Performance of llama.cpp with Vulkan #10879
Replies: 193 comments 292 replies
-
|
AMD FirePro W8100
|
Beta Was this translation helpful? Give feedback.
-
|
With the latest updates:
|
Beta Was this translation helpful? Give feedback.
-
|
With FA:
|
Beta Was this translation helpful? Give feedback.
-
|
AMD RX 470
|
Beta Was this translation helpful? Give feedback.
-
|
With the latest updates:
|
Beta Was this translation helpful? Give feedback.
-
|
With FA:
|
Beta Was this translation helpful? Give feedback.
-
|
i got the mining edition 8 gb ran slightly better
build: e288693 (6242) |
Beta Was this translation helpful? Give feedback.
-
|
ubuntu 24.04, vulkan and cuda installed from official APT packages.
build: 4da69d1 (4351) vs CUDA on the same build/setup
build: 4da69d1 (4351) |
Beta Was this translation helpful? Give feedback.
-
|
Macbook Air M2 on Asahi Linux ggml_vulkan: Found 1 Vulkan devices:
|
Beta Was this translation helpful? Give feedback.
-
|
For the record I think this is slow on the HoneyKrisp side rather than llama.cpp |
Beta Was this translation helpful? Give feedback.
-
|
Can you share how you got vulkan to build on Asahi? I can't seem to get cmake to notice it. |
Beta Was this translation helpful? Give feedback.
-
|
Spoke too soon, got it working! |
Beta Was this translation helpful? Give feedback.
-
|
Gentoo Linux on ROG Ally (2023) Ryzen Z1 Extreme ggml_vulkan: Found 1 Vulkan devices:
|
Beta Was this translation helpful? Give feedback.
-
|
ggml_vulkan: Found 4 Vulkan devices:
|
Beta Was this translation helpful? Give feedback.
-
|
Cool setup! Could you also post the result of 1, 2 and 3 7900 XTX GPUs? You can use only the first GPU with |
Beta Was this translation helpful? Give feedback.
-
|
|
Beta Was this translation helpful? Give feedback.
-
|
For this multi GPU case getting Vulkan to support #6017 pipeline parallelism might help improve the prompt processing speed. |
Beta Was this translation helpful? Give feedback.
-
|
@netrunnereve I updated the commit id in all my result. |
Beta Was this translation helpful? Give feedback.
-
|
build: 0d52a69 (4439) NVIDIA GeForce RTX 3090 (NVIDIA)
AMD Radeon RX 6800 XT (RADV NAVI21) (radv)
AMD Radeon (TM) Pro VII (RADV VEGA20) (radv)
Intel(R) Arc(tm) A770 Graphics (DG2) (Intel open-source Mesa driver)
|
Beta Was this translation helpful? Give feedback.
-
|
@netrunnereve Some of the tg results here are a little low, I think they might be debug builds. The cmake step (at least on Linux) might require |
Beta Was this translation helpful? Give feedback.
-
|
I've added My release numbers for the RX 470 are basically identical to the ones I posted earlier without the flag.
|
Beta Was this translation helpful? Give feedback.
-
|
Maybe not in your case, but some other results are suspiciously low in tg (for example the RTX 3080) |
Beta Was this translation helpful? Give feedback.
-
|
Build: 8d59d91 (4450)
Lack of proper Xe coopmat support in the ANV driver is a setback honestly.
edit: retested both with the default batch size. |
Beta Was this translation helpful? Give feedback.
-
I put my A770 into a Windows PC and gave Intel GPA and vtune a shot: GPA just crashes most of the time, I couldn't get it to trace anything useful. vtune works, but does not support Vulkan. It just shows some high-level metrics in that case, not really useful sadly. |
Beta Was this translation helpful? Give feedback.
-
I did build it with cmake with build type Release. |
Beta Was this translation helpful? Give feedback.
-
|
In that case it's something else, cause it should be performing similarly to my A770. I suspect the mesa version, there was something in newer mesa versions that slowed down tg on Intel. |
Beta Was this translation helpful? Give feedback.
-
|
A750 has 448 CUs, A770 has 512 CUs I think. Personally, I am not worried about tg. I am worried about pp here. The gemm batch quickly saturates my GPU. |
Beta Was this translation helpful? Give feedback.
-
|
@0cc4m https://gitlab.freedesktop.org/mesa/mesa/-/issues/12585 |
Beta Was this translation helpful? Give feedback.
-
|
Here's something exotic: An AMD FirePro S10000 dual GPU from 2012 with 2x 3GB GDDR5. build: 914a82d (4452)
|
Beta Was this translation helpful? Give feedback.
