Alex Bradbury

At Igalia since March 2022.

Cambridge (UK)

POSTS

Dec 2, 2025

QEMU-based instruction execution counting

Although analysing performance by way of instruction counting has obvious limitations, it can be helpful (especially when combined with appropriate analysis scripts) to get rapid feedback on the impact of...

Dec 1, 2025

Minipost: Olmo 3 training cost

Recently I jotted down some notes on LLM inference vs training costs for DeepSeek and I wanted to add on an additional datapoint for training cost based on the recently...

Nov 30, 2025

Minipost: Benchmarking the Hetzner AX102 vs CCX53

I recently had reason to do a quick comparison of the performance of the Hetzner AX102 dedicated server and the high-end ‘dedicated’ CCX53 VPS on Hetzner Cloud and thought I...

Nov 29, 2025

Minipost: LLM inference vs training costs for DeepSeek

Tl;dr: Based on published data from DeepSeek, we can estimate it takes something like ~70 days of inference traffic (served by DeepSeek themselves, ignoring any other providers) to match the...

Media

Improvements to RISC-V Vector Code Generation in LLVM

Jul 7, 2025 Slides (354 KB PDF)

Lessons learned from leveling up RISC-V LLVM testing

Jun 9, 2025

RISC-Y Business

Jan 31, 2025

Toots

Dec 3, 2025

I’ve started writing up the flow I use for capturing instruction execution frequency data from benchmarks using QEMU. First we...

Oct 14, 2025

Unknown whether this is due to changes in default buffer size in sox, the Linux audio stack, or otherwise, but...

Jul 21, 2025

Pour one out for NativeClient, finally removed from the LLVM tree last week https://github.com/llvm/llvm-project/commit/0d2e11f3e834

Jun 26, 2025

An alternative algorithm for enforcing per-author paper submission limits. https://arxiv.org/abs/2506.20141AI driving algorithms research, but perhaps not as expected.

Commits

Nov 27, 2025

Team

Alex Bradbury

Cambridge (UK)

POSTS

QEMU-based instruction execution counting

Minipost: Olmo 3 training cost

Minipost: Benchmarking the Hetzner AX102 vs CCX53

Minipost: LLM inference vs training costs for DeepSeek

Media

Improvements to RISC-V Vector Code Generation in LLVM

Lessons learned from leveling up RISC-V LLVM testing

RISC-Y Business

Toots

Commits

[LLVM] Revert "[ShrinkWrap] Modify shrink wrapping to accommodate functions …

[LLVM] [RISCV][NewPM] Port RISCVCodeGenPrepare to the new pass manager (#168…

[LLVM] [AArch64][test] Improve pr166870.ll test case (#168194)

[LLVM] Revert "[VPlan] Expand WidenInt inductions with nuw/nsw" (#168080)

Alex Bradbury

Cambridge (UK)

Follow me

POSTS

Media

Toots

Commits