I've come across a little more material regarding the forthcoming GTX 1060 release. There are some possibly important points of contrast to the pattern of the 1080 and 1070 recent releases.
While for those cases review samples were distributed well in advance of first shipping date, and the NDA embargo on reviewers discussing their findings was a week or so before release date, in this case I think the reviewer samples went out later, and in any case the strongly rumored review embargo release time is on the same day as first shipments.
Possibly a bigger difference is in the Founders Edition matter. For the 1080 and 1070, about five of the major card makers sold FE cards boxed with their own logo, but built, so far as anyone has mentioned, identically to the Nvidia design. Near first shipment date, these completely dominated the available card supply, especially for the 1080, and distinct designs have only crept into the marketplace slowly. Also for both of these cards Nvidia took the position that this Founders Edition was something they expected to remain available throughout the life of the card type.
For the 1060 I've seen plausible claims that the "limited" FE card is only being made by Nvidia (which probably means assembled by a Chinese contract assembly house at their direction, quite possibly Hon Hai=Foxconn) and only available from them. However a range of card house cards are expected to be available for order on release day. And the "limited" FE card is slated to disappear from the market in perhaps a month.
Lastly, and of special interest for Einstein folks, I've seen multiple ways of saying that Nvidia has given the 1060 a better shake than it did the 960 as regards memory performance. This applies whether you compare within family (1060 to 1070 appears to have a much higher fraction than did 960 to 970) or relationship to previous generation (1060 from 960 nominally appears to be a big memory performance bump). I think the 960 was deemed a disappointing card for Einstein performance by more than one user, but this gives me some hope for better things this time.
So I am steadfast in my intention to buy a 1060 as soon as I can, and report trial results here. I am less committed than I was to specifically getting the Founders Edition, though I remain concerned about the character of the fan noise on the non-FE cards. But surely some of them will be OK, and I have potential places to put them that are not nearly so well supplied with cool air by the PC case as is my current 1070 arrangement.
Meanwhile, in the RX 480 power matter, one reviewer at Techpowerup has tested the new driver release and deems that it successfully shifted power consumption off the motherboard connection sufficiently, and that in compatibility mode (a defaulted off user option) it finally meets the stated total card power consumption.
i also have Nvidia 1080 and the Nvidia inspector does not work so far when you try to adjust the memory clock in P2. I'm currently at 5100Mhz which is achieved by +500 overclock. I hope the new version of Nvidia inspector recognise the memory correctly so you can adjust the speed acording in P2 as my 1080 is able to go up to 5750Mhz.
I've been able to raise my 1080's memory clock using PrecisionX OC
Are you able to adjust independently the 1080 memory clock used in the P0(gaming) state and in the P2 (distributed computing) state using PrecisionX?
Running which load are you running the memory clock at 5450? Which tool are you relying on for reporting that value?
For my 1070, the one thing I have seen to work and which I have been using has been to use Nvidia Inspector to specify an offset clock for the P0 state while the card was at the moment in the P8 state. This same offset then also affected the P2 state. A serious concern with my method is that it sets the P0 state clock rate higher than might work. I'd be very happy to be able to influence P2 clock rates (memory and core) independently of the rates for other states.
While GTX 1060 launch day and review embargo release day are supposedly both next Tuesday, July 19, I followed up on a post claiming two places were accepting pre-orders, and placed an order for a Zotac GTX 1060 card at a nominal price of just under $260 plus moderate shipping. So perhaps the claimed $250 MSRP for this one is less of a fib than the claimed MSRP for 1080, 1070, and 480.
Or maybe sabrePC will never ship against my order. Time will tell.
I've been looking at NowInStock listings for the 1080, 1070, and RX 480, and the recent pattern seems to be that GTX 1080 and RX 480 listings get cleaned out very shortly after they are posted, but that for several days now there have been multiple Green (e.g. accepting orders claimed to be against actual stock) listings for the 1070 at any given moment. On the other hand, I've not spotted any 1070s anywhere near the stated $380 MSRP. But if availability stays good, I'd expect transaction prices to come down a little.
I currently predict the GTX 1060 to be slower than the RX 480 on Einstein BRP6/CUDA55, perhaps rather substantially, but to be rather a low lower in system level power consumption and system resource requirement (with the current applications and drivers). I don't have an RX 480 on order now, so probably will not provide a direct comparison on the same host for that, but do plan to swap some 1060 (maybe even the one I just pre-ordered, if that comes through) into the host I used to provide 1070 Einstein data.
I've been able to raise my 1080's memory clock using PrecisionX OC
Are you able to adjust independently the 1080 memory clock used in the P0(gaming) state and in the P2 (distributed computing) state using PrecisionX?
Running which load are you running the memory clock at 5450? Which tool are you relying on for reporting that value?
Since I've installed the 1080, it's mostly been running in P2 since it's mostly been doing BOINC work. While running BOINC, I can use PrecisionX OC to adjust the memory clock above what it would normally operate at in P2 (small correction: currently running at 5467MHz). Both PrecisionX OC and Nvidia Inspector report the same value (within 1MHz) when I adjust values this way. I have not been able to adjust the core clock speed at all using PrecisionX OC or NVI.
I've had limited testing in P0, mostly looking at clock values when gaming (Fallout 4). It seems this game is not pushing my card to the max as it does not boost above base clock in game, although I can tell a noticeable improvement over the 980Ti in terms of frame smoothness.
I recall when the 780Ti was first released it seemed to often not boost its core clock in BOINC applications. This was later remedied, whether by the applications or the drivers I don't recall.
I also recently acquired a 1070 and have found that it always runs at a core boost of over 1900MHz, ranging from 1923MHz to 1967MHz (it varies, not sure why at the moment; well within temp and power settings). I am currently running the memory on this card at 4293MHz as reported by PrecisionX OC and NVI.
P.S. I should probably mention that both my 1080 and 1070 are EVGA cards. I don't know if this makes a difference in my ability to adjust the memory clocks with EVGA PrecisionX OC.
I picked up an EVGA 1070 FTW model directly from EVGA. I thought I was pre-ordering, and then they surprised me by shipping the next day.
I'm currently working through overclocking with PrecisionX, and am approaching stability at 2050 core (boost), 4600 / 9200 memory, steady 69c / 32% fan on the default fan curve running 4x FGRP6. I get consistent computation errors when I run the memory higher, but I haven't yet tested the limits of the core clock. It is currently set to +100 over its factory value, which is already 100+ over a Founder's.
HWMonitor reports 4x using 94% GPU capacity, 83% frame buffer capacity, memory controller 12%. I don't have power draw numbers yet. Best guess, it will be in the 200k daily range, but my stats are all over the place as I type this because of stopping, starting, differing mix of concurrent units, crashes.
And if I may say so, it's a really good-looking card with internal lighting, supposedly RGB but I haven't bothered changing it from default white.
I picked up an EVGA 1070 FTW model directly from EVGA.
Yours is on a Windows 10 system, as is mine. I've got a thread going over in problems and bug reports regarding a memory leak which gives me currently about a two day time limit after reboot. If you could check your system for the symptoms of the problem, that would add a useful scope data point.
I picked up an EVGA 1070 FTW model directly from EVGA.
Yours is on a Windows 10 system, as is mine. I've got a thread going over in problems and bug reports regarding a memory leak which gives me currently about a two day time limit after reboot. If you could check your system for the symptoms of the problem, that would add a useful scope data point.
I'll monitor the system. I'm on a fresh reboot from last night after a driver crash, and have settled on a +100 +800 OC. It's running 4x BRP6 1.57, a few Climate@Home units that will finish soon, and the other CPU tasks pending the next round of GW searches are SETI (Lunatics) apps. No Gamma-Ray Pulsar Search, and OC was only done with PrecisionX, so another difference is nVidia Inspector is not running.
You're using about 6 GB of the cards memory but hardly any memory bandwidth? Did you swap those numbers?
And as I said before, be careful with very high memory overclocks - the card may not crash but become slower as you approach the hard limit. This was reported around 5500 MHz for the GTX 1080 in gaming reviews.
I've come across a little
)
I've come across a little more material regarding the forthcoming GTX 1060 release. There are some possibly important points of contrast to the pattern of the 1080 and 1070 recent releases.
While for those cases review samples were distributed well in advance of first shipping date, and the NDA embargo on reviewers discussing their findings was a week or so before release date, in this case I think the reviewer samples went out later, and in any case the strongly rumored review embargo release time is on the same day as first shipments.
Possibly a bigger difference is in the Founders Edition matter. For the 1080 and 1070, about five of the major card makers sold FE cards boxed with their own logo, but built, so far as anyone has mentioned, identically to the Nvidia design. Near first shipment date, these completely dominated the available card supply, especially for the 1080, and distinct designs have only crept into the marketplace slowly. Also for both of these cards Nvidia took the position that this Founders Edition was something they expected to remain available throughout the life of the card type.
For the 1060 I've seen plausible claims that the "limited" FE card is only being made by Nvidia (which probably means assembled by a Chinese contract assembly house at their direction, quite possibly Hon Hai=Foxconn) and only available from them. However a range of card house cards are expected to be available for order on release day. And the "limited" FE card is slated to disappear from the market in perhaps a month.
Lastly, and of special interest for Einstein folks, I've seen multiple ways of saying that Nvidia has given the 1060 a better shake than it did the 960 as regards memory performance. This applies whether you compare within family (1060 to 1070 appears to have a much higher fraction than did 960 to 970) or relationship to previous generation (1060 from 960 nominally appears to be a big memory performance bump). I think the 960 was deemed a disappointing card for Einstein performance by more than one user, but this gives me some hope for better things this time.
So I am steadfast in my intention to buy a 1060 as soon as I can, and report trial results here. I am less committed than I was to specifically getting the Founders Edition, though I remain concerned about the character of the fan noise on the non-FE cards. But surely some of them will be OK, and I have potential places to put them that are not nearly so well supplied with cool air by the PC case as is my current 1070 arrangement.
Meanwhile, in the RX 480 power matter, one reviewer at Techpowerup has tested the new driver release and deems that it successfully shifted power consumption off the motherboard connection sufficiently, and that in compatibility mode (a defaulted off user option) it finally meets the stated total card power consumption.
Hello guys, i also have
)
Hello guys,
i also have Nvidia 1080 and the Nvidia inspector does not work so far when you try to adjust the memory clock in P2. I'm currently at 5100Mhz which is achieved by +500 overclock. I hope the new version of Nvidia inspector recognise the memory correctly so you can adjust the speed acording in P2 as my 1080 is able to go up to 5750Mhz.
Thanks
I've been able to raise my
)
I've been able to raise my 1080's memory clock using PrecisionX OC. Currently have it running around 5450MHz, though I could probably go higher.
RE: I've been able to raise
)
Are you able to adjust independently the 1080 memory clock used in the P0(gaming) state and in the P2 (distributed computing) state using PrecisionX?
Running which load are you running the memory clock at 5450? Which tool are you relying on for reporting that value?
For my 1070, the one thing I have seen to work and which I have been using has been to use Nvidia Inspector to specify an offset clock for the P0 state while the card was at the moment in the P8 state. This same offset then also affected the P2 state. A serious concern with my method is that it sets the P0 state clock rate higher than might work. I'd be very happy to be able to influence P2 clock rates (memory and core) independently of the rates for other states.
While GTX 1060 launch day and
)
While GTX 1060 launch day and review embargo release day are supposedly both next Tuesday, July 19, I followed up on a post claiming two places were accepting pre-orders, and placed an order for a Zotac GTX 1060 card at a nominal price of just under $260 plus moderate shipping. So perhaps the claimed $250 MSRP for this one is less of a fib than the claimed MSRP for 1080, 1070, and 480.
Or maybe sabrePC will never ship against my order. Time will tell.
I've been looking at NowInStock listings for the 1080, 1070, and RX 480, and the recent pattern seems to be that GTX 1080 and RX 480 listings get cleaned out very shortly after they are posted, but that for several days now there have been multiple Green (e.g. accepting orders claimed to be against actual stock) listings for the 1070 at any given moment. On the other hand, I've not spotted any 1070s anywhere near the stated $380 MSRP. But if availability stays good, I'd expect transaction prices to come down a little.
I currently predict the GTX 1060 to be slower than the RX 480 on Einstein BRP6/CUDA55, perhaps rather substantially, but to be rather a low lower in system level power consumption and system resource requirement (with the current applications and drivers). I don't have an RX 480 on order now, so probably will not provide a direct comparison on the same host for that, but do plan to swap some 1060 (maybe even the one I just pre-ordered, if that comes through) into the host I used to provide 1070 Einstein data.
RE: RE: I've been able to
)
Since I've installed the 1080, it's mostly been running in P2 since it's mostly been doing BOINC work. While running BOINC, I can use PrecisionX OC to adjust the memory clock above what it would normally operate at in P2 (small correction: currently running at 5467MHz). Both PrecisionX OC and Nvidia Inspector report the same value (within 1MHz) when I adjust values this way. I have not been able to adjust the core clock speed at all using PrecisionX OC or NVI.
I've had limited testing in P0, mostly looking at clock values when gaming (Fallout 4). It seems this game is not pushing my card to the max as it does not boost above base clock in game, although I can tell a noticeable improvement over the 980Ti in terms of frame smoothness.
I recall when the 780Ti was first released it seemed to often not boost its core clock in BOINC applications. This was later remedied, whether by the applications or the drivers I don't recall.
I also recently acquired a 1070 and have found that it always runs at a core boost of over 1900MHz, ranging from 1923MHz to 1967MHz (it varies, not sure why at the moment; well within temp and power settings). I am currently running the memory on this card at 4293MHz as reported by PrecisionX OC and NVI.
P.S. I should probably mention that both my 1080 and 1070 are EVGA cards. I don't know if this makes a difference in my ability to adjust the memory clocks with EVGA PrecisionX OC.
I picked up an EVGA 1070 FTW
)
I picked up an EVGA 1070 FTW model directly from EVGA. I thought I was pre-ordering, and then they surprised me by shipping the next day.
I'm currently working through overclocking with PrecisionX, and am approaching stability at 2050 core (boost), 4600 / 9200 memory, steady 69c / 32% fan on the default fan curve running 4x FGRP6. I get consistent computation errors when I run the memory higher, but I haven't yet tested the limits of the core clock. It is currently set to +100 over its factory value, which is already 100+ over a Founder's.
HWMonitor reports 4x using 94% GPU capacity, 83% frame buffer capacity, memory controller 12%. I don't have power draw numbers yet. Best guess, it will be in the 200k daily range, but my stats are all over the place as I type this because of stopping, starting, differing mix of concurrent units, crashes.
And if I may say so, it's a really good-looking card with internal lighting, supposedly RGB but I haven't bothered changing it from default white.
Gamboleer wrote:I picked up
)
Yours is on a Windows 10 system, as is mine. I've got a thread going over in problems and bug reports regarding a memory leak which gives me currently about a two day time limit after reboot. If you could check your system for the symptoms of the problem, that would add a useful scope data point.
RE: Gamboleer wrote:I
)
I'll monitor the system. I'm on a fresh reboot from last night after a driver crash, and have settled on a +100 +800 OC. It's running 4x BRP6 1.57, a few Climate@Home units that will finish soon, and the other CPU tasks pending the next round of GW searches are SETI (Lunatics) apps. No Gamma-Ray Pulsar Search, and OC was only done with PrecisionX, so another difference is nVidia Inspector is not running.
RE: HWMonitor reports 4x
)
You're using about 6 GB of the cards memory but hardly any memory bandwidth? Did you swap those numbers?
And as I said before, be careful with very high memory overclocks - the card may not crash but become slower as you approach the hard limit. This was reported around 5500 MHz for the GTX 1080 in gaming reviews.
MrS
Scanning for our furry friends since Jan 2002