OS
Pool
ASIC Firmware
ASIC Hub
More
blog
Blog
News and Articles
forum
Forum
Top discussions, announcements, help
Show all articles
Referral Program
referral
Increase your income with Hive. Invite your friends and earn real cryptocurrency!
Get your referral link

Not enough PCI address space for a large BAR

AMD Cards
,
#1

Good afternoon: I have a significant problem and I haven’t been able to solve it, even after searching online.

Throughout 2025, I’ve been mining ERGO with my two rigs without much trouble. The MH/s were stable and there were hardly any issues.

But right on December 31st, both rigs decided to “die” and are no longer working.

I’ve tried:
1 - I have the latest version of HiveOS.

2 - I’ve updated my AMD drivers to the latest version.

3 - I’ve tried changing the cryptocurrency and the miner, but it still doesn’t work.

After extracting several logs through Hive Shell, I get the following:

  • Not enough PCI address space for a large BAR
  • BAR failed to assign

What happened to cause both my rigs to “stop working” simultaneously? Was there an update that I’m unaware of and haven’t been able to find online? Has anyone else experienced this problem and could offer some help?

Any help would be greatly received.

Happy New Year, and thank you very much.

Log:

Setup Miner…
Cuda driver detected.
Number of Cuda supported GPUs: 0
OpenCL driver detected.
Number of OpenCL supported GPUs: 0

Connecting to pool…
DNS over HTTPS resolve succeeded
Connected to es.ergo.herominers.com(152.114.194.200):1180 (TLS enabled)
TLS Handshake success
Subscribed to stratum server
New extranonce received: c46e
Subscribed to extranonce
New extra nonce received: c46e
Authorized worker: xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx.MinerTwo
New target received: 0000000100000000 (Diff 1)
New job received: 0 Height: 1692019 Target: 0000000112E0BE82…
Start Mining…
All devices deselected or failed compatiblity check. Closing lolMiner
lolminer exited (exitcode=0), waiting to cooldown a bit

#2

One for instance Cpu Mining has a HUGE Pages param in miners. These just mean how many threads you’re using. 128 threads is generally the Max HUGE Pages allocated. Typically it’s for more than one Physical CPU on the motherboard. also things like Memory timing only work on AMD.
So,

this depends on your gpu setup. to Re-Size the “Bar” the mem allocated inside the GPU. a quick AI chat gets you this:

" Resizable BAR (“Base Address Register”) is a PCI Express feature that changes how the CPU accesses your GPU’s VRAM."

GPU Architecture Resizable BAR Needs 4G Decoding Notes
Pascal (P104‑100) :x: No Optional Always 256 MB BAR
Turing GTX :x: No Optional No ReBAR
Turing RTX :heavy_check_mark: Yes :heavy_check_mark: Yes First NVIDIA ReBAR GPUs
Ampere :heavy_check_mark: Yes :heavy_check_mark: Yes Full support
Ada (RTX 40) :heavy_check_mark: Yes :heavy_check_mark: Yes Full support
GCN / Polaris / Vega :x: No Optional No ReBAR
RDNA1/2/3 :heavy_check_mark: Yes :heavy_check_mark: Yes Full support

:jigsaw: 2. Architectures that DO support Resizable BAR

These GPUs can expose their entire VRAM to the CPU (instead of 256 MB chunks):

:heavy_check_mark: NVIDIA

  • Turing RTX (RTX 2060/2070/2080/2080 Ti)
  • Ampere (RTX 3060/3070/3080/3090)
  • Ada Lovelace (RTX 40‑series)

:heavy_check_mark: AMD

  • RDNA1 (RX 5500/5600/5700 series)
  • RDNA2 (RX 6000 series)
  • RDNA3 (RX 7000 series)

These GPUs require:

  • Above 4G Decoding = ON
  • Resizable BAR = ON
  • CSM = OFF

:jigsaw: 1. Architectures that DO NOT support Resizable BAR

These GPUs are permanently limited to a 256 MB BAR window:

:x: NVIDIA

  • Kepler (GTX 600/700)
  • Maxwell (GTX 900)
  • Pascal (GTX 10‑series, P104‑100, P106‑100, P102‑100)
  • Turing GTX (GTX 1650/1660 series)

:x: AMD

  • GCN 1.0 → 4.0 (HD 7000 → RX 500 series)
  • Vega 56/64
  • Radeon VII

Hope this answers your questions.

#3

Thanks for your answer. I downgrade the version to 202312 and it works again.

#4

when you run a resize bar gpu you’ll notice the differences are with mem. the whole vram in one go on a gpu with no resize bar is really easy to achieve. just set the on board igp peg cache l1 l2 l3 switch to max size. enter into the os the settings for every 1gigabyte of GPU RAM times 2 gigabytes of CPU SRAM. every time the gpu requested mem size is a 256 bit chunk. ultimately the bandwidth for pcie 3.0 is x16 4gigabytes per second.

1gb ram gpu
2 gigabytes ram cpu
4 times 256 bit chunk mem swap
4 gigabytes max

max on board sram in Consumer computer systems running mining hardware is around 128GB CPU RAM

highest blockchain dag size is around 16gb

4/((4*256)/(16gpux2cpu))
=32 gigbytes
with a page size of single or double core modes. 1 or 2 or 3 etc. core and engine multiplies into threaded mem resource preloaded for cpu ready gui aware services.
32 times 2 (two) core in one cpu) is 64 gigabytes of VRAM.

And it has nothing to do with have resize bar on. it does something else.

instead of dumping gpu ram into cpu sram in 256 mb chunks. it dump all. in staging for better branding context user experiences it’s a shortcut.

rebar on is a shortcut when you can’t set it your self.
not hard. sram and vram are not the same as cache on a chipset.

the thing is virtual.

there is more to it.

increased size cache of your local user machine Memory and Address Space Limits
virtual space of a 32 bit process is user 4 GB X86
virtual space of a 64 bit process is user 128TB 64-bit
V-space of 32 bit for kernel is 4GB X86
V-space of 64 bit for kernel is 128GB 64-bit
non paged pool user and kernel 16GB
system cache user and kernel 16GB

it just matters if you have a single or double or more core and engine count or multipliers.