Author

Topic: [ANN] cudaMiner & ccMiner CUDA based mining applications [Windows/Linux/MacOSX] - page 741. (Read 3426922 times)

member
Activity: 84
Merit: 10
I can check my 780 on pci 3 x16 if you guys want those results
That would help tremendously. What kind of risers do you have? Can you try a 16x to 16x riser or even a 1x to 16x in that PCI-E 3.0 versus the card being put into the slot directly? Thanks.

I missed the riser part sorry. Would test it but it means ripping my water cooler apart :-(
Ah, no worries. Would still be curious if 3.0 slots yield any benefit versus 2.0 slots for cudaminer if there was any way you had a means of doing that without messing up the water cooling.
sr. member
Activity: 266
Merit: 250
anyone tried mining grid? is it profitable?
newbie
Activity: 43
Merit: 0
Is anyone having issues compiling on linux?

Ran autogen, then configure --with-cuda=/opt/cuda-5.5

Breaks with this at the end:
Code:
checking for pthread_create in -lpthread... yes
checking for SSL_library_init in -lssl... yes
checking for EVP_DigestFinal_ex in -lcrypto... yes
./configure: line 6142: syntax error near unexpected token `,'
./configure: line 6142: `LIBCURL_CHECK_CONFIG(, 7.15.2, ,'

I thought maybe curl wasn't installed, and actually it wasn't, but after installing it I got the same error even after rebooting.

This is pulled from the git, with the latest release having been only 10 hours ago when I pulled it last night Smiley Went and checked around for any extra/unmatched ` , ; ' etc but didn't find any Sad  I am not really a bash scripter

This is debian wheezy 64bit by the way.
sr. member
Activity: 350
Merit: 250
I can check my 780 on pci 3 x16 if you guys want those results
That would help tremendously. What kind of risers do you have? Can you try a 16x to 16x riser or even a 1x to 16x in that PCI-E 3.0 versus the card being put into the slot directly? Thanks.

I missed the riser part sorry. Would test it but it means ripping my water cooler apart :-(
legendary
Activity: 1154
Merit: 1001
The bandwidth will also vary depending on the number of installed card.
My motherboard (Z77 G45), will allow one card to run at 16x bandwidth (pcie 3)
but 2 cards will run at 8x bandwidth
3 cards (using the pci-e 2.0) will make the first cards run at 8x bandwidth and the two other at 4x

What tool do you use to confirm this slow down?

I got this one my MB manual Wink
me too.
However, hwinfo64 show the actual bandwidth used by the card (gpu-z also I think)

I'm now using two 750 TI's on a motherboard with PCIe 1.0, one is sitting on a x16 slot, the other one on a x4 slot (IIRC, the equivalent of a PCIe 3.0 x1 slot), I have no bottleneck if I run all my mining with '-H 2'. I'm reaching the same speeds as everyone else here, roughly 310 to 320 kh/s scrypt or 160 mh/s kekkak with moderate overclocking. Without the '-H 2', I go straight to 250 kh/s or so in scrypt.

(hash rates quoted are for a single card, and both reach the same figures with -H 2)

Cheers,
Myagui

Edit: Will check & report on my bus usage tonight.
legendary
Activity: 1400
Merit: 1050
The bandwidth will also vary depending on the number of installed card.
My motherboard (Z77 G45), will allow one card to run at 16x bandwidth (pcie 3)
but 2 cards will run at 8x bandwidth
3 cards (using the pci-e 2.0) will make the first cards run at 8x bandwidth and the two other at 4x

What tool do you use to confirm this slow down?

I got this one my MB manual Wink
me too.
However, hwinfo64 show the actual bandwidth used by the card (gpu-z also I think)
member
Activity: 84
Merit: 10
would the 4x slots  be enough to keep stable rates on 750ti's?
I did try a 1x-16x riser inside a 4x slot. That yielded 5-6% decreased performance in hashrates, but as far as stability, it is rock solid on +135 core +600 mem overclocking.
newbie
Activity: 48
Merit: 0
would the 4x slots  be enough to keep stable rates on 750ti's?
member
Activity: 84
Merit: 10
I can check my 780 on pci 3 x16 if you guys want those results
That would help tremendously. What kind of risers do you have? Can you try a 16x to 16x riser or even a 1x to 16x in that PCI-E 3.0 versus the card being put into the slot directly? Thanks.
newbie
Activity: 59
Merit: 0
The bandwidth will also vary depending on the number of installed card.
My motherboard (Z77 G45), will allow one card to run at 16x bandwidth (pcie 3)
but 2 cards will run at 8x bandwidth
3 cards (using the pci-e 2.0) will make the first cards run at 8x bandwidth and the two other at 4x

What tool do you use to confirm this slow down?

