1450 and only doing about 475. I think my TC is around 15k area. 475 is still really nice. I need to undervolt these cards but I haven't even had them a week yet.

Brother is using 1100core/1250mem, TC 21568, stock volts, 470 kh/s, 450 WU/m. This is on coinhuntr fyi.
 
Brother is using 1100core/1250mem, TC 21568, stock volts, 470 kh/s, 450 WU/m. This is on coinhuntr fyi.

I haven't played with TC much. I'll have to try upping TC and backing the RAM down.
 
I haven't played with TC much. I'll have to try upping TC and backing the RAM down.

I read most of "The Stilts" thread on the LTC forums about memory timings. Good info in general about memory clocks and timings. Higher isn't always better, as the vbios adjust timings to adjust for higher and lower clock speeds. Sometimes lower clock + tighter timings = more bandwidth.
 
Most of my 270(x) cards run from 465-495KH/s its just a matter of finding the sweet spot. It depends on the card and rig but most of mine are running 14-1500 memory and 1050-1150 on clocks. I have one XFX card that actually runs at 1030/1420 and gets 485KH/s. If I raise or lower clocks even 20Mhz, the hash goes down to 430KH/s. I think there is a definite sweet spot for most cards and its not just pushing them as fast as possible.
 
I wonder what this says about the Maxwell architecture in general. If this is merely the first implementation of this new architecture and we're already seeing these nice energy and mining numbers (beats my brother's 760 in mining, even), then I'm intrigued to see what more potent cards can do with it. But it'll be a little while before they lift the lid on the 860 and whatnot, probably.
 
An alternative to AMD cards would be nice. Since they are generally out of stock everywhere for the good ones. Or out outrageously expensive.
 
Well it will be interesting to see for sure, IMHO there is only so much squeezing they can do before power will be needed or limit of the process will be met. low leakage decent speed high leakage fast speed, I think this very much applies to power use and so forth, there is only so much they can squeeze out of a very high end part, and these are not high end they are low power draw, so looks amazing in comparison and it, but the real test will be the very high end and I honestly do not thin they will drop temps, power or increase performance by this massive margin seems very unlikely UNLESS they manage to do something else with the card to make it same power but vastly different then what we were expecting it to be and cost $2k LOL
 
GK110 is actually more efficient than GK104, however I think that is partly because 104 is bus starved more often. You see it with GK110 when overclocking aswell, just ask Face2Face about spinning your wheels increasing power draw for little return when running a 780 at 1400 core but only 1650 mem.

Still has me interested, especially if the bus will choke the 750 Ti when it comes to core OC. Seeing many cards hitting 1400+ but not backed up be amazing vram OCs thus resulting in pretty meh core scaling in games.

I went and flashed my 780 to 1250 mem, then used offset to drop it down even more to 1125 (any lower and it will sys lock). Results are not suprising to me, even when messing with the config using T48x4 and a stiff 60C limit and quite fan curve (54% @ 60C) I'm getting 510 kh/s @ 260w at the wall which is around 1032/1125, vram actually doesn't help at all and will even hinder kh/s after a certain point due to my temp/fan limited setup.

Not the kh/w of my lovely 7950s but I'm not under volting it just yet.
 
Last edited:
not bad not bad. I Get 700 kh/s with my 780 lightning @ 375 watt.

3 of those 750ti's would do around 750 kh/s @ 280 watts. Call me impressed. Cant wait for big die maxwell.

Of course AMD Still would be the best bet for mining over 750ti's. 1 290x can go close to 1000 kh/s At around 280-300watt.
 
If you dont mind me asking, what's your config?

Tried something different and got up to 560 kh/s @ 265w, its not great but it isn't awful either.

I had a 750 Ti FTW in my cart but decided to wait for some results.
 
If you dont mind me asking, what's your config?

Tried something different and got up to 560 kh/s @ 265w, its not great but it isn't awful either.

