Top all-time · #601
Tim Dettmers
Personal blog · timdettmers.comaihardware
Tim Dettmers's personal blog at timdettmers.com covers ai and hardware. Since 2016, 102 different Hacker News readers have linked it 116 times, most often “Which GPU(s) to Get for Deep Learning?”. It is still writing: the last post was on 2026-09-21, and you can follow it by feed.
Often recommended · 16 of 40 sampled comments recommend it?
Why HN links it
HN readers link his GPU buying guide for deep learning as the reference for which card to buy, plus his posts on quantization and whether AGI will happen.
- Recommended for · 16 comments
- Called 'gold standard', 'fantastic', 'amazing'; 'covers a lot of ground and lays out choices quite well'.
Summarised by Claude from a sample of 40 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
Making deep learning accessible.
Most-linked posts?
- Which GPU(s) to Get for Deep Learning?24 people · 24 mentions · first linked 2023-01-30
- Which GPUs to get for deep learning9 people · 10 mentions · first linked 2020-09-07
- Brain vs. Deep Learning (2015)7 people · 7 mentions · first linked 2016-03-21
- Emergent Features in 6.7B+ Param Transformers6 people · 6 mentions · first linked 2022-08-18
- Deep Learning Hardware Limbo5 people · 5 mentions · first linked 2017-12-21
- Why AGI Will Not Happen5 people · 5 mentions · first linked 2025-12-10
- Which GPU(s) to Get for Deep Learning4 people · 6 mentions · first linked 2017-04-10
- GPUs for Deep Learning in 2023 – An In-depth Analysis4 people · 4 mentions · first linked 2023-01-16
- Which GPU(s) to Get for Deep Learning4 people · 4 mentions · first linked 2016-06-30
- Which GPU(s) to Get for Deep Learning3 people · 3 mentions · first linked 2017-03-19
What Hacker News says?
I'm only planning at this point - so I don't know but am very interested. I see the RTX 3080 reviewed as the most cost effective chip you can get for deep learning. I have the impression a lot of research is moving to lower precision also. https://timdettmers.com/2020/09/07/which-gpu-for-deep-learni...
Comment linking /2020/09/07/which-gpu-for-deep-learning/
on “Nvidia cripples cryptocurrency mining on RTX 3080 and 3070 cards”Submitted Which GPUs to get for deep learning as a story · 224 points
HN discussionMulti-GPU in desktop chassis gets crazy pretty quickly. If you don't care about aesthetics and can figure out both the power delivery and PCI-E lane situation, https://timdettmers.com/2023/01/30/which-gpu-for-deep-learni... has an example that will make a Bitcoin mining rig look clean. Water cooling can get you down to 2x slot height, with all of the trouble involved in water cooling. NVIDIA really segmented the mar…
Comment linking /2023/01/30/which-gpu-for-deep-learning/
on “Meta AI releases Code Llama 70B”Submitted Frontier AI on Your Own Hardware as a story · 185 points
HN discussionThis is a tough question to answer, because it depends a lot on what you want to do! One way to approach it may be to look at what models you want to run and check the amount of VRAM they need. A back-of-the-napkin method taken from here[0] is: VRAM (GB) = 1.2 * number of parameters (in billions) * bits per parameter / 8 The 1.2 is just an estimation factor to account for the VRAM needed for things that aren't model…
Comment linking /2023/01/30/which-gpu-for-deep-learning/
on “Llama.vim – Local LLM-assisted text completion”- nl · 2022-08-18
Submitted Emergent Features in 6.7B+ Param Transformers as a story · 185 points
HN discussion Needs a bit of an update but pretty good article https://timdettmers.com/2020/09/07/which-gpu-for-deep-learni...
Comment linking /2020/09/07/which-gpu-for-deep-learning/
on “GPU mining no longer profitable after Ethereum merge”Submitted Brain vs. Deep Learning (2015) as a story · 170 points
HN discussionHasn't been updated since 2020, but Tim Dettmer's guide [0] is pretty much the gold standard for optimizing what to buy for which area of DL/ML you're interested in. The pricing has changed thanks to GPU prices coming back down to earth a bit, but what to look out/how much ram you need for which task hasn't. Check out the "TL;DR advice" section then scroll back up for detail info on why and common misconceptions. Fo…
Comment linking /2020/09/07/which-gpu-for-deep-learning/
on “Open-source rival for OpenAI’s DALL-E runs on your graphics card”Submitted Which GPU(s) to Get for Deep Learning as a story · 166 points
HN discussionNot just SLI, but also non-SLI multi-GPU clusters which are common in machine learning, e.g.: http://timdettmers.com/2014/11/09/model-parallelism-deep-lea... And even if you don't use GPUs, now there are M.2 SSDs which commonly use 4X PCIe lanes, see https://en.wikipedia.org/wiki/M.2
Comment linking /2014/11/09/model-parallelism-deep-learning/
on “AMD Gives More Zen Details”Submitted My Experience and Advice for Using GPUs in Deep Learning: Which GPU to get as a story · 158 points
HN discussionThis blog post might shed some light on the matter. If I'm understanding it correctly, it claims there are emergent features on the LLM weights that make it easier to "compress" the floats into smaller bits without losing much precision. https://timdettmers.com/2022/08/17/llm-int8-and-emergent-fea... Note that 2 bit quantization is generally regarded as too aggressive. Generally 4bits+ achieves a good tradeoff, see…
Comment linking /2022/08/17/llm-int8-and-emergent-features/
on “Brave Leo now uses Mixtral 8x7B as default”Submitted Understanding Convolution in Deep Learning (2015) as a story · 142 points
HN discussionNvidia 1080 is a good one. Take a look at this write up. http://timdettmers.com/2014/08/14/which-gpu-for-deep-learnin... its updated in July 2016 with info about the later versions.
