DaCM Posted August 27, 2016 Share Posted August 27, 2016 (edited) Laptop Specs i7 3610QM 8GB RAM GTX 660m Windows 10 Anniversary 64bit 1080p screen eGPU Components Zotac GTX 1060 Mini 550W Chieftec PSU PE4C v3 mPCIe Installation Steps - Uninstall mobile nvidia drivers, reboot - Take out wifi card to free up mPCIe slot - Install PE4C connector into the slot (with the GTX 1060 and PSU already set up with the PE4C) - Reboot, install desktop nvidia driver, reboot At this point the 1060 was properly recognised by the system (no errors at all), without the need to use Setup 1.3. There is no need to use any of the delay switches on the PE4C either, it is simple plug and play. The in-built 660m is not disabled this way though, and it was the GPU used for Optimus/internal screen output. I wanted to test internal screen performance using the 1060, so I proceeded to use Setup 1.3. - Install Setup 1.3, boot into it - Set eGPU port (#4 in my case) to GEN2 - Set endpoint to 4.0 - Disable dGPU - Ignore dGPU - iGPU + eGPU compaction - Chainload - test run (MBR) This way, the 660m is completely disabled and the 1060 can output to the internal screen. Throughout my testing GEN2 speeds were maintained by the PE4C without issue, no freezes, driver crashed, BSODs. I have performed some tests on the internal screen, but results are quite underwhelming unfortunately. In 3DMark 11 the system achieved P93xx points, and 34FPS in Unigine Heaven 4.0 (extreme) on the internal screen. (Normal scores are P17000 and 80+ FPS.) Game performance is very clearly limited by PCIe bandwidth as well (GPU usage often below 50%), but the framerates are interestingly tending to VSync framerates, such as 30 or 45 FPS. Fallout 4 Ultra: 29.9 FPS, often dipping as low as 23 FPS Fallout 4 Medium: 29.9 FPS, rarely dipping lower Doom Ultra (Vulkan): Constant stuttering (couldn't measure FPS as RivaTuner overlay didn't work) Doom High (Vulkan): Constant stuttering Witcher 3 High: 29 FPS, often dipping to low 20s No Man's Sky Ultra: Consistently well over 60 FPS, smooth Just Cause 2 (regardless of settings): around 25 FPS Team Fortress 2 Max: 45 FPS, microstuttering Borderlands 2 Ultra: 50-60 FPS, smooth Dawn of War II Retribution Benchmark on Ultra: 40 FPS average Overall, there is a roughly 60% performance loss on the internal screen compared to normal GTX 1060 performance. Furthermore, some (even undemanding) games are rendered unplayable due to the stuttering the lack of bandwidth causes. I would not recommend this setup for internal screen use. On the external screen, results are roughly as expected, with only a slight performance loss. I achieved 61.4FPS in Unigine Heaven 4.0 (extreme) on an external screen. Game performance is also much better: Fallout 4 Ultra: 60 FPS smooth, rare dips into 50s Witcher 3 Ultra: 25 - 45 FPS, dips are visible Witcher 3 High: 45 - 60 FPS, smooth Doom Ultra (Vulkan): Couldn't measure FPS, but very smooth No Man's Sky Ultra: Over 60, smooth Just Cause 2 (regardless of settings): 40 - 45 FPS Dawn of War II Retribution Benchmark on Ultra: 60 FPS average Guild Wars 2: The game uses an old engine and is horribly optimised. It seems to be using a lot of PCIe bandwidth, and performance is hardly playable at any setting, with constant stuttering. (The internal 660m performs better) Lastly, I haven't tested if not using Setup 1.3 and having both nvidia cards run would result in a performance loss on the external screen. I would guess that it would not make a difference though, as the 660m would drive the internal screen, while the 1060 the external one. So overall, if you are using an external screen (which you should be, considering the performance), the MSI GE60 (2012) is plug and play. Edited August 27, 2016 by DaCM 3 Quote Link to comment Share on other sites More sharing options...
akeean Posted September 3, 2016 Share Posted September 3, 2016 Thanks for the summary. 2012 means it's connected via a x1 PCIe Gen 2 lane?What resolution did you test on?This one fares much better & closer to expectation on a x1 pcie connection. Stupid question: Is optimus compression enabled? This should save quite some bandwith&might explan the performance loss. I don't know if it would work without the compression enabled at all, but maybe this points you towards a fix.I'm not on eGPU yet, so not speaking from experience, but looking for a similar setup&reading up on the matter. Will be with a PCIe 3 system and a i7 4700MQ.Sent from my XT1058 using Tapatalk Quote Link to comment Share on other sites More sharing options...