-
|
Very interesting, and looks like it's pretty close to the W8100 in tg despite being a dual GPU card. Your backend scales pretty well with layer splitting which is why I find it worthwhile to run my RX470 and W8100 together (I end up getting results that are close to the average of both cards).
|
Beta Was this translation helpful? Give feedback.
-
|
Latest arch with For the sake of consistency I run every bit in a script and also build every target from scratch (for some reason kill -STOP -1
timeout 240s $COMMAND
kill -CONT -1
ggml_vulkan: Found 1 Vulkan devices:
ggml_vulkan: 0 = Intel(R) Iris(R) Xe Graphics (TGL GT2) (Intel open-source Mesa driver) | uma: 1 | fp16: 1 | warp size: 32 | matrix cores: none
build: ff3fcab (4459)
This bit seems to underutilise both GPU and CPU in real conditions based on
|
Beta Was this translation helpful? Give feedback.
-
|
Unless you reduce the number of GPU layers, threads and openblas/non-openblas is not gonna make any difference. Try it with ngl 0, then only prompt processing is accelerated using Vulkan, the rest runs on CPU. This is often a good setting for integrated GPUs. |
Beta Was this translation helpful? Give feedback.
-
|
That's something I didn't think about, with
build: ba8a1f9 (4460)
|
Beta Was this translation helpful? Give feedback.
-
|
It seems latest patches has improved the results a bit: ggml_vulkan: Found 1 Vulkan devices:
|
Beta Was this translation helpful? Give feedback.
-
|
A few months later and I get:
build: f3a4b16 (5568) I run it on Linux (Arch with lama.cpp-vulkan-git package compiled by GCC 15). From my tests, only Vulkan backend ( I'm curious why I cannot go over 6 t/s. Is this an issue with the newer llama.cpp version or with my OS configuration? |
Beta Was this translation helpful? Give feedback.
-
|
Intel ARC A770 on Windows:
build: ba8a1f9 (4460) |
Beta Was this translation helpful? Give feedback.
-
Single GPU VulkanRadeon Instinct MI25 ggml_vulkan: 0 = AMD Radeon Instinct MI25 (RADV VEGA10) (radv) | uma: 0 | fp16: 1 | warp size: 64 | matrix cores: none
build: 2739a71 (4461) Radeon PRO VII ggml_vulkan: 0 = AMD Radeon Pro VII (RADV VEGA20) (radv) | uma: 0 | fp16: 1 | warp size: 64 | matrix cores: none
build: 2739a71 (4461) Multi GPU Vulkanggml_vulkan: 0 = AMD Radeon Pro VII (RADV VEGA20) (radv) | uma: 0 | fp16: 1 | warp size: 64 | matrix cores: none
build: 2739a71 (4461) ggml_vulkan: 0 = AMD Radeon Pro VII (RADV VEGA20) (radv) | uma: 0 | fp16: 1 | warp size: 64 | matrix cores: none
build: 2739a71 (4461) Single GPU RocmDevice 0: AMD Radeon Instinct MI25, compute capability 9.0, VMM: no
build: 2739a71 (4461) Device 0: AMD Radeon Pro VII, compute capability 9.0, VMM: no
build: 2739a71 (4461) Multi GPU RocmDevice 0: AMD Radeon Pro VII, compute capability 9.0, VMM: no
build: 2739a71 (4461) Layer split
build: 2739a71 (4461) Row split
build: 2739a71 (4461) Single GPU speed is decent, but multi GPU trails Rocm by a wide margin, especially with large models due to the lack of row split. |
Beta Was this translation helpful? Give feedback.
-
|
What is the power profile for this MI25? Mine is 110W but its running slower than yours on git from today. |
Beta Was this translation helpful? Give feedback.
-
|
Mine defaults to 220w. |
Beta Was this translation helpful? Give feedback.
-
|
AMD Radeon RX 5700 XT on Arch using mesa-git and setting a higher GPU power limit compared to the stock card.
I also think it could be interesting adding the flash attention results to the scoreboard (even if the support for it still isn't as mature as CUDA's).
|
Beta Was this translation helpful? Give feedback.
-
|
There is no Vulkan flash attention support (except with coopmat2 on very new nvidia drivers). What you're measuring here is a CPU fallback. |
Beta Was this translation helpful? Give feedback.
-
|
I see, I was sure about the CPU fallback but didn't know there was no flash attention support at all. |
Beta Was this translation helpful? Give feedback.
-
|
I tried but there's nothing after 1 hrs , ok, might be 40 mins... Anyway I run the llama_cli for a sample eval...
