Contribute
Register

Report your NVIDIA Web Driver and CUDA issues and I'll pass them on to the developers

Status
Not open for further replies.
As my tests, Kelper base GT630/GT730 also have sluggish issue with native or Web driver, only found on SKL/KBL + GT730 system, IVY/Haswell + GT730 system does not have this issue. Even IVY Celeron 1037U + GT730 has better UI performance than i3 7100 + GT730. There are no relevant to SMBIOS, I tried many SMBIOS, results are same.

Native GT730 driver seems to be better than Web driver, UI performance are acceptable, most fo the apps response and window resizing are OK except during video playback (sluggish issue normally found during video playback), but some still slightly sluggish.
 
No difference. Still lagging like crazy...

I've discovered another thing in the code of NVDAStartupWeb.

It seems to look for a new bootarg "nvda_drv_vrl" and take values between inclusive 0 and 4 for it. Not sure what it does.

It's not present on my 2012 rMBP which does not lag with the 156 Webdriver, so I'm not thinking it's relevant to our issues . . .
 
Confirmed. I just tested my HP with iMac18,3 SMBIOS. Smooth as silk. The issue has absolutely nothing to do with choice of system definition.

As my tests, Kelper base GT630/GT730 also have sluggish issue with native or Web driver, only found on SKL/KBL + GT730 system, IVY/Haswell + GT730 system does not have this issue. Even IVY Celeron 1037U + GT730 has better UI performance than i3 7100 + GT730. There are no relevant to SMBIOS, I tried many SMBIOS, results are same.

Native GT730 driver seems to be better than Web driver, UI performance are acceptable, most fo the apps response and window resizing are OK except during video playback (sluggish issue normally found during video playback), but some still slightly sluggish.

Here's the next horrible idea: You both have well-confirmed that it's not connected to SMBIOS. Cool.

You know, Clover can fake CPUs too . . . ;p

FakeCPUID. Now, Faking Haswell and newer is generally not generally recommended due to new modes and instructions added on Haswell. But that should be fine for Skylake and newer, since they will possess the Haswell+ instructions. In fact, the CPUID we might try is the one from Broadwell (0x0306D0). And if that does not work, try Haswell (0x0306C0). Not sure it's a good idea to go back before that . . .
 
Last edited:
In fact, the CPUID we might try is the one from Broadwell (0x0306D0). And if that does not work, try Haswell (0x0306C0). Not sure it's a good idea to go back before that . . .

I use HWP for PM, always use 0x0306a0 + nullcpupowermanagement to fully disable SKL/KBL XCPM, Fake CPU ID will not correct this issue, issue should caused by display/MTL VAbundle when IQSV is off, when IQSV is on sluggish issue may caused by display/MTL VAbundle, xxxx.framework or dyld_share_cache, use SMBIOS (like MacPro5,1) not supported by AppleGVA should greatly reduce sluggish issue, UI performance should be acceptable but IQSV HW decoding/encoding feature will never work.
 
I use HWP for PM, always use 0x0306a0 + nullcpupowermanagement to fully disable SKL/KBL XCPM, Fake CPU ID will not correct this issue, issue should caused by display/MTL VAbundle when IQSV is off, when IQSV is on sluggish issue may caused by display/MTL VAbundle, xxxx.framework or dyld_share_cache, use SMBIOS (like MacPro5,1) not supported by AppleGVA should greatly reduce sluggish issue, UI performance should be acceptable but IQSV HW decoding/encoding feature will never work.

Thing is people are getting this issue even when the iGPU is not active.

And I have Shiki faking iMac14,2 for AppleGVA. But it's still sluggish . . .

VDADecoderChecker still reports IQSV working even while the lag occurs. In fact, for me at at least, videos generally play fine and are unaffected by the lag, I think, because the iGPU is partially saving the system.

Edit: I do agree this explains the improvement pastrychef saw on MacPro5,1. But even fixing this does not totally fix the lag on 156/157. Real Macs have this lag on 156/157. And even with patched up IQSV and iGPU cooperation, it's still dizzyingly bad. Fixing things up with Shiki only fixes that "general High Sierra lag" people have always complained about. This is far worse.
 
Last edited:
Thing is people are getting this issue even when the iGPU is not active.

And I have Shiki faking iMac14,2 for AppleGVA. But it's still sluggish . . .

VDADecoderChecker still reports IQSV working even while the lag occurs. In fact, for me at at least, videos generally play fine and are unaffected by the lag, I think, because the iGPU is partially saving the system.

Yes, when IQSV is off, UI still sluggish.

With IQSV on, video playback is good, but during video playbcack, system UI, others apps response and windows resizing are lagging, not much helpful.
 
Yes, when IQSV is off, UI still sluggish.

With IQSV on, video playback is good, but during video playbcack, system UI, others apps response and windows resizing are lagging, not much helpful.

Same as my experience. :/

And what's funniest is that downgrading to 104-106 works. So it's not something that changed fundamentally in macOS. It's got to somehow be something that changed with the driver . . .

And it's not library validation. The symptoms are wrong. Transparency still works. And re-enabling it fixes nothing.

I'm just trying to understand why it only affects Skylake and newer, including Skylake and newer real macs with eGPU. That doesn't make sense . . .
 
Last edited:
I use HWP for PM, always use 0x0306a0 + nullcpupowermanagement to fully disable SKL/KBL XCPM, Fake CPU ID will not correct this issue, issue should caused by display/MTL VAbundle when IQSV is off, when IQSV is on sluggish issue may caused by display/MTL VAbundle, xxxx.framework or dyld_share_cache, use SMBIOS (like MacPro5,1) not supported by AppleGVA should greatly reduce sluggish issue, UI performance should be acceptable but IQSV HW decoding/encoding feature will never work.

I'd also love to say that FakeCPUID won't work, but, um, it seems like Coffee Lake is unaffected by the lag, unless Kaby Lake is faked I presume . . .

The Coffee Lake id is 0x906EA. Might be worth a shot . . .
 
Status
Not open for further replies.
Back
Top