I had a 750 Ti FTW in my cart but decided to wait for some results.

latest version of cudaminer 2-9-14

cudaminer.exe -i 0 -C 1 -H 1 -m 1 -l T12x32

usually mine at 1411/6708
 
ExtremeTech benched the new card

www.extremetech.com/gaming/176785-n...per-efficient-quiet-a-serious-threat-to-amd/3

121w gaming (apparently it's the whole system?)

Hashrates:
Litecoin.png


LTC mining efficiency is equal with the 270, not better. (decent on it's own, not ground breaking since this is AMDs old arch)
LiteCoinEfficiency.png


The card Nvidia shipped us had a hard-coded 135MHz overclocking limit, but we had no problem hitting that speed with another 200MHz on memory as well. Power consumption remains rock-bottom.

Is that just a sample, or are all of them neutered like that? I assume nv neutered the volt modding anyway.
 
Last edited:
As if taking total system power for 1 card is anywhere near realistic for mining scenario.
 
As if taking total system power for 1 card is anywhere near realistic for mining scenario.

That's what I was wondering thus the very clear question about it. I didn't see it near the chart.

I agree the CPU uses less power mining but the GPUs use more than gaming in some games (at least 290x crossfire in BF4 uses less than mining in my experience).
 
As tested by Tomshardware, Nvidia's Kepler and Maxwell TDP is far more strict than AMD/ATi's for GCN.

One would need to mod the card's artificial TDP limit somehow to increase the power draw further.

07-Power-Consumption-Peak.png
 
Last edited:
As if taking total system power for 1 card is anywhere near realistic for mining scenario.

More interested in seeing what happens with optimizations and settings played with.

Obviously with a fixed platform power draw the 750 Ti has a disadvantage, but I can do 430 kh/s at 265w or 508 kh/s at 265w or 560 kh/s at 265w just by changing a few settings in cudaminer. Point here is obviously even disadvantaged by platform power I would imagine the 750 Ti will see a decent improvement on its kh/w.
 
More interested in seeing what happens with optimizations and settings played with.

Obviously with a fixed platform power draw the 750 Ti has a disadvantage, but I can do 430 kh/s at 265w or 508 kh/s at 265w or 560 kh/s at 265w just by changing a few settings in cudaminer. Point here is obviously even disadvantaged by platform power I would imagine the 750 Ti will see a decent improvement on its kh/w.

One needs to test the maximum recognized number of cards Nvidia's driver allows as well.

Most of AMD/ATi's drivers only allow 4 (with some problems) and up to 6 on a select few drivers.

I have no idea how much Nvidia's drivers allow.
 
More interested in seeing what happens with optimizations and settings played with.

Obviously with a fixed platform power draw the 750 Ti has a disadvantage, but I can do 430 kh/s at 265w or 508 kh/s at 265w or 560 kh/s at 265w just by changing a few settings in cudaminer. Point here is obviously even disadvantaged by platform power I would imagine the 750 Ti will see a decent improvement on its kh/w.

Good points. In the end I'm not sure about the interest in the 750 ti, other than it being NV's first decent and acceptable khash/w ratio. It's so weak that in the end most people are probably going for something faster.

What it will come down to is general hashrates that can be achieved (not golden cards) on a stand off (for the informed miner).

In the end I just hope it means 20nm will not be as far apart as past generations. Now we've seen a little taste of maxwell hashing, it will be fun to see hawaii's successor too (whatever it's called).

If NV does come close to the hashing of AMD, (provided crypto currencies don't crash) gamers are going to get squeezed and priced out even further.
 
Good points. In the end I'm not sure about the interest in the 750 ti, other than it being NV's first decent and acceptable khash/w ratio. It's so weak that in the end most people are probably going for something faster.

What it will come down to is general hashrates that can be achieved (not golden cards) on a stand off (for the informed miner).