DaCM Posted September 7, 2016 Author Share Posted September 7, 2016 I haven't tried Firestrike, but I have done some calculations since and it turns out that 1080p screen output at 30FPS uses up almost exactly half of the 5Gbps bandwidth available on a 2.0 1x link, which explains the massive performance degradation and stuttering on the internal screen. Lowering the resolution in games improves performance on the internal screen drastically. Optimus compression is enabled on the internal screen, but due to the 1080p resolution the performance can't really be helped I think. (To send 1080p 60FPS to the internal screen the GPU would use up 100% of the available bandwidth and not be able to receive anything to process at all, which is not possible, of course.) Everything is fine on the external screen though, which is why I would suggest to do a similar setup only if you intend to use one. 1 Quote Link to comment Share on other sites More sharing options...
Another Tech Inferno Fan Posted September 8, 2016 Share Posted September 8, 2016 (edited) MSI Afterburner has a PCIe bus utilisation monitoring feature. Perhaps you could draw up a table showing PCIe bus utilisation vs screen res and internal/external monitor? Edited September 8, 2016 by Arbystrider 1 Quote Link to comment Share on other sites More sharing options...
DaCM Posted September 12, 2016 Author Share Posted September 12, 2016 I did think about that, but the Afterburner monitoring is definitely not accurate. I am not sure if it is because it excludes the bandwidth used for sending the picture back to the internal screen or because of some other reason, but the utilisation reading never goes above 60%, even when it is clearly the bottleneck. 1 Quote Link to comment Share on other sites More sharing options...
Another Tech Inferno Fan Posted September 13, 2016 Share Posted September 13, 2016 (edited) 18 hours ago, DaCM said: the utilisation reading never goes above 60%, even when it is clearly the bottleneck. This is happening to me as well - The bus utilisation never goes beyond 65% in any program that uses Optimus. I figured this was coincidence that my 580 never used anything more than 65%, and that a faster card that yielded more performance would utilise more bandwidth. I concluded this based on the fact that the 580 would consistently show 100% GPU core utilisation within Afterburner during such loads, though I've learned not to trust even that. There desperately needs to be someone with a massive collection of video cards and can benchmark every one of them on x1.2Opt to see which cards aren't held back by bandwidth, and what the bandwidth requirement is to get the ideal performance out of each card. I'd do the grunt work of all that myself if I had the resources. Edited September 13, 2016 by Arbystrider Quote Link to comment Share on other sites More sharing options...
Another Tech Inferno Fan Posted September 14, 2016 Share Posted September 14, 2016 Hate to double-post, but in light of new information I feel this is warranted. I was playing TOXIKK and I noticed that bus utilisation consistently was above 67%. Sometimes as high as 96%. I managed to reproduce this using MSI Kombustor's Furry Donut stress test. UTIL, % shows core utilisation, BUS, % shows PCIe bus utilisation. From this I think I can conclude that the PCIe x1.2 interface is not the performance-limiting factor in my eGPU config, and that Optimus doesn't in fact use an entire 33-40% of the PCIe x1.2 interface's bandwidth. What of you and your GTX 1060, OP? What is your bus utilisation figure under Furmark? Quote Link to comment Share on other sites More sharing options...