Meanwhile OpenBLAS |
Beta Was this translation helpful? Give feedback.
-
|
Even at below 1t/s |
Beta Was this translation helpful? Give feedback.
-
|
Truth is ... (0.79 tokens per second), 3788.63 ms / 3 tokens So it's not even...it just slower... |
Beta Was this translation helpful? Give feedback.
-
|
Radeon 680M (Ryzen 5 6800H), LPDDR5 6400 MT/s
build: ee09828 (6795) |
Beta Was this translation helpful? Give feedback.
-
|
RTX 3070 Laptop GPU (8 GB / GDDR6 / 256 bit) Driver Version: 580.76.05
build: ceff6bb (6783) |
Beta Was this translation helpful? Give feedback.
-
|
RX Vega 56
build: 66b0dbc (6791) |
Beta Was this translation helpful? Give feedback.
-
|
I know that's an old run but interestingly this is beating the Vega 64 by quite a bit. |
Beta Was this translation helpful? Give feedback.
-
|
|
Beta Was this translation helpful? Give feedback.
-
|
|
Beta Was this translation helpful? Give feedback.
-
|
Another Strix Halo 395+ (128GB). Also with the optimizations alluded to in the Strix Halo benchmarking guide. radv, mesa 26.0.0 amdvulkan |
Beta Was this translation helpful? Give feedback.
-
|
AMD Radeon RX 480 $ llama-bench --device vulkan2 -ngl 100 -m llama-2-7b.Q4_0.gguf -fa 0,1
build: 0bcb40b (6833) |
Beta Was this translation helpful? Give feedback.
-
|
AMD Ryzen 5 5600H with Vega 7 iGPU. The package is capped at 35W. 2x 32GB DDR4 3200MHz (generic). Slightly more surprising is this: |
Beta Was this translation helpful? Give feedback.
-
|
I get ~11 tps running GPT-OSS on RAM, but only 6 tps when running 7B. This could have been attributed to the smaller number of active parameters of GPT-OSS. |
Beta Was this translation helpful? Give feedback.
-
|
@pt13762104 Yes, most likely. Memory bandwidth is a limiting factor here I think. However those models make local LLMs feasible on platforms that were never thought to be sufficient for it. Here's another example, 7B MoE, running completely on the iGPU and staying relatively cool. |
Beta Was this translation helpful? Give feedback.
-
You can uncap TDP for package / iGPU. https://github.com/JamesCJ60/Universal-x86-Tuning-Utility/releases I use this, and on my humble R5 4650U I GPU runs so fast you can play BG3 in full HD with decent FPS. |
Beta Was this translation helpful? Give feedback.
-
I know, but (a) this is a tiny mini PC that fits in my palm - it simply doesn't provide the cooling or voltage regulators; (b) the wall wart delivers 55W max; and (c) this isn't Windows (but POR can be edited in the BIOS). |
Beta Was this translation helpful? Give feedback.
-
Wall wart hahahahahaha made me laugh aloud |
Beta Was this translation helpful? Give feedback.
-
|
b6840, macOS Sequoia AMD Radeon RX 6900 XT, eGPU, TB3, iMac Pro ./llama.cpp/build/bin/llama-bench -m ./GGUF/llama-2-7b.Q4_0.gguf -ngl 100 -fa 0,1 -sm none -mg 0
AMD Radeon Pro Vega 64, internal, iMac Pro ./llama.cpp/build/bin/llama-bench -m ./GGUF/llama-2-7b.Q4_0.gguf -ngl 100 -fa 0,1 -sm none -mg 1
FA ALL: ./llama/llama-bench -m ./GGUF/llama-2-7b.Q4_0.gguf -ngl 100 -fa all -sm none -mg 0
./llama/llama-bench -m ./GGUF/llama-2-7b.Q4_0.gguf -ngl 100 -fa all -sm none -mg 1
|
Beta Was this translation helpful? Give feedback.
-
|
Radeon Vega 6 APU
(AMD Ryzen 5 PRO 4650U) .\llama-bench.exe -m llama-2-7b.Q4_0.gguf -ngl 100 -fa 0,1
FA ALL .\llama-bench.exe -m llama-2-7b.Q4_0.gguf -ngl 100 -fa all
|
Beta Was this translation helpful? Give feedback.
-
|
AMD Radeon 880M (Ryzen AI 9 365) Tested with RADV and AMDVLK. % AMD_VULKAN_ICD=AMDVLK ./build-vk/bin/llama-bench -m llama-2-7b.Q4_0.gguf -fa 0,1
% AMD_VULKAN_ICD=RADV ./build-vk/bin/llama-bench -m llama-2-7b.Q4_0.gguf -fa 0,1
build: c55d53a (6854) Edit: it's reported as 890M but it's 880M. |
Beta Was this translation helpful? Give feedback.
