speed-bench: add NVIDIA RTX PRO 6000 Blackwell results#256
Open
imbibekk wants to merge 1 commit into
Open
Conversation
Numbers collected on a single NVIDIA RTX PRO 6000 Blackwell Server
Edition (97 GB VRAM, compute capability sm_120), CUDA 13.0, built with
`make cuda-generic`. Only GPU 0 is used since `ds4_gpu_init` selects a
single device.
Sweep (default flags):
./ds4-bench -m ds4flash.gguf \
--prompt-file speed-bench/promessi_sposi.txt \
--ctx-start 2048 --ctx-max 65536 --step-incr 2048 \
--gen-tokens 128 \
--csv speed-bench/rtx_pro_6000_blackwell.csv
The README row reports the 8192-token frontier so it sits next to the
existing DGX Spark GB10 row (7047 tokens) for direct comparison. The
full 2048..65536 sweep is in the CSV.
Caveat: at long context the engine logs
"CUDA q8 fp16 cache budget exhausted; using q8 kernels" because the
80.76 GiB model plus context buffers leave only about 1 GiB free on a
97 GiB card, so all frontiers reflect the fallback q8 kernel path
rather than the q8/fp16 cached fast path.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Numbers collected on a single NVIDIA RTX PRO 6000 Blackwell Server Edition (97 GB VRAM, compute capability sm_120), CUDA 13.0, built with
make cuda-generic. Only GPU 0 is used sinceds4_gpu_initselects a single device.Sweep (default flags):
./ds4-bench -m ds4flash.gguf
--prompt-file speed-bench/promessi_sposi.txt
--ctx-start 2048 --ctx-max 65536 --step-incr 2048
--gen-tokens 128
--csv speed-bench/rtx_pro_6000_blackwell.csv
The README row reports the 8192-token frontier so it sits next to the existing DGX Spark GB10 row (7047 tokens) for direct comparison. The full 2048..65536 sweep is in the CSV.
Caveat: at long context the engine logs
"CUDA q8 fp16 cache budget exhausted; using q8 kernels" because the 80.76 GiB model plus context buffers leave only about 1 GiB free on a 97 GiB card, so all frontiers reflect the fallback q8 kernel path rather than the q8/fp16 cached fast path.