vulkan: Tune GDN kernel, fix Intel performance by 0cc4m · Pull Request #29476 · ggml-org/llama.cpp

0cc4m · 2026-09-26T10:48:56Z

Overview

I noticed an opportunity for tuning and also that Intel's performance was terrible. This PR fixes that.

Test-backend-ops perf improvements

RTX 3090

ubatch	before	after	Δ
2048	1666	1562	−6.3%
4096	3381	3169	−6.3%
8192	6835	6513	−4.7%
16384	13827	13352	−3.4%
32768	29911	27975	−6.5%

AMD Radeon Pro VII was already optimal.

Intel A770

ubatch	before	after	speedup
64	679.7	73.7	9.2×
256	3478.6	282.3	12.3×
512	6959.3	661.6	10.5×
1024	13900.3	1322.1	10.5×
2048	27753.8	2630.8	10.6×
4096	55511.9	5313.2	10.4×
8192	111023.5	10437	10.6×
16384	222400.8	24863.7	8.9×
32768	445054.2	46746.4	9.5×

Model benchmark (Qwen 3.6 35B Q2_K)

Prompt processing, t/s

GPU	ubatch	before	after	Δ
RTX 3090	512	3926.8	3878.4	−1.2%
RTX 3090	1024	4584.7	4616.5	+0.7%
RTX 3090	2048	5121.1	5154.8	+0.7%
Pro VII	512	785.2	787.9	+0.3%
Pro VII	1024	951.1	952.6	+0.2%
Pro VII	2048	1026.5	1027.7	+0.1%
A770	512	686.6	865.6	+26.1%
A770	1024	791.7	1050.3	+32.7%
A770	2048	852.7	1174.4	+37.7%

Token generation, t/s

GPU	before	after	Δ
RTX 3090	~141	~142	flat
Pro VII	89.8	89.6	flat
A770	49.97	54.18	+8.4%

Requirements

I have read and agree with the contributing guidelines
AI usage disclosure: YES

SleepinDevil · 2026-09-28T10:36:51Z

Those performance increase metrics look great on the Xe, nice work!!

Is this expected to have some improvements on Xe3 as well?

I built this PR and benchmarked Qwen3.6-27B Q6_K & Qwen3.8-27B Q8_K_L (UD) and note a very slight PP regression on the Intel B70 running on Windows with this PR applied. I am on the latest driver: 32.0.101.8976

PR 29476 - Qwen 3.6 27B Q6_K

llama-bench  -p 8192  -n 128  -d 0,8192,65536  -r 3  -fa 1  --delay 10  -ngl 99  --device Vulkan1  -m ..\LLM-models\Qwen3.6-27B-MTP-Q6_K.gguf
ggml_vulkan: Found 2 Vulkan devices:
ggml_vulkan: 0 = NVIDIA GeForce RTX 3080 Ti (NVIDIA) | uma: 0 | fp16: 1 | bf16: 1 | fp4: 0 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: NV_coopmat2
ggml_vulkan: 1 = Intel(R) Arc(TM) Pro B70 Graphics (Intel Corporation) | uma: 0 | fp16: 1 | bf16: 0 | fp4: 0 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: KHR_coopmat

model	size	params	backend	ngl	fa	dev	test	t/s
qwen35 27B Q6_K	21.30 GiB	27.32 B	Vulkan	99	1	Vulkan1	pp8192	666.49 ± 1.38
qwen35 27B Q6_K	21.30 GiB	27.32 B	Vulkan	99	1	Vulkan1	tg128	20.85 ± 0.06
qwen35 27B Q6_K	21.30 GiB	27.32 B	Vulkan	99	1	Vulkan1	pp8192 @ d8192	406.09 ± 0.26
qwen35 27B Q6_K	21.30 GiB	27.32 B	Vulkan	99	1	Vulkan1	tg128 @ d8192	20.20 ± 0.02
qwen35 27B Q6_K	21.30 GiB	27.32 B	Vulkan	99	1	Vulkan1	pp8192 @ d65536	102.89 ± 0.09
qwen35 27B Q6_K	21.30 GiB	27.32 B	Vulkan	99	1	Vulkan1	tg128 @ d65536	16.60 ± 0.00

build: b266105b5 (11226)

