AlphaFold Performance: molecule size, speed, memory, and GPU

Tom Goddard
January 12, 2022, updated February 5, 2024

The table shows ColabFold 1.5.5 is about 2 to 10 times faster than AlphaFold 2.3.2.

AlphaFold version 2.1.1 has limitations in the protein or complex size that can be predicted limited mostly by available GPU memory. Also predictions of large structures can take more than 10 hours. Here are examples to gauge what is possible.

AlphaFold 2.1.2 switched OpenMM minimization to use GPU while AlphaFold 2.1.1 used CPU. This makes runs faster. Times below are for 2.1.1 unless stated otherwise.

Maximum Structure Size

How many amino acids can a protein or complex contain before AlphaFold runs out of memory? And how often does it make wrong predictions of large complexes? Here are examples of runs on large structures.

PDBProteins / uniqueTotal amino acidsGPU memory (GB)Nvidia GPUTime (hours)ComputerNotes
CNAG_01613
CNAG_06362		2 / 2467848A4065UCSF clusterColabFold 1.5.5. Completed, two plausible folds. Homologs of UNC79 and UNC80 in PDB 7sx3.
CNAG_01613
CNAG_06362		2 / 2467848A40103UCSF clusterAlphaFold 2.3.2. Completed, two plausible folds. Homologs of UNC79 and UNC80 in PDB 7sx3.
7KIY3 / 3378048A4016.5UCSF clusterColabfold 1.5.5. Completed, wrong packing, and monomers have lots of self-clashes. Image.
6IG91374448A4035UCSF clusterCompleted, correct fold, used PDB templates. Image.
3GHG + 8T2V5/5320448A407.5UCSF cluster1ColabFold 1.5.5. Fibrin and integrin each predicted correctly but no binding identified. Image.
6YEJ1319948A4028UCSF cluster1Each of 2 domains predicted correctly, but relative position wrong. Image.
7JGD1266048A4013UCSF clusterCompleted. Misaligned domains. Image.
7LHW1252748A4017UCSF clusterCompleted, roughly correct with many shifted domain positions. Image.
6UM1124992430901Desktop PCFailed in jax CUDA_ERROR_ILLEGAL_ADDRESS, probably out of GPU memory. Log.
8DIT3 / 3249848A403UCSF clusterColabFold 1.5.5. Completed. Trimer not packed correctly but domains of proteins are mostly correct.
8DIT3 / 3249824309012Desktop PCAlphaFold 2.3.1. Completed. Trimer not packed correctly but domains of proteins are mostly correct.
8DIT3 / 3249848A4016UCSF clusterAlphaFold 2.3.1. Completed. Trimer not packed correctly but domains of proteins are mostly correct.
7M1P1227348A407.5UCSF clusterAlphaFold 2.1.2 and reduced databases which does not use hhblits. Mostly correct prediction, 1A RMSD for 1900 atoms, N terminal domains shifted 10A. Image.
7M1P1227348A403.7UCSF clusterFailed making sequence alignment in hhblits "ERROR: did not find 569 match states in sequence 1 of tr|A0A1H9CGJ3|A0A1H9CGJ3_9FIRM". AlphaFold 2.1.1. Log.
7M1P122732430904.5Desktop PCFailed making sequence alignment. Log.
7ALP120842430908Desktop PC2Completed. Misaligned domains. Image.
7BAN119322430901.3Desktop PCFailed making sequence alignment in hhblits. Log.
7P3T6 / 1181248A4012.5UCSF clusterCorrect prediction, mostly 1A rmsd. Image.
8T2V2 / 2177048A401UCSF clusterColabFold 1.5.5. Correct prediction. Image.
6X6U4 / 215922430903.5Desktop PCCorrect prediction, mostly 1A rmsd. Image.
7vku4 / 415272430901.5Desktop PCAlphafold 2.3.1 using local ColabFold installation, no templates, no minimization. Part correct 0.9 A rmsd, beta barrels not joined.
7vku4 / 4152716T44Google ColabFailed after computing 2 models. Alphafold 2.3.1 using ColabFold web site, no templates, no minimization. Said "session crashed after using all available RAM". Also said "Session timed out". Run with ColabPro.
