Bitcoin Forum

I'm really stuck trying to diagnose this problem.

The problem is straightforward to describe.  I have 6xVEGA 64s.  I go to Wattman to enable/adjust HBCC for each card.  When I click "apply" the screen will flicker once and the setting will take.  However for 2/6 of the cards it just crashes Wattman.  When I open Wattman up the setting remains unchanged for that card.

At one point all cards were working fine.  It was not until I rebooted that I got into this state and now I cannot get out of it.

It gets a bit weirder.  I've tried the following on both the latest blockchain drivers and the latest non-blockchain drivers. 
- Safe Mode
- DDU, remove all (also tried same steps with AMD Clean Utility)
- Boot with 1 GPU
- Install drivers
- Shutdown, connect another GPU, boot, wait for drivers to see new card
- repeat last step until all 6 cards are working

Again 2 cards will unable to adjust HBCC.  I get no error message, no details as to why, the setting just refuses to adjust for two of the cards.

I am running:
- Windows 10
- 32GB RAM
- 64GB Virtual Memory (required for xmr-stak-amd setup getting 1900H+).


Question:
- Is there an error log AMD spits out I can try and look at?
- Windows records no event logs when HBCC adjustment causes Wattman to crash.
- Bios issue?
- What if anything can I do to continue to diagnose this?

Any other info I can provide to help others help me?

TLDR; Help me please! My rig is short 1200H because two cards won't enable HBCC.

More info.

So the two problematic cards are plugged into PCI-E 3.0 1x slots (#2 and #3 on my mobo in order from top to bottom).

So I swapped a working HBCC card with one of the ones that I cannot enable HBCC on.

It seems whichever two cards are plugged into these PCI-E slots are problematic as the swapped card now cannot enable/adjust HBCC.

So I verified this a few more times.  It appears whatever two cards I have plugged into the #2 and #3 pci-e slot cannot enable HBCC. 

This problem also goes away entirely if I just run 4 GPUs, then those slots are fully functional.

It can be that HBCC technology only supported by modern cpu's pci-e lanes (ryzen, modern intel) and can have problems if used with chipset pci-e lanes ... just a guess ... I've read something from AMD documentation ...

Hmmm.  I suppose this is some important info given your response.

Mobo:
MSI Z270-A Pro ATX LGA1151

CPU:
INTEL® CELERON® Processor G3930

Perhaps the CPU is the problem?

I think your cpu is not a problem. The problem can be the way how motherboard bios or windows decided to connect you 5-th or 6-th gpu to the system - directly to cpu pci-e lanes or to southbridge switch.

I don't know if it is our case but you can read here https://github.com/RadeonOpenCompute/ROCm
Search for Supported CPUs chapter

So reading up on ROCm support CPUs I noticed it claims to require Gen3 PCIE. 

My bios I had set to Gen 2.  I changed it to Gen 3.  Did not help. 

Given at one point both these cards worked.  That one point was the very time I booted with them plugged in. 

So it is very likely a Windows issue.  I should try a complete clean install. 

Is there anyway to check if the pci-e lane is directly to cpu pci-e lanes or to southbridge switch?  I examined Windows System Information and could not tell from there.

you tries use Linux?
on linux i have autooverclock and autofan on nvidia
i heard AMD-gpupro driver is good work on Vega now

More info.

I followed the power table mods seen here (https://bitcointalk.org/index.php?topic=2002025.msg22158580#msg22158580)

It worked on 4 of the 6 cards.  The same two cards when I go hit "reset" after reboot retain the default stock settings.  All other cards are functioning nicely with those power table mods.

So, the same two cards refuse to read their registry settings from the same location as all the others.  Definitely something up with Windows I think.  I might try Linux.

Hey there,

just registered here because I got the exact same problem with 6x VEGA 56... 4 running with HBCC and 2 not.. AMD Settings crashes every time after activating hbcc on any card.