PR 29476 - Qwen 3.8 27B Q8_K_L

llama-bench  -p 8192  -n 128  -d 0,8192,65536  -r 3  -fa 1  --delay 10  -ngl 99  --device Vulkan1  -m ..\LLM-models\Qwen3.8-27B-UD-Q8_K_L.gguf
ggml_vulkan: Found 2 Vulkan devices:
ggml_vulkan: 0 = NVIDIA GeForce RTX 3080 Ti (NVIDIA) | uma: 0 | fp16: 1 | bf16: 1 | fp4: 0 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: NV_coopmat2
ggml_vulkan: 1 = Intel(R) Arc(TM) Pro B70 Graphics (Intel Corporation) | uma: 0 | fp16: 1 | bf16: 0 | fp4: 0 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: KHR_coopmat

model	size	params	backend	ngl	fa	dev	test	t/s
qwen35 27B Q8_0	26.11 GiB	27.32 B	Vulkan	99	1	Vulkan1	pp8192	684.79 ± 3.12
qwen35 27B Q8_0	26.11 GiB	27.32 B	Vulkan	99	1	Vulkan1	tg128	18.11 ± 0.01
qwen35 27B Q8_0	26.11 GiB	27.32 B	Vulkan	99	1	Vulkan1	pp8192 @ d8192	411.94 ± 0.10
qwen35 27B Q8_0	26.11 GiB	27.32 B	Vulkan	99	1	Vulkan1	tg128 @ d8192	17.61 ± 0.00
qwen35 27B Q8_0	26.11 GiB	27.32 B	Vulkan	99	1	Vulkan1	pp8192 @ d65536	102.93 ± 0.03
qwen35 27B Q8_0	26.11 GiB	27.32 B	Vulkan	99	1	Vulkan1	tg128 @ d65536	14.79 ± 0.00

build: b266105b5 (11226)

Current source code build - Qwen 3.6 27B Q6_K

llama-bench  -p 8192  -n 128  -d 0,8192,65536  -r 3  -fa 1  --delay 10  -ngl 99  --device Vulkan1  -m ..\LLM-models\Qwen3.6-27B-MTP-Q6_K.gguf
ggml_vulkan: Found 2 Vulkan devices:
ggml_vulkan: 0 = NVIDIA GeForce RTX 3080 Ti (NVIDIA) | uma: 0 | fp16: 1 | bf16: 1 | fp4: 0 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: NV_coopmat2
ggml_vulkan: 1 = Intel(R) Arc(TM) Pro B70 Graphics (Intel Corporation) | uma: 0 | fp16: 1 | bf16: 0 | fp4: 0 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: KHR_coopmat

model	size	params	backend	ngl	fa	dev	test	t/s
qwen35 27B Q6_K	21.30 GiB	27.32 B	Vulkan	99	1	Vulkan1	pp8192	682.68 ± 2.50
qwen35 27B Q6_K	21.30 GiB	27.32 B	Vulkan	99	1	Vulkan1	tg128	20.83 ± 0.05
qwen35 27B Q6_K	21.30 GiB	27.32 B	Vulkan	99	1	Vulkan1	pp8192 @ d8192	411.53 ± 0.34
qwen35 27B Q6_K	21.30 GiB	27.32 B	Vulkan	99	1	Vulkan1	tg128 @ d8192	20.26 ± 0.01
qwen35 27B Q6_K	21.30 GiB	27.32 B	Vulkan	99	1	Vulkan1	pp8192 @ d65536	103.13 ± 0.09
qwen35 27B Q6_K	21.30 GiB	27.32 B	Vulkan	99	1	Vulkan1	tg128 @ d65536	16.61 ± 0.00

build: 4da6337 (11223)