7S621144116P1002Google ColabFailed, out of GPU memory allocating 19 GB. Colab resource monitor shows GPU (not CPU) memory is exhausted. 5000 aligned sequences. Log.
7Q5Z1143616P1001Google ColabFailed using ColabFold after predicting two models. CUDA_ERROR_ILLEGAL_ADDRESS. Log.
3GHG3 / 3143448A400.7UCSF clusterColabfold 1.5.5. Correct fold. Image.
7E5N4 / 1129248A404UCSF clusterCompleted with AlphaFold 2.1.2. Wrong packing. Image.
7E5N4 / 1129248A4011UCSF clusterCompleted with AlphaFold 2.1.1. Wrong packing. Image.
7E5N4 / 112922430902.5Desktop PCCompleted with AlphaFold 2.1.2. Wrong packing. Image.
7E5N4 / 112922430903.5Desktop PCFailed predicting model 2 with OpenMM error writing PDB "The number of positions must match the number of atoms". AlphaFold 2.1.1. Identical error on second try. Log.
7QFP1126916P1002Google ColabCompleted with ColabFold with some wrong domain positions. Image.
6XMP1123916P1005Google ColabFailed. Only 1 of 5 models completed, others gave out of memory. One completed model ran out of memory in energy minimization. 6000 aligned sequences.
7KTT1114216P1003Google ColabFailed. Only 2 of 5 models completed, others gave out of memory allocating 13 GB. One completed model ran out of memory in energy minimization. 1500 aligned sequences. Log. Image.
6Z032 / 1107816P1005Google Colab3Completed with wrong complex, proteins on top of each other. Image.
5N5F10 / 198016P1002.5Google ColabCorrect prediction, 0.4 A rmsd. Image.
6Z1J3 / 38242430901.5Desktop PCCorrect prediction, mostly 0.5 A rmsd. Image.
6Z1J3 / 382448A402.5UCSF clusterCorrect prediction, mostly 0.5 A rmsd.
6Z1J3 / 382416P1005Google ColabCorrect prediction, mostly 0.5 A rmsd.
7fc71962430900.02Desktop PCAlphafold 2.3.1 using local ColabFold installation, no templates, no minimization. Correct prediction, mostly 0.8 A rmsd.
7fc719616T40.08Google ColabAlphafold 2.2.2. ColabFold run on Google Colab through ChimeraX, no templates, no minimization. Correct prediction, mostly 0.8 A rmsd. 1 minute to install ColabFold on Google Colab and 4 minutes to run 5 models.
7fc719648A400.02UCSF clusterColabFold 1.5.5, no templates, no minimization. Correct prediction, mostly 0.7 A rmsd. Sequence alignment done in advance with --msa-only option.
7fc719648A400.8UCSF clusterAlphafold 2.3.1. Correct prediction, mostly 0.7 A rmsd.
  1. UCSF cluster: Wynton cluster, using Nvidia A40 GPUs with 48 GB video memory, CPU info N/A, Alphafold databases on parallel BeeGFS file system.
  2. Desktop PC: Intel Core i9-10850K CPU @ 3.60GHz, 64 GBytes memory, Nvidia RTX 3090 / 24 GBytes, AlphaFold databases on 4 TB Samsung 870 QVO SATA 3 SSD drive.
  3. Google Colab: Run from ChimeraX (1.4 daily build December 2021) menu Tools / Structure Prediction / AlphaFold. Using Colab Pro paid service with Nvidia GPUs P100, K80, or T4 with 16 GB identified with nvidia-smi command from Colab shell. Small AlphaFold databases streamed from web, no templates.

Speed with GPU vs without GPU

AlphaFold can be run without a GPU using the AlphaFold Docker --use_gpu=False option or the run_alphafold.py script --gpu_devices=-1 option. Here are some tests with small proteins comparing runs without GPU to with GPU.

PDBProteins / uniqueTotal amino acidsGPU memory (GB)Nvidia GPUTime (hours)ComputerNotes
6Z1J3 / 3824No GPU17UCSF clusterCorrect prediction.
6Z1J3 / 382448A402.5UCSF clusterCorrect prediction, mostly 0.5 A rmsd.
7B761125No GPU2.7UCSF clusterWrong fold. Image.
7B76112548A400.5UCSF clusterWrong fold.
7B761125No GPU1Desktop PCWrong fold.
7B7611252430900.3Desktop PCWrong fold.