Jump to content

sNullp

Registered User
  • Posts

    12
  • Joined

  • Last visited

About sNullp

  • Birthday 03/21/1984

Recent Profile Visitors

The recent visitors block is disabled and is not being shown to other users.

sNullp's Achievements

Settling In

Settling In (2/7)

11

Reputation

  1. My L502x + PE4C + GTX970 setup has some annoying boot up problem, so I spent some time doing experiments and some research, here's what I found regarding these two special signal lines. Basically they are related to how a motherboard initialize a device. Per https://pcisig.com/sites/default/files/specification_documents/MiniEx_M2_ECR_PERST_CLKREQ_Power-up-Reqts_20140918_Final.pdf, we can learned that for a normal motherboard + pcie device, when you press the power button the 3.3V power is firstly established. When the device probes the 3.3V power, it will issue CLKREQ# by lowering its voltage to ground. When the motherboard detects a CLKREQ, it will try to establish the reference clock, then issue PERST# by raise it to 3.3V to let the device know it. After that the device will pick the reference clock and start communicating with the motherboard. For PE4c semantics (http://www.bplus.com.tw/PDF/PE4C_V30_20150129A.pdf), the CLKREQ# is not connected to the GPU at all. It will simply connected to the CLKREQ# delay circuit, which, allowing you to choose how many seconds you want before it lower the wire to ground. However, for PERST#, it connects the motherboard, the GPU and the PERST# delay circuit all together. The PERST# delay circuit will try to raise the signal only after your specific delay time, but it is easy to realize that when the motherboard raise the line to 3.3V first, it doesn't matter anymore how PERST# delay circuit behave. The problem my laptop (L502x) has is, its motherboard doesn't 100% honor the PCIe standard. More specifically, it uses a global PERST# line, so it will be raised to 3.3V after about 0.5s after power up, globally, including the port you used to connect GPU. Moreover, its BIOS is buggy that it can't support external GPU during boot so enabling GPU while booting will hang the laptop. Also, raising PERST# before CLKREQ# goes to ground is an undefined behavior, and will result the GPU freeze and becoming undetectable. So it looks like a dilemma: If we don't delay any signal, the GPU will be correctly connected upon power up, then BIOS hangs. If we delay CLKREQ# using the on-board circuit, the laptop will still issue PERST# 0.5s after power up so the GPU will freeze. If we try to change PERST# delay switch, I'm sorry, the laptop motherboard is faster than the delay circuit so nothing will change at all. However, right now you might also come up with the solution yourself. If we simply put a small piece of electric tape, on line 22 on the bottom side (PCI Express Mini Card (Mini PCIe) pinout diagram @ pinoutsguide.com) of the mPCIe connector, which blocks motherboard's PERST# signal. We can successfully enable PE4C's PERST# delay circuit. Then, pick a combination that CLKREQ# will goes down before PERST# raise up, you are done! Now the GPU would be detected automatically upon boot without any hot plugging! ps. PE4L has similar PERST# delay circuit implementation, its CLKREQ# is named as DETECT# but it doesn't offer any delay circuit. The 1x/(2x,4x,8x,16x) switch would either connect CLKREQ# to the card, or to the ground at the very first.
  2. I performed more tests even including try similar things on a PE4L. Basically PE4L is indeed a subset of PE4C. They behave similarly on hot plugging and both hung on boot when CLKRUN delay is off. As a conclusion, the more reliable way to workaround the problem is still a delayed power link, not the CLKRUN delay. To explain this, I have came up with a theory: upon boot or hot plugging pci bus will send something to GPU to correctly initialize it, and this signal will also be blocked by the CLKRUN delay if set. For GPU side, power up + no init signal will result in a locked up state, in which PCIe soft reset won't help. So what we need is this initialization signal (can't be re-issued using setup 1.x). However accepting the signal on boot is impossible because without CLKRUN the laptop will hang. So the only option left is hotplugging (also triggers the initialization signal), and it is easier to hotplugging the DA-2 side rather than mPCIE side. Update: I finally solved the problem perfectly using a tiny piece of electric tape, please see http://forum.techinferno.com/enclosures-adapters/10812-pe4x-series-understanding-clkreq-perst-delay.html#post142689 for details!