Current source code build - Qwen 3.8 27B Q8_K_L

llama-bench  -p 8192  -n 128  -d 0,8192,65536  -r 3  -fa 1  --delay 10  -ngl 99  --device Vulkan1  -m ..\LLM-models\Qwen3.8-27B-UD-Q8_K_L.gguf
ggml_vulkan: Found 2 Vulkan devices:
ggml_vulkan: 0 = NVIDIA GeForce RTX 3080 Ti (NVIDIA) | uma: 0 | fp16: 1 | bf16: 1 | fp4: 0 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: NV_coopmat2
ggml_vulkan: 1 = Intel(R) Arc(TM) Pro B70 Graphics (Intel Corporation) | uma: 0 | fp16: 1 | bf16: 0 | fp4: 0 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: KHR_coopmat

model	size	params	backend	ngl	fa	dev	test	t/s
qwen35 27B Q8_0	26.11 GiB	27.32 B	Vulkan	99	1	Vulkan1	pp8192	704.18 ± 3.66
qwen35 27B Q8_0	26.11 GiB	27.32 B	Vulkan	99	1	Vulkan1	tg128	18.11 ± 0.01
qwen35 27B Q8_0	26.11 GiB	27.32 B	Vulkan	99	1	Vulkan1	pp8192 @ d8192	418.20 ± 0.07
qwen35 27B Q8_0	26.11 GiB	27.32 B	Vulkan	99	1	Vulkan1	tg128 @ d8192	17.60 ± 0.00
qwen35 27B Q8_0	26.11 GiB	27.32 B	Vulkan	99	1	Vulkan1	pp8192 @ d65536	103.39 ± 0.09
qwen35 27B Q8_0	26.11 GiB	27.32 B	Vulkan	99	1	Vulkan1	tg128 @ d65536	14.79 ± 0.00

build: 4da6337 (11223)

0cc4m · 2026-09-28T11:36:26Z

I don't have access to any Xe2 or Xe3 hardware, so I cannot tell you anything about that. I would expect them to benefit too.

TheBlueMatt · 2026-09-28T12:21:51Z

What's the test-backend-ops perf -o for this?

0cc4m · 2026-09-28T12:44:16Z

I didn't keep the tests, but it was just additional perf cases, something like this:

test_cases.emplace_back(new test_gated_delta_net(GGML_TYPE_F32, 16, 128,  2048, 1, 2)); // MMA PP-2048
test_cases.emplace_back(new test_gated_delta_net(GGML_TYPE_F32, 16, 128,  4096, 1, 2)); // MMA PP-4096
test_cases.emplace_back(new test_gated_delta_net(GGML_TYPE_F32, 16, 128,  8192, 1, 2)); // MMA PP-8192
test_cases.emplace_back(new test_gated_delta_net(GGML_TYPE_F32, 16, 128, 16384, 1, 2)); // MMA PP-16384
test_cases.emplace_back(new test_gated_delta_net(GGML_TYPE_F32, 16, 128, 32768, 1, 2)); // MMA PP-32768

TheBlueMatt · 2026-09-28T15:28:19Z

On a B65:

before:
  GATED_DELTA_NET(type=f32,head_count=32,head_size=128,n_seq_tokens=1,n_seqs=1,v_repeat=1,permuted=0,kda=0,K=1):              129056 runs -     7.93 us/run -     4160 kB/run -  500.60 GB/s
  GATED_DELTA_NET(type=f32,head_count=16,head_size=64,n_seq_tokens=1,n_seqs=1,v_repeat=1,permuted=0,kda=0,K=1):               343980 runs -     2.96 us/run -      528 kB/run -  170.44 GB/s
  GATED_DELTA_NET(type=f32,head_count=32,head_size=128,n_seq_tokens=1,n_seqs=1,v_repeat=1,permuted=0,kda=1,K=1):              112490 runs -     9.24 us/run -     4176 kB/run -  431.11 GB/s
  GATED_DELTA_NET(type=f32,head_count=32,head_size=128,n_seq_tokens=64,n_seqs=1,v_repeat=1,permuted=0,kda=0,K=1):               8178 runs -   135.40 us/run -     8208 kB/run -   57.82 GB/s
  GATED_DELTA_NET(type=f32,head_count=32,head_size=128,n_seq_tokens=256,n_seqs=1,v_repeat=1,permuted=0,kda=0,K=1):              1634 runs -   617.03 us/run -    20544 kB/run -   31.77 GB/s
  GATED_DELTA_NET(type=f32,head_count=32,head_size=128,n_seq_tokens=512,n_seqs=1,v_repeat=1,permuted=0,kda=0,K=1):               908 runs -  1331.31 us/run -    36992 kB/run -   26.52 GB/s
  GATED_DELTA_NET(type=f32,head_count=32,head_size=128,n_seq_tokens=1024,n_seqs=1,v_repeat=1,permuted=0,kda=0,K=1):                      481 runs -  2688.74 us/run -    69888 kB/run -   24.84 GB/s
  GATED_DELTA_NET(type=f32,head_count=4,head_size=128,n_seq_tokens=64,n_seqs=1,v_repeat=1,permuted=0,kda=0,K=1):               16380 runs -   118.65 us/run -     1026 kB/run -    8.25 GB/s
  GATED_DELTA_NET(type=f32,head_count=4,head_size=128,n_seq_tokens=256,n_seqs=1,v_repeat=1,permuted=0,kda=0,K=1):               8190 runs -   464.97 us/run -     2568 kB/run -    5.27 GB/s
  GATED_DELTA_NET(type=f32,head_count=4,head_size=128,n_seq_tokens=512,n_seqs=1,v_repeat=1,permuted=0,kda=0,K=1):               7257 runs -   925.73 us/run -     4624 kB/run -    4.76 GB/s
  GATED_DELTA_NET(type=f32,head_count=4,head_size=128,n_seq_tokens=1024,n_seqs=1,v_repeat=1,permuted=0,kda=0,K=1):              3841 runs -  1847.93 us/run -     8736 kB/run -    4.51 GB/s
  GATED_DELTA_NET(type=f32,head_count=32,head_size=128,n_seq_tokens=64,n_seqs=1,v_repeat=1,permuted=0,kda=1,K=1):               7276 runs -   202.80 us/run -     9224 kB/run -   43.38 GB/s
  GATED_DELTA_NET(type=f32,head_count=16,head_size=128,n_seq_tokens=2048,n_seqs=1,v_repeat=2,permuted=0,kda=0,K=1):                      327 runs -  5308.05 us/run -   102912 kB/run -   18.53 GB/s
  GATED_DELTA_NET(type=f32,head_count=16,head_size=128,n_seq_tokens=4096,n_seqs=1,v_repeat=2,permuted=0,kda=0,K=1):                      167 runs - 11283.41 us/run -   201728 kB/run -   17.12 GB/s
  GATED_DELTA_NET(type=f32,head_count=16,head_size=128,n_seq_tokens=8192,n_seqs=1,v_repeat=2,permuted=0,kda=0,K=1):                       85 runs - 21377.87 us/run -   399360 kB/run -   17.95 GB/s
  GATED_DELTA_NET(type=f32,head_count=16,head_size=128,n_seq_tokens=16384,n_seqs=1,v_repeat=2,permuted=0,kda=0,K=1):                      43 runs - 42926.09 us/run -   794624 kB/run -   17.92 GB/s
  GATED_DELTA_NET(type=f32,head_count=16,head_size=128,n_seq_tokens=32768,n_seqs=1,v_repeat=2,permuted=0,kda=0,K=1):                      22 runs - 86225.55 us/run -  1585152 kB/run -   18.06 GB/s