Let me know if you can figure something out :)

Would love to use linux, because I hate that gui stuff on windows, but as far as I know there is no HBCC support on linux yet...

What effect does the HBCC have on mining? From everything I read it is a technology with no real use. In lab tests they had to absolutely destroy the card running 8k resolution benchmarks to even see HBCC in use. In real world gaming and workstation applications it is 100% useless.

The effect on xmr-stak-amd is 1300H/S vs 1900H/S, big deal.. :(

Yup, exactly that.  I am losing 1200H because 2 cards won't enable HBCC.  I mean not the end of the world but I hate having problems I cannot solve.  

My next move is likely a clean install of Windows.

Given I tried OLDComer's power table mod and it took perfectly on 4/6 cards.  The same 2 cards that I cannot enable HBCC on simply ignore it.  The registry section for those 2 cards has the soft power table entry and it is correct.  Yet those 2 cards just reset to AMD Stock settings when I hit reset in Wattman while the other 4 jump to OLDComer's settings.  

Clearly this is a clue or related to the same HBCC problem (or an annoying coincidence).

So I am thinking this problem is solvable.  I am thinking Windows has done something wrong with the card.

That's why I hate Windows... this randomness sucks  :-[
Please let me know if it can fix the problem... :)

If not we have to wait for HBCC support in the linux drivers I guess...

More info.

I got a dummy hdmi plug.  So far my VEGA 64s have not needed one.  They always run fine monitor or not.

While xmr-stak-amd was mining I plugged a dummy plug into one of the two problematic cards.  Everything froze, screen went black, flickered a bit, then came back.  xmr-stak-amd looked stalled but eventually came back to life.

Check hash rates.  The one card I plugged the dummy plug into was hashing at 1900H but only for ~10sec!!  Then it dropped back down.

So, I figure lets reboot with the hdmi dummy plug connected from the start.  If I do this, I cannot enable HBCC on any card as the machine always crashes. 

So the best I can get now is to get everything all setup for 1900H (clock speeds, HBCC toggled).  Then once xmr-stak is running I can shove a dummy plug into a card hashing low and get ~10s of 1900H before it drops down to 1300H. 

Ok, weird... Thanks for the info, I'll try tomorrow with a second monitor - let's see what happens...  ???

Exact same issue, when plugged in PCI-E 1x any alteration to HBCC even though it is shown in AMD Control Center has not effect. Hash is 1300 instead of the 1900 it should reach.

I have also run into a bug that is preventing me from installing 6 Vega on a MSI Z97 Gaming 5 board.

I have substituted it for an R9 290 and the BSOD on driver installs for the 6th card have went away.

Seems like 4x Vega Rigs are trouble free though.

Does anyone know of a way to compare the PCIE lane setup across GPU's or at least how Windows sees it?

According to GPU-Z they are all PCIE 3.0 16x.  I need more details than that. 

Has anyone managed to figure this out yet? I'm thinking about adding 2 more Vega to my rig of 4 (and fill out the 2 empty slots in this server case).

I'm guessing from this post I absolutely should not be doing that?

I have been unable to locate a solution. 

More info though.  If I do a completely clean install with 1 GPU connected.  Then boot with all 6.  Disable crossfire.  I can enable HBCC on 5/6 cards before reboot.  Reboot once and it is 4/6 cards forever.  Given we have to reboot to install the soft power tables I don't even consider this a half a solution. 

I am sure the problem is addressable by AMD. 

For now mine monero with four and mine something else with the other two.

How do you make them mine monero with 4 and something else with the other two?

A friend with 5 found a solution, that is working (at least for him). I am going to try it with the next restart.
After reboot - change virtual ram size. In his case he changed from 90 to 80. But I guess any change would do the trick. And then move sliders and confirm the HBCC size.