Comment linking /2014/08/14/which-gpu-for-deep-learning/
on “Radeon Instinct – Optimized Machine and Deep Learning”Submitted A Full Hardware Guide to Deep Learning as a story · 134 points
HN discussion17 years doesn't seem too far off. Tim Detmer's detailed analysis puts it anywhere from 2037 to 2078, with his preference the latter (or later). http://timdettmers.com/2015/07/27/brain-vs-deep-learning-sin... A key assumption I don't share with Detmer is that we must match the brain's raw computational power to have a chance at human-level AI. From an evolutionary point of view, most of the brain's computations sure…
Comment linking /2015/07/27/brain-vs-deep-learning-singularity/
on “Deep Learning 2016: The Year in Review”Submitted Deep Learning Hardware Limbo as a story · 92 points
HN discussionI read somewhere[1] that because we know far more now about human neurons than we did five years ago, modern CNN's are better compared to a _single_ neuron (but are probably still simpler than them). I'm far from an expert in brain biology, but what I took away from that is that it's specious to make these comparisons based on our current knowledge of either field. [1]: http://timdettmers.com/2015/07/27/brain-vs-dee…
Comment linking /2015/07/27/brain-vs-deep-learning-singularity/
on “Deep Learning 2016: The Year in Review”Submitted Why AGI Will Not Happen as a story · 57 points
HN discussionYes, ANNs are inspired by the brain. Here is a list of properties that ANNs shared with CPUs that are different from brains: * Synchronized activation vs. asynchronous / partially synchronous activation * Digital signals vs. analog signals * Instantaneous transmission of signals vs. delay imposed by axon and dendrite length * Uniform signal vs. use of various neurotransmitter signals * Rapid activation speed (GHz) v…
Comment linking /2015/07/27/brain-vs-deep-learning-singularity/
on “Neurogenesis Deep Learning”Submitted GPUs for Deep Learning in 2023 – An In-depth Analysis as a story · 25 points
HN discussionThey have invested and it works pretty well. But CUDA has such a huge lead and is the default. Most users dont care about the hardware but they also dont care enough about the cost. And the really price sensitive users are running their own clusters under their desks using gaming GPU’s. A more detailed explanation (ROCm section) https://timdettmers.com/2020/09/07/which-gpu-for-deep-learni...
Comment linking /2020/09/07/which-gpu-for-deep-learning/
on “Nvidia Announces A100 80GB GPU for AI”Submitted Deep Learning Hardware Limbo as a story · 13 points
HN discussionI commented earlier, but figured to add some more: I'm rooting for AMD and also looking to replace a 2012 MBP with an AMD laptop. The defacto cards for ML work seem to be NVDIA (eg their RTX line, https://timdettmers.com/2020/09/07/which-gpu-for-deep-learni... ) Does anyone know if pytorch/sci-kit might offer more support to AMD cards?
Comment linking /2020/09/07/which-gpu-for-deep-learning/
on “AMD Announces Ryzen “Zen 3” and Radeon “RDNA2” Presentations for October”Submitted Which GPU(s) to Get: My Experience and Advice for Using GPUs in Deep Learning as a story · 6 points
HN discussionI like this blog to understand NVidia perf stats (search for "Normalized GPU deep learning performance"): https://timdettmers.com/2020/09/07/which-gpu-for-deep-learni... It seems that their consumer GPUs also have a gap
Comment linking /2020/09/07/which-gpu-for-deep-learning/
on “PyTorch on Apple M1 MAX GPUs with SHARK – faster than TensorFlow-Metal”Submitted Which GPU(s) to Get for Deep Learning as a story · 6 points
HN discussionHere's a writeup on tpu vs got throughput: https://timdettmers.com/2018/10/17/tpus-vs-gpus-for-transfor... The underlying principle is that even if the silicon isn't strictly faster, you can shove a dumb amount of data through with the right architecture. And efficient training with data parallelism means that it's a worthwhile strategy.
Comment linking /2018/10/17/tpus-vs-gpus-for-transformers-bert/
on “Google’s Cloud TPU Pods are now publicly available in beta”Submitted Which GPU(s) to Get for Deep Learning: My Experience and Advice as a story · 5 points
HN discussion
