[SOLVED] Nvidia GeForce GTX 1080/1070

[objective-x]

It works
16x
View attachment 247860

Thanks for checking! What's the Link Speed under the PCI section, though? That's what determines whether it's running at PCIe 1 or 2 speeds.

 

[laxattack]

I just want to say after lurking and following this thread, I am happy with the results!

 

[Fl0r!an]

Just to confirm, even with the Nvidia web drivers, un-flashed cards still run at PCIe 1.0 speeds in a Mac Pro, correct?

In OS X the WebDrivers will enable PCIE 2.0 in MacPro4,1 & 5,1. In Windows it'll be limited to PCIE 1.1 speed unless it's flashed.

This only applies to genuine Macs though, it has never been an issue with Hackintoshs.

 

[Fl0r!an]

MVC posted some benchmarks with 1080 & Titan running Valley@5K, performance increase compared to Maxwell is awesome: http://www.macvidcards.com/blog/5k-valley-testing

CUDA on the other hand doesn't look too impressive, 1080 gets beaten by the old 980Ti in most benchmarks: http://barefeats.com/cmp_pascal.html
GTX 1080 OpenCL performance is below RX 480.

 

[akashbanerjeeab]

https://www.tonymacx86.com/threads/need-help-with-audio-and-cpu-core-count.219256/

If some can help please do so Thanks.

 

[masterkain]

amagad it works on a z170 omg omg

had to install agpdfix tho, otherwise black screen

rocket league still **** but that's the poor implementation

 

[joevt]

joevt, I fixed the audio problem with the vodoohda software. Install it, reboot, choose the third digital out (HDMI) option and you should be ready to go.

I think the option to choose will depend on the connector the display is connected. I would like to figure out how to get HDMI audio working correctly using an SSDT edit (using Toleda's audio guide or scripts), or possibly HDMIAudio.kext (which I think just does mostly what the SSDT edits do), or a modified version of the kext. That way, the name of the audio device will be the display, and only connected displays will appear in the list.

Do you have any problem with your mouse pointer changing between the left/right part of the screen? This is driving me crazy!

Yes, this is a common problem even with Apple MacBook Pros, and it occurs when using any display where the left and right side use a different DisplayPort stream (4K MST displays, and 5K dual cable displays). You can find a few examples of people having this problem in Apple's discussion forums.

It works
16x

What's the Link Speed under the PCI section, though? That's what determines whether it's running at PCIe 1 or 2 speeds.

The speed (either 2.5 GT/s or 5 GT/s or 8 GT/s) is listed in the PCI section, but only if your DSDT has the changes that indicate the PCIe slot devices. And anyway, the graphics card may idle at 2.5 GT/s and then speed up when doing rendering, so in that case you need to read the actual PCIe status register while doing rendering to get the speed. In mac OS, you would use lspci to dump the registers while the card is doing 3d work, find the register value, and interpret the bit fields using the PCI specification. In Windows, you would use GPU-Z, click the "?" icon to start a rendering test, which should cause the speed to inrease from the idle value to the full speed value.

 

[asiga]

CUDA on the other hand doesn't look too impressive, 1080 gets beaten by the old 980Ti in most benchmarks: http://barefeats.com/cmp_pascal.html
GTX 1080 OpenCL performance is below RX 480.

The only unexpected numbers there are from the 1080. On the other hand, the 1080-Ti is getting a similar LuxMark score as the Pascal Titan X scores from the LuxMark site (22335 vs 24536), which is a really *great* score, specially if you consider that the Titan X score was obtained in Windows, whose OpenCL drivers tend to be more optimized than the rest of platforms.

There's definitely something very weird with the 1080 in the benchmarks from that page. The 1080-Ti gets the expected results, though.

BTW: Too bad they only tested the simple Luxball scene! The test deserved the other more complex scenes!

 

[kpoloboy916]

The only unexpected numbers there are from the 1080. On the other hand, the 1080-Ti is getting a similar LuxMark score as the Pascal Titan X scores from the LuxMark site (22335 vs 24536), which is a really *great* score, specially if you consider that the Titan X score was obtained in Windows, whose OpenCL drivers tend to be more optimized than the rest of platforms.

There's definitely something very weird with the 1080 in the benchmarks from that page. The 1080-Ti gets the expected results, though.

BTW: Too bad they only tested the simple Luxball scene! The test deserved the other more complex scenes!

Seems to match around my GTX 1080. Super weird. seems like the 1080 isn't performing up to par. what is also weird, is the GTX 1070 score higher than GTX 1080 in benches as well.

attached below are my GTX 1080 benches.

 

[objective-x]

In OS X the WebDrivers will enable PCIE 2.0 in MacPro4,1 & 5,1. In Windows it'll be limited to PCIE 1.1 speed unless it's flashed.

Ah thanks for the clarification... I had it the other way around! Thankfully based on some benchmarks I found, a 1060 shouldn't be bottlenecked at 16x PCIe 1.1.

The speed (either 2.5 GT/s or 5 GT/s or 8 GT/s) is listed in the PCI section, but only if your DSDT has the changes that indicate the PCIe slot devices. And anyway, the graphics card may idle at 2.5 GT/s and then speed up when doing rendering, so in that case you need to read the actual PCIe status register while doing rendering to get the speed. In mac OS, you would use lspci to dump the registers while the card is doing 3d work, find the register value, and interpret the bit fields using the PCI specification. In Windows, you would use GPU-Z, click the "?" icon to start a rendering test, which should cause the speed to inrease from the idle value to the full speed value.

Thanks for this info! I didn't realize the link speed can change depending on utilization.