  3. Currently I experienced no problem even if I OCed 150mhz off the GPU. So I'm not considering this hacking at this step. I'm more eager to solve the laptop blank boot screen problem if DA-2 is connected to the PE4c. See my comments in the PE4C thread.
  4. Try hot resetting and initializing eGPU port before pci compaction in setup 1.x.
  5. More experiments showed that although my BIOS would disable the pcie port if idle, that is not the root reason why the switches won't work. Let's assume BIOS side is okay as I found a way to force them enabled. However, I found that if DA-2 is connected and switches are at any position that won't hang, when I power up the laptop, it was the GPU side - not the laptop side, would have the problem. Basically that will result GPU fan running at maximum speed and undetectable (even you disconnect DA-2 and replug again). I speculate when machine boots up it sends some special command that makes the board or the GPU stuck at this weird mode. Only hot plug the pcie end would solve the problem (disconnecting pcie will shut down GPU side).
  6. LOL, I could manually plug DA-2 without this . But what I want is something without manual involvement. I'm thinking of using a time delay relay on the power link.
  7. Tried but failed . For now the only way to make sure it will work is plug a valid pcie device(including pe4c with DA-2 unplugged) on that port upon boot and switch (hot plug) to GPU before setup 1.3 then do a pcie hot reset. It seems to me that it would be best if I can have a power switch on PE4C.
  8. Very unfortunately that doesn't work for me. It has two switches, SW1 and SW2. Changing SW1 has 0 effect to me, while, for SW2: position #1 (off) will hang the machine when boot position #2 and #3 would successfully let the machine boot, but the GPU would never be detected, unless I do a hot plug. I guess the reason behind is that my BIOS will automatically disable the port if no device is connected upon boot. But don't know how to disable this feature. Thoughts?
  9. I had similar problem on L502x, and I think I found the reason: The GTX 970 would show up with an incorrect hardware id when it is just powered up. If you power the card up, wait for at least 1 second, and reset the pcie port, you will see the hardware id change. After that it is guaranteed that you will got a 970 detected.
  10. Hi there, While PE4C has several advantages, how to correctly "hot plug" it like previous PE4L? My DELL L502x won't boot if eGPU is connected, so hot plugging is the only way. With PE4L, you could power the card up while not plug the mPCIe end into the computer. However with PE4C the DA-2 port won't turn green unless the mPCIe is connected to a running computer. This makes me nervous because I want to make sure power is running before I do hot plug to prevent hardware damage, but, can't find a correct way to do it with PE4C.
  11. Thank you, Dschijn! This makes sense to me that all data points suggesting DA-2 can drive a over-clocked GPU come from Europe. Waiting for more inputs here, though, to see which brand I should buy for my DA-2 + PE4C setup.
  12. I've done some research and found it is suggested that not letting the GPU exceed ~165W, which is calculated by 220W * 75% (DA-2 efficiency estimate). The post also mentioned down clocking and TDP throttling to avoid BSOD to confirms this theory. If this theory is true, it seems to me that only ZOTAC GTX 970 with stock configuration could fit the power requirement. However I found some inconsistency data points here: http://forum.techinferno.com/implementation-guides-apple/9821-2014-15-mbp-iris-gtx970%4016gbps-tb2-akitio-thunder2-win8-1-osx10-10-%5Bdschijn-2%5D.html (DA-2 can provide maybe 240W) http://forum.techinferno.com/implementation-guides-apple/10099-2012-15-macbook-pro-gtx970%4010gbps-tb1-akitio-thunder2-win7-osx10-10-3-%5Bbsohn%5D.html (200W + 12.5%) http://forum.techinferno.com/diy-e-gpu-projects/10124-gtx970-power-consumption-discussion-2.html (211W) http://forum.techinferno.com/enclosures-adapters/9426-220w-dell-da-2-ac-adapter-discussion-9.html (#89, ASUS GTX 970) So, I guess I'm slightly confused. If you also have a DA-2 + GTX970 configuration, could you post which brand you are using and are you down-clocking / throttle the maximum TDP to reach a stable behavior? Thanks!
×
×
  • Create New...

Important Information

By using this site, you agree to our Terms of Use. We have placed cookies on your device to help make this website better. You can adjust your cookie settings, otherwise we'll assume you're okay to continue.