This is a re-post of a thread topic that apparently did not survive the recent forum migration.
I just received a replacement NUC7i5DNHE from Intel (3 days late: thanks UPS) and not only is the original reported issue still present, this NUC now reproduces additional issues already reported by other forum members:
The replacement unit came shipped with 0052 BIOS, and the NVMe half speed issue was already present. Updating to latest 0059 BIOS made no difference. I even tried power button F4 BIOS recovery but it didn't appear to work. I did a F7 BIOS update to 0059 and none of the above issues changed.
Going through this replacement procedure also cost me:
It has become apparent that the original issue I reported is a BIOS bug, not a HW defect. Intel did not diagnose this correctly, so replacing my NUC7i5DNHE unit not only did not fix the issue, now I've got additional issues that may be actually related to the hardware that did not exist before. I regret having gone through the replacement procedure, and now I am extremely displeased.
I'd like to request Intel to ship me back the exact NUC7i5DNHE unit I returned, and I will ship this replacement unit with more problems back to Intel.
Why did you lose a Windows 10 PRO license? You could have reactivated the original license on the replacement hardware. The only case where the license will not reactivate on replacement hardware is if the original system was upgraded from Windows 7, 8 or 8.1 - and we know this wasn't the case because these O/S were never supported on this hardware.
Forget about the Win10 license. Also forget about the HDD activity light as I got it working by re-configuring the power button LED.
Now I just want the reduced link speed for NVMe PCIe SSD fixed.
It's pretty obvious I received a refurbished unit from Intel, with an older motherboard rev. than the returned unit. Looking carefully at the PCIe SSD throughput (I use "winsat disk" in elevated command prompt) it has become clear that the actual link speed (between SSD and PCIe host controller) is Gen 2 x4, instead of the expected Gen 3 x4.
I have tested the following NVMe PCIe M.2 SSDs with this NUC7i5DNHE, and they all have less than 1800MB/s sequential read throughput:
Performance of the Samsung SSDs is clearly bottlenecked by the Gen 2 x4 link speed negotiated for the M.2 2280 slot. Visual BIOS continues to report Gen 3 (8GT/s) x4 with these SSDs, but I suspect this displayed speed is the one reported by the SSD, not the actual negotiated link speed. Gen 2 x4 has a max throughput of 20GT/s, which correlates to actual throughput measurements being capped at ~1800MB/s. This is an out-of-spec situation for the NUC7i5DNHE (Dawson Canyon series) that needs a fix.
I can't tell whether the fix to restore PCIe Gen 3 x4 link speed for supporting NVMe SSDs should be in BIOS or hardware, but it should be one of these two. Since the returned NUC7i5DNHE unit had no trouble running NVMe SSDs at Gen 3 x4, I suspect the difference may be in hardware. I can provide SA numbers for returned and replacement units if it helps.
Intel, your move next.
Attached are SSU reports and Samsung Magician performance benchmark and WinSAT disk runs for the following 3 NUC / SSD configurations:
The exact same 970 PRO 512GB PCIe NVMe SSD was installed into the NUC7i5DNHE and NUC7i5BNH, along with the exact same pair of Samsung 4GB 1Rx16 PC4-2400T-SC0-11 DDR4 SODIMMs (M471A5244CB0-CRC). Installed OS is Windows 10 1803 for 970 PRO 512GB SSD and Windows Server 2019 for 960 PRO 2TB SSD,
The NUC7i5DNHE (with 0059 BIOS) gets sequential read & write numbers less than 1800MB/s with either Samsung SSD. In contrast, the NUC7i5BNH running the same 970 PRO SSD gets a sequential read number in the 3500MB/s range.
NUC7i5DNHE with Samsung 970 PRO 512GB SSD:
NUC7i5DNHE with Samsung 960 PRO 2TB SSD:
NUC7i5BNH with Samsung 970 PRO 512GB SSD:
The NUC7i5DNHE returned to Intel had a motherboard SA # of J57626-508. This NUC fully supported PCIe Gen 3 x4 speed with NVMe SSDs.
The NUC7i5DNHE replacement unit I received has a motherboard SA # of J63721-503. This NUC runs all my NVMe SSDs at PCIe Gen 2 x4 speed.
I have some additional data to report.
I recently bought a NUC7i5DNBE bare board that came shipped with 0044 BIOS. This is my second Dawson Canyon NUC.
The BIOS of this board was updated with the following sequence:
0044 -> 0052 (jumper method) -> 0053 -> 0055 -> 0056 -> 0057 -> 0058 -> 0059
When doing the 0052 BIOS update with jumper method, an additional item is displayed: "Descriptor area". This appears to correlate with the PCIe speed fix for NVMe SSDs.
This NUC board now runs installed NVMe SSDs at PCIe Gen3 x4 speed, so I get sequential read speeds of up to 3500MB/s with my Samsung 970 PRO 1TB SSD.
In contrast, my first NUC7i5DNHE NUC, which is a replacement unit shipped to me by Intel, apparently missed out going through the 0052 BIOS update with jumper method, so even with latest 0059 BIOS installed, it is stuck running PCIe NVMe SSDs at Gen 2 x4 speed.