For me it worked with 5 once, when I upped the virtual ram from 64 to 80. That info I got from him have not tried yet, as I am very busy today, but will do with the next restart.
In any case, to make 6 work, it seems we need 96 or more virtual ram, and I simply don't have that much space on the SSD. Someone with bigger SSD should try it.

Isolate the two bads cards, launch Claymore with -di flag to only use those two cards.


You need ~8GB per card of virtual memory.  Just launch xmr-stak-amd with 1 GPU and watch your virtual memory usage jump.  It is linear with each card you add.  Also the virtual memory only appears important when launching xmr-stak-amd.  I will try messing with virtual memory settings and restarting my miner when I get a chance to see if that helps.   With 4xVEGA 64 @ 1900H I need 32GB Virtual Memory.  When I add the other two cards (that refuse HBCC settings) my virtual memory needs hit 48GB.  So it appears all cards are indeed requested 8GB extra memory with our intensity settings however 2 of them are stuck at the low 'non-hbcc' hashrate of 1300H.

TLDR; I don't think we need 96GB of Virtual Mem.  I believe it is 8GB per card.

Well, the 5th card had never worked for me before I set it from 64 to 80 ... So I don't know. You may be right, you may be not :)
A thing to note - the ONE time it worked for me, I had disabled onboard video, and connected a monitor to one of the cards.
The guy who has these 5 working, told me that he has all of the other 4 connected one to another with HDMI cables, and the 5th to a monitor. Still after restart it worked for him on 3 only. After restart and change to vram size, all 5 came back online at full speed.

Btw - about crossfire - I noticed, that there is a setting for it only when the card is connected to a Monitor.
But I noticed strange thing - the first card had it ON. I disabled it. Then connected another two cards and they had both off. Have not tried on the others after it. Is that normal?

Thats why these are called beta drivers. ;)

BTW, anybody tried driver clean reinstall with all cards plugged in?

Is it better for Vega to use the new 17.9.3 drivers if you have 4 vegas and 2 rx580. It is impossible to get the HBCC working on more than 2 vegas even having 16 Gb RAM + 80Gb virtual mem with the blockchain drivers...

On blockchain drivers, 8GB of system RAM, 48GB of Virtual mem.  I can enable HBCC on 4/6 Vega 64's without issue.  I can in rare cases after a full clean reinstall get 5/6 cards to work.  RAM and VMEM is not the issue here.  Shot my Virtual mem up to 96GB and my system mem up to 32GB.  Same outcome. 4/6 can HBCC, 2 cannot.

I'm just getting into the vega+cryptonight scene.. So is this issue about only 4 out of 6 vega gpus having low hashrate specific to a few users or ALL windows users currently?

probably all. including me too.

I once had 5/6 hashing at around 1800/card, but I couldn't repeat that ever again.
I'm back on normal driver now on ethash, I got fed up with the buggy BC driver.

BTW I could turn on HBCC for all 6 cards, but only 4 of them had the slider.

I was thinking about buying some vega 64 or 1080ti how does the vega hash on other algo's

The issue is 2/6 cards are hashing monero as if HBCC is disabled (so ~1300H).

4/6 cards are hashing proper (~1900H).

At the moment I am just not mining monero with the 2 that hash low due to HBCC not enabling.

This issue appears widespread.  Other forums have other people complaining about the same.  

The best I've had is 5/6 working but only after a full clean install, first reboot after the clean install and I am back to 4/6.

I should probably move to AMD forums and start asking them for help.  The right people might see it and fix it officially.  I am positive it is a driver issue and can be addressed.

anyone know if it's possible with 1 rig to have some amd (vega) gpus mining xmr and also have nvidia gpus mining equihash? thinking of going this route as a bandaid for this HBCC problem

Good thing when I ordered parts for my new rig I grabbed a 500 GB SSD which I never do. Thanks to all in this thread you may have saved me a headache.