In the end I just hope it means 20nm will not be as far apart as past generations. Now we've seen a little taste of maxwell hashing, it will be fun to see hawaii's successor too (whatever it's called).

If NV does come close to the hashing of AMD, (provided crypto currencies don't crash) gamers are going to get squeezed and priced out even further.

Nvidia's (as in Nvidia + AIB) capacity to produce cards is 2-3 times that of AMD/ATi at a minimum, with likely being higher than that if we are just counting high end skus.

I have a feeling that AMD/ATi and Nvidia combined can absorb all of the crypto demand without much markup.

Crypto hardware demand is directly regulated by aggregate difficulty of all the coins in relation to their costs and prices.

Even with just AMD/ATi's 7870 and above, most of the demand has already been met.
 
usually mine at 1411/6708

Wow, those are some redonkulous clocks. Mine hits a wall at about 1250 core. What are your temps like? Curious how good that H55 is at cooling.

I get "exceeds warp kernel limit" with 12x32.

Same here.

Good points. In the end I'm not sure about the interest in the 750 ti, other than it being NV's first decent and acceptable khash/w ratio. It's so weak that in the end most people are probably going for something faster.

What it will come down to is general hashrates that can be achieved (not golden cards) on a stand off (for the informed miner).

It'll be interesting too see if more performance can be squeezed out of Maxwell. Higher core cards are going to be interesting.
 
The chance that 750 Ti is bandwidth limit in scrypt mining is high.

Minimum bandwidth needed for Hawaii at near ideal timings is 512-bit wide ~1375 mhz for 950 kh/s.

That's ~352 GB/s for 950 kh/s
That's ~2.9 kh/s for 1 GB/s

Assuming 128-bit wide 1500 mhz is ~96 GB/s
750Ti @ 128-bit wide 1500 mhz
2.9 x 96 = ~278.4 kh/s

I doubt there's much more that can be squeezed out with that memory interface.

Assuming there's a 192-bit wide 1750 mhz Maxwell coming out, it's theoretical limit would be ~ 487.2 kh/s
 
635 kh/s @ 230 GB/s

735 kh/s @ 326 GB/s

mining3_zps376add00.png~original



42% more bandwidth ~ 16% hash increase.

I imagine I would need to test 267 GB/s to see if it gets the same hash as 320 GB/s and thus it stopped scaling after no longer needed the additional bandwidth.

And here I didn't think vram made much of a difference...

Same memory increase from 1058 core netted a giant 8 kh/s increase!

j7fiJKO.png


mining4_zpsab9c1ea2.png~original
 
Last edited:
230 x 2.9 = 667

I'm talking about theoretical limits, not guaranteed minimums.

780 hashes slowly because it hashes slowly, bandwidth can only be a bottleneck, not some sort of amplifier.
 
nvidia-geforce-gtx-750-ti-maxwell-oc-mining.jpg


We have already gotten about 265 KH/s with the default settings of the card, but overclocking it has increased the performance up to almost 300 KH/s whill still in the 60W power consumption (maintained by the power limiter of the graphics card

I'm really itching to get the FTW model to see what it can do.
 
108.288 x 2.9 = 314.0352

You should just custom order cards with 1750 mhz GDDR5 :p

112 x 2.9 = 324.8

2000 mhz GDDR5

128 x 2.9 = 371.2
 
Waiting for the 750 Ti Classified King Pin Edition with 1900MHz Samsung vram.

Hopefully your math gets destroyed though, even still 300 kh/s at ~60w is crazy efficiency.
 
I'm just pessimistic we'll get a high enough memory bandwidth Maxwell for scrypt mining until the 320+bit 1750 mhz ones come out (Probably GM204) for those who need the density.
Higher density has it's own power efficiency benefits, as the other board power is basically flat from 0-6 cards.

Higher density also allows for water cooling with less overhead, which is useful if you can pipe the water to a radiator that's outside to save on A/C power costs.
 
Back
Top