Unfortunately, starting with BIOS 0053 there is a BIOS rollback lockout that prevents BIOS from being rolled back to 0052 or earlier, due to an Intel ME firmware update addressing a critical security issue. Because of this, I was only able to roll the BIOS back to 0053, but not 0052. All attempts to flash back to 0052 using jumper recovery method have failed thanks to the rollback lockout.
From what I can tell, there are only two realistic solutions to restore the M.2 2280 slot speed to PCIe Gen 3 x4 for NVMe SSDs:
The speed fix in BIOS 52 should be there in all subsequent BIOS releases. Installing any of them using the Recovery Method should accomplish the same thing. Have you tried this?
That was actually the first thing I tried, but for some reason it didn't work. That misled me to believe the fix was only put into BIOS 0052 which in hindsight didn't make any sense.
I just attempted flashing BIOS 0059 again with jumper removed, this time with one difference: the M.2 NVMe SSD was removed first. Just checked the sequential read speed of a re-installed Samsung SM961 512GB NVMe SSD, and it's back to ~3300MB/s, so it looks like the PCIe Gen3 x4 speed has been properly restored for the M. 2280 slot! 😊
Intel, please consider this throughput issue with PCIe NVMe SSD satisfactorily closed.
The other issue of "Disabling PCIe M.2 SSD slot also disables I219-LM Ethernet" remains for Dawson Canyon, though, but it's something I can live with.
Thanks again, Scott, for the help!
Hhmmm... Yes, it would appear that there is a path to a working solution, but removal of the SSD should not be necessary. I think this should go back to the development team for analysis. If this is the only path to a solution, I would like to understand why and I would like to at least see this documented somewhere (like BIOS release notes).
BTW, I also took my NUC6i5SYH (Swift Canyon) through the exact same recovery procedure to regain PCIe Gen 3 x4 speed for the M.2 2280 slot. This unit was the only one out of three that had depressed throughput for NVMe SSDs. Some time ago, I went through the jumper recovery procedure with this unit but the throughput didn't go back up. Then, yesterday, after achieving success with my NUC7i5DNHE, I tried the same procedure with this NUC6i5SYH, also with the NVMe SSD removed first, and it worked successfully as well.
So we have some empirical data suggesting the jumper recovery method is ineffective in restoring PCIe Gen 3 x4 speed of M.2 2280 slot for two different NUC models, unless the installed NVMe SSD is physically removed first.
I believe this behavior should not be too hard for Intel (development team) to reproduce. Just start with a Dawson Canyon NUC with something like 0044 or 0048 BIOS. Attach an NVMe M.2 SSD that talks PCIe Gen3 x4, boot to Windows and measure the depressed SSD sequential read throughput, do the jumper recovery procedure to update to latest BIOS with the SSD still attached, then re-measure the sequential read throughput for any improvement. If no change, remove the SSD and go through the jumper recovery procedure again, then measure throughput again.
I completely agree that this requirement (of removing the PCIe M.2 SSD) should be documented especially if it can be confirmed as being necessary.
Agreed, if this works (will have to try it for my case that was somehow garbled on this site), then it should be in the notes to prevent a lot of customer hassle. I also had done the jumper recovery post 0052 and it did not fix the issue, but I did not remove the nvme. Will report back.
UPDATE - it does work with the nvme removed. Please update the BIOS notes to reflect this and save yourselves support cases.
I have reviewed the latest updates from Intel.
Regarding issue #1 (Disabling PCIe M.2 Slot also disables the LAN adapter) I still do not know why we have different repro results. I now have two NUC7i5DNHE units running the same BIOS that repro this issue with 100% consistency.
For issue #2 (Running BIOS update (Jumper procedure) with the SSD attached doesn't update the PCIe Gen 2 speed), I can no longer repro this issue, as ever since I succeeded in getting the link speed back up to PCIe Gen3 for the NVMe SSD, it has not gone back down with subsequent normal (F7) BIOS updates, even without removing the installed PCIe M.2 SSD first. I think any Dawson Canyon unit that does not repro the PCIe Gen2 speed limit for PCIe M.2 SSD is not a good candidate for investigating this issue. If a unit is already demonstrating PCIe Gen3 speed, then it is free from this issue and no amount of flashing (normal or jumper) will drop the the link speed back to PCIe Gen2. A unit has to be "stuck" at PCIe Gen2 speed to be a good candidate for the BIOS update testing.
I can believe that issue #2 may be related to the PCIe M.2 SSD brand/model I'm using, specifically Samsung models 970 PRO, 960 PRO, 950 PRO, etc.
In any case, ever since I discovered a workaround (by first removing the installed PCIe M.2 SSD prior to jumper-based BIOS update) my NUC7i5DNHE units no longer suffer from issue #2, so at least for me this issue can be closed. Also, as mentioned previously, I can live with issue #1 not getting fixed so it can be considered closed as well.
Thank you Intel support team for your attention and assistance.
Hello to all!
We had 17 NUC's, all with problem regarding the issue #2 - SSD running on a PCIe Gen 2 Speed. We opened a thread hoping to solve the issue. The thread was lost during forum migration, and was never brought back ( https://forums.intel.com/s/question/0D50P000049OkncSAC/missing-discussion-performance-problem-with-s... ). The issue was never resolved.
If you are unable to bring back the discussion which has all the info the community provided inside, please don't expect from us to go in details all over again. We already started to RMA NUC's, as this was easier solution in the end.