Ok 8 Vega 56 and even tho it says hbcc is working on 6 it's not. Tested 3 and hashes at 1840 per with 8 they hover around 1350 for all. This is more than likely a combo of amd drivers and perhaps pcie bios settings or lane limitation. Someone more creative than us hopefully figures it out and shares.

Yeah no luck fixing this so far.  Has nothing to do with Virtual mem or ram at this point.  8GB of ram to enable HBCC.  8GB of Virtual mem per card to run xmr-stak-amd with the intensity settings to hit 1900H.

Now something weird I've noticed.

Roughly every 24 hours of mining the two cards that refuse to enable HBCC and hash at 1300H suddenly hash at 1900H for ~10min then drop back down.

I am going to try to enable to web client for xmr-stak-amd and write a scraper to monitor the hashrate more closely than I can by randomly checking up on it.  Also appears you can have it write the hashrate out to a file, that might be easiest.

The only correlation I can find for the random gain of +1200H once a day is it appears to time with

Just to be clear with 6xVEGA 64.

4@1900H + 2@1300H (due to HBCC not enabling), total of 10200H.  Randomly every day it jumps to 11500H for a short period.  The only correlation I can find in the system logs is Windows Defender Updates.

Very strange.

I've seen that too, as randomly my max hashrate is 11500 instead of the usual 10000 that I am on. But no idea if that was for a second, minute or more.
The worse thing about this shit is that you cannot combine 4 vegas with other AMD cards, as it still does not work afterwards on all 4. So either 4 vega or you should add nvidia cards, but that does not really sound like a realistic option, as currently nvidia cards are too overpriced.

Or maybe , the whole thing is a visual glitch, and the 5 cards where I see a HBCC slider enabled, are indeed hashing at around 1900 ea. As how can you explain this? http://prntscr.com/gwokm2
I've had such values , over the 10000-10020 that I see on the console the whole time since my last restart 5 days ago. And on top of that I have only 96% good results .

Same with 96% good result. 

Your screenshot basically mirrors my exact avg hash rates on nanopool, ~10500.  My reported is almost always 10200 but on nanopool it always averages 10500 or higher. 

I imagine we both have the same setup using the soft power table mod.

check the web stats of the xmr stak using another device on your network, not the rig itself, so you can read the values real time without touching the console to see whats up. I have not done taht yet, as I need to have physical access to the router.

we have a bit different settings for sure, but its all odd ... I have modded 5 cards - 1220/1100. The sixt is manually adjusted with wattman to 1220/1030 , but its showing even better results, drawing probably more power, but it is what it is.
Also, I am at 1800/1600 intensity, as the thing was crashing after a while on 2016/1600, so I moved it a bit under, and its stable for around a week like this.

Have you tried using some ridiculous amount of VM like 256GB, or say 128GB or real memory ?
Would be interesting to know if it makes a difference.

do you try the new driver 17.10.2? probl. the problem is gone?

I have just tried this with the same outcome, 4/6 cards will enable. What I have noticed is that 2/4 of the cards display as "unknown" under the PCI Bus in System whereas 4/6 show as PCI-E v3.0. So the cards/driver cannot read the PCI bus properly?

https://bitcointalk.org/index.php?topic=2315166.0

Even better, a true community hero here:
https://bitcointalk.org/index.php?topic=2002025.msg23619746#msg23619746

i donno bout you guys but i use this.

https://www.monitortests.com/forum/Thread-Custom-Resolution-Utility-CRU

restart the graphic driver by running restart.exe file and bam!hope it helps  ;D

Ive been unsuccessful running more than 4 cards per rig.

My solution was to buy more motherboards and run 4 cards per rig. This has increased my cost, but there hasn't been anything yet i've tried that has worked.

I've had as many as 8 in a rig working fine, but HBCC enabled only on 4 of them, and i've tried this across multiple motherboards, different things.


I have one potential solution i will try when i get time to take a few rigs offline for a few hours. We'll see...

Use this: https://bitcointalk.org/index.php?topic=2315166 or write it yourself