douirc Posted September 17, 2016 Share Posted September 17, 2016 (edited) I'll chime in in case it's helpful. I'm running X230 i5 8GB with PE4C V3.0 Expresscard and a Gigabyte GTX 1060 6GB mini connected to a PB258q. I set Windows 10 to output to Display 2 only (external). I'm using GPU-Z to measure Bus Interface Load and reaching a max of 73% during gaming. My synthetic score is below. I'm looking for ways to improve the performance even further. Unigine Heaven Benchmark 4.0 FPS: 32.2 Score: 810 Min FPS: 7.5 Max FPS: 63.4 System Platform: Windows NT 6.2 (build 9200) 64bit CPU model: Intel(R) Core(TM) i5-3320M CPU @ 2.60GHz (2594MHz) x2 GPU model: Intel(R) HD Graphics 4000 10.18.10.4358/NVIDIA GeForce GTX 1060 6GB 21.21.13.6909 (4095MB) x1 Settings Render: Direct3D11 Mode: 2560x1440 8xAA fullscreen Preset Custom Quality Ultra Tessellation: Extreme Edited September 17, 2016 by douirc Quote Link to comment Share on other sites More sharing options...
DaCM Posted September 18, 2016 Author Share Posted September 18, 2016 (edited) On 2016.09.14. at 0:20 PM, Arbystrider said: Hate to double-post, but in light of new information I feel this is warranted. I was playing TOXIKK and I noticed that bus utilisation consistently was above 67%. Sometimes as high as 96%. I managed to reproduce this using MSI Kombustor's Furry Donut stress test. UTIL, % shows core utilisation, BUS, % shows PCIe bus utilisation. From this I think I can conclude that the PCIe x1.2 interface is not the performance-limiting factor in my eGPU config, and that Optimus doesn't in fact use an entire 33-40% of the PCIe x1.2 interface's bandwidth. What of you and your GTX 1060, OP? What is your bus utilisation figure under Furmark? I got a consistent 95% readout (3rd value in the 2nd row of the overlay). If you look at the FPS counter though, Afterburner is showing 100FPS, while Furmark is showing 50FPS. Since 60FPS in 1080p would practically use 100% of the available bandwidth, Afterburner is the incorrect one. 50 FPS, however, only takes up around 80-85%, and considering that Furmark is a very simple benchmark in terms of visuals, it is very possible that the data sent to the GPU only takes up the remaining approximately 10%. Looking at Furmark results from other 1060s in normal setups though, the result they produce is an average of 65 FPS, which would be impossible on a 1.2 internal screen setup due to the bandwidth limit: So in Furmark it is possible to clearly demonstrate that a bottleneck is caused by the bandwidth, but this is the only program I have seen so far where this is the case. I can see the same FPS difference caused by a bottleneck in games compared to normal GTX 1060 setups, but the Afterburner PCI Util readout never goes above 60-70% in those. I would still say that the Afterburner readout is incorrect in most cases though, because there is a clear jump in performance when I use the exact same setup on an external display, so there is no other explanation for it. It might be worthwhile to look up how Afterburner measures the PCI bus utilisation, because maybe a portion of the data is not being measured for some reason. Edited September 18, 2016 by DaCM Quote Link to comment Share on other sites More sharing options...
kosowski Posted April 25, 2017 Share Posted April 25, 2017 On 27/8/2016 at 2:31 PM, DaCM said: Laptop Specs i7 3610QM 8GB RAM GTX 660m Windows 10 Anniversary 64bit 1080p screen eGPU Components Zotac GTX 1060 Mini 550W Chieftec PSU PE4C v3 mPCIe Installation Steps - Uninstall mobile nvidia drivers, reboot - Take out wifi card to free up mPCIe slot - Install PE4C connector into the slot (with the GTX 1060 and PSU already set up with the PE4C) - Reboot, install desktop nvidia driver, reboot At this point the 1060 was properly recognised by the system (no errors at all), without the need to use Setup 1.3. There is no need to use any of the delay switches on the PE4C either, it is simple plug and play. Hi! Sorry for using an old thread, have a question about it. So, without disabling the dGPU you had it working with standard BIOS or was it necessary to modify it in order to whitelist the card? I have the same laptop model and I'm thinking about doing the same . Thanks! 1 Quote Link to comment Share on other sites More sharing options...
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.