-
|
The prompt processing of these two cords seems very low. Using cuda I am getting almost 500t/s for the gtx 1070 and 572t/s on the p104-100 with FA. Nvidia GeForce GTX 1070 ggml_vulkan: 0 = NVIDIA GeForce GTX 1070 (NVIDIA) | uma: 0 | fp16: 0 | bf16: 0 | warp size: 32 | shared memory: 49152 | int dot: 0 | matrix cores: none
build: ad51c0a (6948) Nvidia P104-100 ggml_vulkan: Found 3 Vulkan devices:
build: ad51c0a (6948) |
Beta Was this translation helpful? Give feedback.
-
|
Nvidia Pascal GPU's prompt processing should be faster once you figure out why int dot is not working. It's either an outdated driver, or an outdated Vulkan SDK that was used for compiling the version you are benchmarking. It should show up was |
Beta Was this translation helpful? Give feedback.
-
|
Thank you for the information. I believe I am using the latest nvidia drivers (575.57.08). How do I check for the vulkan sdk? I don't remember installing one. I just git pulled the llama.cpp and compiled it with the right flags (cuda and vulkan support because of my mixed setup (2 pascal card and a rx 480 that I had lying around when I built this box)). The radeon rx 480 is also showing int dot: 0 so is it possible it could also have a faster prompt processing if I find the cause? |
Beta Was this translation helpful? Give feedback.
-
|
vulkaninfo --summary could give more data. Here's what I get: |
Beta Was this translation helpful? Give feedback.
-
AMD Radeon 780M (AMD Ryzen 7 8845HS w/ Radeon 780M Graphics)Mesa 25.2.5 VulkanAMD_VULKAN_ICD=RADV GGML_VK_VISIBLE_DEVICES=0 .build/bin/llama-bench --model ~/work/.models/llama-2-7b.Q4_0.gguf -fa 0,1
build: a5c07dc (6949) AMD VulkanAMD_VULKAN_ICD=AMDVLK GGML_VK_VISIBLE_DEVICES=1 .build/bin/llama-bench --model ~/work/.models/llama-2-7b.Q4_0.gguf -fa 0,1
build: a5c07dc (6949) |
Beta Was this translation helpful? Give feedback.
-
NVIDIA GeForce RTX 4060 Laptop GPUDriver version 580.95.5.0GGML_VK_VISIBLE_DEVICES=0 .build/bin/llama-bench --model ~/work/.models/llama-2-7b.Q4_0.gguf -fa 0,1
build: a5c07dc (6949) |
Beta Was this translation helpful? Give feedback.
-
Powercolor Red Devil 7900XTXAdrenalin 25.10.2
ggml_vulkan: Found 1 Vulkan devices:
ggml_vulkan: 0 = AMD Radeon RX 7900 XTX (AMD proprietary driver) | uma: 0 | fp16: 1 | bf16: 1 | warp size: 64 | shared memory: 32768 | int dot: 1 | matrix cores: KHR_coopmat
build: 2f0c2db (6957) Vs Aug 6: 7.39% faster, and 6.34% faster with FA. Not only that, this is the new high score for 7900XTX. |
Beta Was this translation helpful? Give feedback.
-
That's just for prompt processing, inference is literally 30% faster 😁 |
Beta Was this translation helpful? Give feedback.
and others
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
This is similar to the Apple Silicon benchmark thread, but for Vulkan! We'll be testing the Llama 2 7B model like the other thread to keep things consistent, and use Q4_0 as it's simple to compute and small enough to fit on a 4GB GPU. You can download it here.
Instructions
Either run the commands below or download one of our Vulkan releases. If you have multiple GPUs please run the test on a single GPU using
-sm none -mg YOUR_GPU_NUMBERunless the model is too big to fit in VRAM.Share your llama-bench results along with the git hash and Vulkan info string in the comments. Feel free to try other models and compare backends, but only valid runs will be placed on the scoreboard.
If multiple entries are posted for the same setup I'll prioritize newer commits with substantial Vulkan updates, otherwise I'll pick the one with the highest overall score at my discretion. Performance may vary depending on driver, operating system, board manufacturer, etc. even if the chip is the same. For integrated graphics note that the memory speed and number of channels will greatly affect your inference speed!
Vulkan Scoreboard (Click on the headings to expand the section)
Llama 2 7B, Q4_0, no FA
Llama 2 7B, Q4_0, FA enabled
Beta Was this translation helpful? Give feedback.
All reactions