after:
  GATED_DELTA_NET(type=f32,head_count=32,head_size=128,n_seq_tokens=1,n_seqs=1,v_repeat=1,permuted=0,kda=0,K=1):              145188 runs -     7.03 us/run -     4160 kB/run -  564.17 GB/s
  GATED_DELTA_NET(type=f32,head_count=16,head_size=64,n_seq_tokens=1,n_seqs=1,v_repeat=1,permuted=0,kda=0,K=1):               352170 runs -     2.90 us/run -      528 kB/run -  173.64 GB/s
  GATED_DELTA_NET(type=f32,head_count=32,head_size=128,n_seq_tokens=1,n_seqs=1,v_repeat=1,permuted=0,kda=1,K=1):              128560 runs -     8.02 us/run -     4176 kB/run -  496.59 GB/s
  GATED_DELTA_NET(type=f32,head_count=32,head_size=128,n_seq_tokens=64,n_seqs=1,v_repeat=1,permuted=0,kda=0,K=1):               8178 runs -   148.35 us/run -     8208 kB/run -   52.77 GB/s
  GATED_DELTA_NET(type=f32,head_count=32,head_size=128,n_seq_tokens=256,n_seqs=1,v_repeat=1,permuted=0,kda=0,K=1):              3268 runs -   607.05 us/run -    20544 kB/run -   32.29 GB/s
  GATED_DELTA_NET(type=f32,head_count=32,head_size=128,n_seq_tokens=512,n_seqs=1,v_repeat=1,permuted=0,kda=0,K=1):               908 runs -  1952.68 us/run -    36992 kB/run -   18.08 GB/s
  GATED_DELTA_NET(type=f32,head_count=32,head_size=128,n_seq_tokens=1024,n_seqs=1,v_repeat=1,permuted=0,kda=0,K=1):                      481 runs -  4015.98 us/run -    69888 kB/run -   16.63 GB/s
  GATED_DELTA_NET(type=f32,head_count=4,head_size=128,n_seq_tokens=64,n_seqs=1,v_repeat=1,permuted=0,kda=0,K=1):               24570 runs -    40.86 us/run -     1026 kB/run -   23.95 GB/s
  GATED_DELTA_NET(type=f32,head_count=4,head_size=128,n_seq_tokens=256,n_seqs=1,v_repeat=1,permuted=0,kda=0,K=1):               8190 runs -   143.91 us/run -     2568 kB/run -   17.02 GB/s
  GATED_DELTA_NET(type=f32,head_count=4,head_size=128,n_seq_tokens=512,n_seqs=1,v_repeat=1,permuted=0,kda=0,K=1):               7257 runs -   285.51 us/run -     4624 kB/run -   15.45 GB/s
  GATED_DELTA_NET(type=f32,head_count=4,head_size=128,n_seq_tokens=1024,n_seqs=1,v_repeat=1,permuted=0,kda=0,K=1):              3841 runs -   566.82 us/run -     8736 kB/run -   14.70 GB/s
  GATED_DELTA_NET(type=f32,head_count=32,head_size=128,n_seq_tokens=64,n_seqs=1,v_repeat=1,permuted=0,kda=1,K=1):               7276 runs -   229.16 us/run -     9224 kB/run -   38.39 GB/s
  GATED_DELTA_NET(type=f32,head_count=16,head_size=128,n_seq_tokens=2048,n_seqs=1,v_repeat=2,permuted=0,kda=0,K=1):                      327 runs -  7924.96 us/run -   102912 kB/run -   12.41 GB/s
  GATED_DELTA_NET(type=f32,head_count=16,head_size=128,n_seq_tokens=4096,n_seqs=1,v_repeat=2,permuted=0,kda=0,K=1):                      167 runs - 15852.84 us/run -   201728 kB/run -   12.18 GB/s
  GATED_DELTA_NET(type=f32,head_count=16,head_size=128,n_seq_tokens=8192,n_seqs=1,v_repeat=2,permuted=0,kda=0,K=1):                       85 runs - 31776.47 us/run -   399360 kB/run -   12.08 GB/s
  GATED_DELTA_NET(type=f32,head_count=16,head_size=128,n_seq_tokens=16384,n_seqs=1,v_repeat=2,permuted=0,kda=0,K=1):                      43 runs - 63808.14 us/run -   794624 kB/run -   12.06 GB/s
  GATED_DELTA_NET(type=f32,head_count=16,head_size=128,n_seq_tokens=32768,n_seqs=1,v_repeat=2,permuted=0,kda=0,K=1):                      22 runs - 127805.27 us/run -  1585152 kB/run -   12.18 GB/s

TheBlueMatt · 2026-09-28T15:32:01Z

Happy to run more tests if you want to suggest a range of gdn_subgroup_size/lanes_per_column to test.

SleepinDevil · 2026-09-28T23:47:18Z

I am happy to run more tests too on my B70 if it helps, I just need to be guided to what commands/tests to do

* vulkan: tune GDN shader * tune for Intel (cherry picked from commit 5c200e0)

…ix (ggml-org#29476) Exactly the 1,865 llama.cpp files the published llama-cpp-sys-2 0.1.158 crate ships (taken from upstream 26394b4, the llama.cpp it vendors), with upstream commit 5c200e0 ("vulkan: Tune GDN kernel, fix Intel performance", ggml-org#29476) applied to ggml/src/ggml-vulkan/ggml-vulkan.cpp. Nothing else changed. Only the crate's file set, because cargo checks out a git dependency's whole tree, and upstream's tools/ui paths exceed Windows MAX_PATH there (longest path here: 82 chars). Parentless, so the submodule fetch transfers only this tree. Retire once a llama-cpp-2 release vendors a llama.cpp that includes 5c200e0.

Points the llama.cpp submodule at Ethoseen/llama.cpp b8f1d912a70a1a05857beea45738d4b268df0841: the llama.cpp files the published 0.1.158 crate ships plus ggml-org/llama.cpp#29476 only. Everything else is the 0.1.158 release. Sightline uses it through [patch.crates-io]; retire once a llama-cpp-2 release carries the fix.

0cc4m added 2 commits September 26, 2026 12:44

vulkan: tune GDN shader

b0979ca

tune for Intel

7d7437a

0cc4m marked this pull request as ready for review September 26, 2026 10:49

0cc4m requested a review from a team as a code owner September 26, 2026 10:49

github-actions Bot added Vulkan Issues specific to the Vulkan backend ggml changes relating to the ggml tensor library for machine learning labels Sep 26, 2026

jeffbolznv approved these changes Sep 26, 2026

View reviewed changes

0cc4m added the merge ready A maintainer can use this label to indicate that they consider the changes final and ready to merge. label Sep 28, 2026

ggerganov merged commit 5c200e0 into master Sep 29, 2026
19 of 22 checks passed

ggerganov deleted the 0cc4m/vulkan-gdn-opt branch September 29, 2026 17:44

ggerganov mentioned this pull request Oct 5, 2026

ggml : bump version to 0.26.0 ggml-org/ggml#1652

Merged

BrewTestBot mentioned this pull request Oct 5, 2026

ggml 0.26.0 Homebrew/homebrew-core#315641

Merged

1 task

unfallen-314 mentioned this pull request Oct 6, 2026

Vulkan: ~45% prompt processing regression on RDNA1 (gfx1010) between b10455 and b11429 (v0.6.0) with a GDN-hybrid model; TG unchanged #30039

Open

edwardyoon pushed a commit to edwardyoon/focus-llama that referenced this pull request Oct 7, 2026

vulkan: Tune GDN kernel, fix Intel performance (ggml-org#29476)

373200b

* vulkan: tune GDN shader * tune for Intel (cherry picked from commit 5c200e0)

gaetan-puleo mentioned this pull request Oct 12, 2026

Merge upstream llama.cpp up to 4a096b8ff (2026-09-30) halo-box/strix-llama.cpp#174

Closed

2 of 5 tasks

Navigation Menu

Repository navigation

vulkan: Tune GDN kernel, fix Intel performance - #29476

vulkan: Tune GDN kernel, fix Intel performance#29476
ggerganov merged 2 commits into
masterfrom
0cc4m/vulkan-gdn-opt

0cc4m commented Sep 26, 2026

Uh oh!

SleepinDevil commented Sep 28, 2026

Uh oh!

0cc4m commented Sep 28, 2026

Uh oh!

TheBlueMatt commented Sep 28, 2026

Uh oh!

0cc4m commented Sep 28, 2026

Uh oh!

TheBlueMatt commented Sep 28, 2026 •

edited

Loading

Uh oh!

TheBlueMatt commented Sep 28, 2026

Uh oh!

SleepinDevil commented Sep 28, 2026

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

Conversation

0cc4m commented Sep 26, 2026

Overview

Test-backend-ops perf improvements

RTX 3090

Intel A770

Model benchmark (Qwen 3.6 35B Q2_K)

Prompt processing, t/s

Token generation, t/s

Requirements

Uh oh!

SleepinDevil commented Sep 28, 2026

PR 29476 - Qwen 3.6 27B Q6_K

PR 29476 - Qwen 3.8 27B Q8_K_L

Current source code build - Qwen 3.6 27B Q6_K

Current source code build - Qwen 3.8 27B Q8_K_L

Uh oh!

0cc4m commented Sep 28, 2026

Uh oh!

TheBlueMatt commented Sep 28, 2026

Uh oh!

0cc4m commented Sep 28, 2026

Uh oh!

TheBlueMatt commented Sep 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

TheBlueMatt commented Sep 28, 2026

Uh oh!

SleepinDevil commented Sep 28, 2026

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

TheBlueMatt commented Sep 28, 2026 •

edited

Loading