I got this one my MB manual Wink
full member
Activity: 140
Merit: 100
The bandwidth will also vary depending on the number of installed card.
My motherboard (Z77 G45), will allow one card to run at 16x bandwidth (pcie 3)
but 2 cards will run at 8x bandwidth
3 cards (using the pci-e 2.0) will make the first cards run at 8x bandwidth and the two other at 4x

What tool do you use to confirm this slow down?
legendary
Activity: 1400
Merit: 1050
Some times ago I tried a 16x by 16x riser with the 780ti, this one was also having performance drop.
sr. member
Activity: 350
Merit: 250
I can check my 780 on pci 3 x16 if you guys want those results
legendary
Activity: 1400
Merit: 1050
The bandwidth will also vary depending on the number of installed card.
My motherboard (Z77 G45), will allow one card to run at 16x bandwidth (pcie 3)
but 2 cards will run at 8x bandwidth
3 cards (using the pci-e 2.0) will make the first cards run at 8x bandwidth and the two other at 4x
member
Activity: 84
Merit: 10
A thought in regards to the PCI-E risers on 1x to 16x.

If the bottleneck is just that 1x doesn't have enough bandwidth because CudaMiner uses lots of bandwidth through the PCI-E slot... if we got PCI-E 3.0 motherboards we would remove that bottleneck no?

PCI-E 2.0 1x bandwidth = 5GT/s
PCI-E 2.0 16x bandwidth = 80GT/s
PCI-E 3.0 1x bandwidth = 80GT/s

Has anyone with 16x to 16x risers confirmed that even those risers are causing decreased performance on these 750 Ti's? What I can confirm on 1x to 16x risers is it decreases performance by about 6%. If we can confirm that 16x to 16x yields the same performance hit then we can eliminate the 1x slot being the bottleneck.

Is it better to plug the 1x->16x speed in a pci-e 3.0 slot or a pci-e 1.0 or it doesn't matter ?

My motherboard only has PCI-E 2.0 slots so I unfortunately can't compare the difference between 1.0 vs 2.0 vs 3.0. A few pages back though Christian did mention something about Cudaminer needing bandwidth across these lanes and that the 1x slot might be what is causing the slight slowdown on risers. Have not been able to isolate if it is the 1x slot being the bottleneck or the risers themselves being the bottleneck. I have tried 2 powered risers so far and they yield the same ~6% performance hit.
legendary
Activity: 1400
Merit: 1050
A thought in regards to the PCI-E risers on 1x to 16x.

If the bottleneck is just that 1x doesn't have enough bandwidth because CudaMiner uses lots of bandwidth through the PCI-E slot... if we got PCI-E 3.0 motherboards we would remove that bottleneck no?

PCI-E 2.0 1x bandwidth = 5GT/s
PCI-E 2.0 16x bandwidth = 80GT/s
PCI-E 3.0 1x bandwidth = 80GT/s

Has anyone with 16x to 16x risers confirmed that even those risers are causing decreased performance on these 750 Ti's? What I can confirm on 1x to 16x risers is it decreases performance by about 6%. If we can confirm that 16x to 16x yields the same performance hit then we can eliminate the 1x slot being the bottleneck.

Is it better (in general) to plug the 1x->16x speed in a pci-e 3.0 slot or a pci-e 1.0 or it doesn't matter  ?
member
Activity: 84
Merit: 10
A thought in regards to the PCI-E risers on 1x to 16x.

If the bottleneck is just that 1x doesn't have enough bandwidth because CudaMiner uses lots of bandwidth through the PCI-E slot... if we got PCI-E 3.0 motherboards we would remove that bottleneck no?

PCI-E 2.0 1x bandwidth = 5GT/s
PCI-E 2.0 16x bandwidth = 80GT/s
PCI-E 3.0 1x bandwidth = 80GT/s

Has anyone with 16x to 16x risers confirmed that even those risers are causing decreased performance on these 750 Ti's? What I can confirm on 1x to 16x risers is it decreases performance by about 6%. If we can confirm that 16x to 16x yields the same performance hit then we can eliminate the 1x slot being the bottleneck.
newbie
Activity: 48
Merit: 0
Price just went up $10 today on newegg for all the ti's once there in your cart for the ones that are left..

ANYONE HAVE HASH RATES FOR THE NON TI VERSION OF THE 750??

I have two Palit 750 StormX OC devices at home, but it will be a while before I can connect them.

Christian


Great.. would be sweet to know the price/hash comparison i have scoured the google machine no luck just people with the same question hehe
newbie
Activity: 2
Merit: 0
Are you sure it runs with 1 card now? You may not have set cud a back up again properly
Yes, It's been hashing for the pass two days on one card.

Anybody else have any ideas why I keep getting "Unable to query CUDA driver version! Is an nVidia driver installed?" when I attempt to run 2 750 Ti. I'm running Ubuntu mini 13.10 and drivers - 334 and cuda is 5.5
hero member
Activity: 756
Merit: 502
Price just went up $10 today on newegg for all the ti's once there in your cart for the ones that are left..

ANYONE HAVE HASH RATES FOR THE NON TI VERSION OF THE 750??

I have two Palit 750 StormX OC devices at home, but it will be a while before I can connect them.

Christian
Jump to: