I'm considering replacing my current old and completely insufficient GTX 650 with the new GTX 1080 when it becomes available. I'm also going to bite the bullet and upgrade to Windows 10 to take advantage of DirectX 12.

I've also considered a GTX 970 as a replacement but comparing current price to performance of the two, the GTX 1080 wins, hands-down.

However, I'm nowhere near an expert when it comes to GPUs, and I'm afraid that I may be making a mistake or bad assumptions.

My system specs are:
O/S: Windows 7 Pro (moving to Windows 10 Pro)
Mobo: ASUS Sabertooth 990FX R2.0, PCIe 2.0 x16
CPU: AMD FX-8350 8-core
Memory: 8GB Corsair Vengeance DDR3 1600
PSU: Corsair HX750 750-watt

Does anyone see a problem with a GTX 1080 on my system?

Is it overkill, or will the new board be fully utilized?

Thanks.

Should be just fine, I think some work units are CPU limited to some degree but I don't think that means you wouldn't see improvement from a faster GPU, some people might mention the fact that your PCIE is only 2.0, but that should make little difference in x16.

Thanks, Skyler.

Does anyone have a GTX 1080 already?

I'm considering replacing my current old and completely insufficient GTX 650 with the new GTX 1080 when it becomes available. I'm also going to bite the bullet and upgrade to Windows 10 to take advantage of DirectX 12.

I've also considered a GTX 970 as a replacement but comparing current price to performance of the two, the GTX 1080 wins, hands-down.

However, I'm nowhere near an expert when it comes to GPUs, and I'm afraid that I may be making a mistake or bad assumptions.

My system specs are:
O/S: Windows 7 Pro (moving to Windows 10 Pro)
Mobo: ASUS Sabertooth 990FX R2.0, PCIe 2.0 x16
CPU: AMD FX-8350 8-core
Memory: 8GB Corsair Vengeance DDR3 1600
PSU: Corsair HX750 750-watt

Does anyone see a problem with a GTX 1080 on my system?

Is it overkill, or will the new board be fully utilized?

Thanks.

I think 16gb of ram should be more efficient if you can swing that too. Microsoft just came out saying anything less than 4gb makes Win10 slow.

Right now they are selling both 1070's and 1080's, most are already sold out or on backorder though. You can check the status here: http://www.nowinstock.net/ Just click on the 1080 or 1070 link. Soon though the secondary market will be free to flood the market with their own versions of these and there should be plenty to go around. Some early reports have said the fan noise can be a bit loud as it changes alot, with some aftermarket programs doing little to change that. One manufacturer is coming out with a water cooled one, all in one package, that should solve that problem though. This thread at Einstein has lots of details about the cards and even some links to some sites that were able to test some of the new cards. NONE though did any testing on running Boinc workunits though:
https://einstein.phys.uwm.edu/forum_thread.php?id=12101

Right now they are selling both 1070's and 1080's, most are already sold out or on backorder though. You can check the status here: http://www.nowinstock.net/ Just click on the 1080 or 1070 link. Soon though the secondary market will be free to flood the market with their own versions of these and there should be plenty to go around. Some early reports have said the fan noise can be a bit loud as it changes alot, with some aftermarket programs doing little to change that. One manufacturer is coming out with a water cooled one, all in one package, that should solve that problem though. This thread at Einstein has lots of details about the cards and even some links to some sites that were able to test some of the new cards. NONE though did any testing on running Boinc workunits though:
https://einstein.phys.uwm.edu/forum_thread.php?id=12101

Read through the Einstein thread. Why anyone would buy NVidia cards for running Einstein is beyond me. Maybe the new ones will perform better but I'm skeptical. I run mostly NVidia but that's because I primarily crunch GPUgrid. Have tried them on other projects and generally they're poor performers compared with AMD. Maybe I just haven't found the right project. Are there any other projects in which NV cards perform well?

Does anyone have a GTX 1080 already?

Installing one now. Titan X were quite underwhelming compared to GTX 980s with GPUGRID long runs on a per core/per clock basis. I have a feeling the GTX 1080 might do the same to the future big-die cards. Slower, but not proportionately so. Might be an efficiency buy (980) over a absolute performance buy (980 Ti/Titan X). I expect GTX 1070 to do the same.

In the UK the only additions that will be available in the next few days (2-4) are founder editions of the GTX1080, and they will cost £619. Manufacturer editions are due in several weeks, up to a month for some. They will cost ~£40 less. No sign of the GTX1070's.

https://www.pugetsystems.com/labs/articles/GTX-1080-CUDA-performance-on-Linux-Ubuntu-16-04-preliminary-results-nbody-and-NAMD-803/
NAMD 2.11 500 time steps of stmv (satellite tobacco mosaic virus - million atom)
GTX 1080 --> 54 sec .81 day/ns
GTX 980 --> 71 sec 1.21 day/ns

PS. CUDA 8 dev kit preview was released a couple of days back.

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Does anyone have a GTX 1080 already?

Installing one now. Titan X were quite underwhelming compared to GTX 980s with GPUGRID long runs on a per core/per clock basis. I have a feeling the GTX 1080 might do the same to the future big-die cards. Slower, but not proportionately so. Might be an efficiency buy (980) over a absolute performance buy (980 Ti/Titan X). I expect GTX 1070 to do the same.

Admin team to upgrade ACEMD - stay tuned.

From host #295749:

<core_client_version>7.6.22</core_client_version>
<![CDATA[
<message>
(unknown error) - exit code -59 (0xffffffc5)
</message>
<stderr_txt>
# GPU [GeForce GTX 1080] Platform [Windows] Rev [3212] VERSION [65]
# SWAN Device 0 :
# Name : GeForce GTX 1080
# ECC : Disabled
# Global mem : 4095MB
# Capability : 6.1
# PCI ID : 0000:01:00.0
# Device clock : 1733MHz
# Memory clock : 5005MHz
# Memory width : 256bit
# Driver version : r368_20 : 36825
#SWAN: FATAL: cannot find image for module [.nonbonded.cu.] for device version 610

NAMD 2.11 500 time steps of stmv (satellite tobacco mosaic virus - million atom)
GTX 1080 --> 54 sec .81 day/ns
GTX 980 --> 71 sec 1.21 day/ns

Great find skgiven.

PS. CUDA 8 dev kit preview was released a couple of days back.

Though the provided driver (branch) is not up to date in the massive 1.2GB package so one has to re-install the most recent driver afterwards. (First CUDA toolkit in a long while without up to date CUDA / graphics driver.)

Read through the Einstein thread. Why anyone would buy NVidia cards for running Einstein is beyond me. Maybe the new ones will perform better but I'm skeptical. I run mostly NVidia but that's because I primarily crunch GPUgrid. Have tried them on other projects and generally they're poor performers compared with AMD. Maybe I just haven't found the right project. Are there any other projects in which NV cards perform well?

The benefit of NV cards is that they can run CUDA and OpenCL. AMD/ATI can't do CUDA (without a lot of jiggery pokery and then only poorly). For those who mainly crunch here NV is the way to go and you have a range of backup GPU projects which you can participate on with the same device. NV cards are good at POEM also. They might not be quite as good as ATI but they are very close (probably within 20% for equivalently priced cards). They used to be, and perhaps still are good at some of the maths projects but that's not my scene. Compared to AMD/ATI the NV cards are bad at MW and poor at Einstein, but good at POEM and excellent here. Based on the specs very little will change in that respect. Unless there is something specific in the GTX1080 and perhaps CUDA8 for Einstein (who are presently still Beta running a CUDA5.5app) I can't see anything changing there.

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Read through the Einstein thread. Why anyone would buy NVidia cards for running Einstein is beyond me. Maybe the new ones will perform better but I'm skeptical. I run mostly NVidia but that's because I primarily crunch GPUgrid. Have tried them on other projects and generally they're poor performers compared with AMD. Maybe I just haven't found the right project. Are there any other projects in which NV cards perform well?

The benefit of NV cards is that they can run CUDA and OpenCL. AMD/ATI can't do CUDA (without a lot of jiggery pokery and then only poorly). For those who mainly crunch here NV is the way to go and you have a range of backup GPU projects which you can participate on with the same device. NV cards are good at POEM also. They might not be quite as good as ATI but they are very close (probably within 20% for equivalently priced cards). They used to be, and perhaps still are good at some of the maths projects but that's not my scene. Compared to AMD/ATI the NV cards are bad at MW and poor at Einstein, but good at POEM and excellent here. Based on the specs very little will change in that respect. Unless there is something specific in the GTX1080 and perhaps CUDA8 for Einstein (who are presently still Beta running a CUDA5.5app) I can't see anything changing there.

I'm currently running 15 NVidia GPUs and of course they work great here as they're the only card supported. Are there any projects though that run both AMD and NVidia where NVidia wins? I can't think of any but possibly there's one/some. Of the projects I've tried (many) there's generally a big gap, with AMD coming out on top. Would like to know of a good project to run these cards on when GPUGrid is out of work. From what I can see of the math projects AMD dominates there too. Not an NV basher as I have 15 and am currently running only 2 AMD. But reality is reality (maybe):

http://www.scientificamerican.com/article/are-we-living-in-a-computer-simulation/

Admin team to upgrade ACEMD - stay tuned.

From host #295749:

<core_client_version>7.6.22</core_client_version>
<![CDATA[
<message>
(unknown error) - exit code -59 (0xffffffc5)
</message>
<stderr_txt>
# GPU [GeForce GTX 1080] Platform [Windows] Rev [3212] VERSION [65]
# SWAN Device 0 :
# Name : GeForce GTX 1080
# ECC : Disabled
# Global mem : 4095MB
# Capability : 6.1
# PCI ID : 0000:01:00.0
# Device clock : 1733MHz
# Memory clock : 5005MHz
# Memory width : 256bit
# Driver version : r368_20 : 36825
#SWAN: FATAL: cannot find image for module [.nonbonded.cu.] for device version 610

Yep, ditto on my host #179637:

<core_client_version>7.6.22</core_client_version>
<![CDATA[
<message>
(unknown error) - exit code -59 (0xffffffc5)
</message>
<stderr_txt>
# GPU [GeForce GTX 1080] Platform [Windows] Rev [3212] VERSION [65]
# SWAN Device 0 :
# Name : GeForce GTX 1080
# ECC : Disabled
# Global mem : 4095MB
# Capability : 6.1
# PCI ID : 0000:05:00.0
# Device clock : 1733MHz
# Memory clock : 5005MHz
# Memory width : 256bit
# Driver version : r368_20 : 36825
#SWAN: FATAL: cannot find image for module [.nonbonded.cu.] for device version 610

Will wait in the wings for an update.

Thank you.

The article skgiven pointed out alluded to such compatibility issues. Probably the admins will need a 10xx card in order to iron out the problems.

Basically the app doesn't recognise the GPU architecture, so it can't get tasks to run.
It should be a simple-ish task to tell the app what this series looks like architecturally, but there could be other issues too. Hopefully they won't have to wait for the CUDA 8 dev kit to be released to get this range up and going, but some app development might be required to get the best out of this range.
Perhaps in a few days it might be useful to resurrect the Beta queue, for GTX1000 series testing? Won't be getting one any time soon, far too expensive in the UK.

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

I have a GTX 1080, but getting computational errors. Any idea when they will fix this?

I expect that only Matt and Gianni are able to update the app, so only they could tell you.
For now you can use your GTX1080 at POEM.

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

On folding@home output range from :
GTX 980 : 350-500K PPD
GTX 980ti : 700-850k PPD
GTX 1080 : 750-1000k PPD

If nvidia keep this scaling can we expect...
GTX 1080ti : 1300-1600K PPD

Read through the Einstein thread. Why anyone would buy NVidia cards for running Einstein is beyond me. Maybe the new ones will perform better but I'm skeptical. I run mostly NVidia but that's because I primarily crunch GPUgrid. Have tried them on other projects and generally they're poor performers compared with AMD. Maybe I just haven't found the right project. Are there any other projects in which NV cards perform well?

The GTX 750 Ti's run great on Einstein. As I posted there, my GTX 750 Tis on singles BRP6-Beta work units are averaging 7800 seconds at 60% TDP as measured by GPU-Z, or 36 watts (Win7 64-bits). That is better than any of the AMD cards I have tried (HD 7790, HD 7870) in terms of efficiency (PPD/watt), though it was a while ago and I don't have the numbers. But things fall down with the GTX 960, apparently because the 128 bit memory bus (the same as on the 750 Ti) can't keep up with the increased number of shaders on the 960. I think that may be more or less true with the higher-end cards also, though I would expect the GTX 970 to be somewhat more efficient than the 980 for that reason also.

Both the GTX 750 Ti and the GTX 960 are very good on POEM too, as skgiven noted above, even though that is OpenCl and designed for the GCN architecture. At least those Nvidia cards are slightly more efficient than the AMD cards as well, though I have not checked out the higher-end cards on POEM. And on Folding, while we are on the subject, the Maxwell architecture is the best; I don't think the AMD cards are nearly as good for efficiency there either, even though they use OpenCl cores now.

I'm currently running 15 NVidia GPUs and of course they work great here as they're the only card supported. Are there any projects though that run both AMD and NVidia where NVidia wins? I can't think of any but possibly there's one/some. Of the projects I've tried (many) there's generally a big gap, with AMD coming out on top. Would like to know of a good project to run these cards on when GPUGrid is out of work. From what I can see of the math projects AMD dominates there too. Not an NV basher as I have 15 and am currently running only 2 AMD.

Both run well at Prime Grid too with Nvidia being faster on some parts and AMD being faster on others.

I'm currently running 15 NVidia GPUs and of course they work great here as they're the only card supported. Are there any projects though that run both AMD and NVidia where NVidia wins? I can't think of any but possibly there's one/some. Of the projects I've tried (many) there's generally a big gap, with AMD coming out on top. Would like to know of a good project to run these cards on when GPUGrid is out of work. From what I can see of the math projects AMD dominates there too. Not an NV basher as I have 15 and am currently running only 2 AMD.

Both run well at Prime Grid too with Nvidia being faster on some parts and AMD being faster on others.

Now we have the $199 RX 480 due to be available on June 27th:

http://wccftech.com/amd-radeon-rx-480-polaris-10-launch/