Редактирование: Whole brain emulation (раздел)

= Large-Scale Emulations So Far =

Simulations marked <span style="color:#0066FF;">blue</span> in are '''human-brain level'''.

{| class="wikitable sortable" border="1" align="center" style="text-align:center;"
! '''Year''' || '''Authors''' || '''Type''' || '''# of Neurons''' || '''# of Synapses''' || '''Hardware and Software''' || '''Slowdown factor''' || '''Timestep''' || '''Notes'''
|-
| 2007 || Hans E. Plesser, Jochen M. Eppler, Agibail Morrison, Markus Diesmann, and Marc-Oliver Gewaltig.<ref>Hans E. Plesser, Jochen M. Eppler, Agibail Morrison, Markus Diesmann, and Marc-Oliver Gewaltig. "Efficient parallel simulation of large‐scale neuronal networks on clusters of multiprocessor computers". Kermarrec A.‐M., Bougé L., and Priol T.: 672‐681, 2007. [http://mindzoo.de/files/plesser-europar07-fullpaper.pdf Available here]</ref> || Integrate-and-fire || 12,500 || 1.56x10<sup>7</sup> || Sun X4100 cluster, 2.4Ghz AMD Opteron, 8GB  RAM, MPI, NEST || 2 || 0.1 ms || Supralinear scaling until 80 virtual processes for 105 neuron/109 synapses.
|-
| 2006 || ikael Djurfeldt, Mikael Lundqvist, Christopher Johansson, Örjan Ekeberg, Martin Rehn, Anders Lansner.<ref>Mikael Djurfeldt, Mikael Lundqvist, Christopher Johansson, Örjan Ekeberg, Martin Rehn, Anders Lansner. "Project report for Blue Gene Watson Consortium Days: massively parallel simulation of brain- scale neuronal network models.". KTH - School of Computer Science and Communication and Stockholm University, 18 October 2006 [http://www‐fp.mcs.anl.gov/bgconsortium/Past%20Results/kth%20bgwreport06.pdf Available here].</ref>  || 6-compartment neurons || 2.2x10<sup>7</sup> || 1.1x10<sup>30</sup> || IBM Warson Research Blue Gene, SPLIT || 5,942 || Unknown || Computing requirements: 11.5 TFLOPS. Spatial delays corresponding 16 cm2 cortex surface.
|- style="color:black; background-color:#0066FF;" |
| 2005 || Izhikevich || Izhikevich neurons, Random synaptic connectivity generated on the fly. (On the fly as in, at every step. It was ''not'' generated at the beginning and then stored, [http://www.izhikevich.org/human_brain_simulation/why.htm because the matrix would've been 10,000 terabytes]). || 10<sup>11</sup> || 10<sup>15</sup> || Beowulf cluster, 27 3GHz processors || 4.2x10<sup>6</sup> (50 days of real time to simulate one second of activity) || 1 ms || Observed Alpha and Gamma rhythms. Holy Toledo!
|-
| 2000 || Howell, Dyhrfjeld-Johnsen et al. || Compartment (Purkinje 4,500 comp./granule 1 comp.) || 16 Purkinje, 244,000 granule cells || Unknown || 128-processor Cray 3TE, PGENESIS  || 4,500 || Unknown || Computing requirements: 76.8 GFLOPS
|-
| 2000 || Howell, Dyhrfjeld-Johnsen et al. || Unknown || 60,000 granule cells, 300 mossy fibers, 300 Golgi cells, 300 stellate cells, 1 Purkinje cell || Unknown || Single workstation 1GB memory, GENESIS || 79,200 || 20 μs || None
|-
| 2007 || Kozlov, Lansner et al. || 5 compartment model || 900 || 10% per hemisegment || SPLIT || Unknown || 100 μs || None
|-
| 2007 || Frye, Ananthanarayanan et al. || Low complexity spiking like Izhikevich neurons, STDP synapses || 8x10<sup>6</sup> || 6300*8x10<sup>6</sup> = 5.04x10<sup>10</sup> || 4096-processor BlueGene/L 256 Mb per CPU || 10 || 1 ms || None
|-
| 2007-2005 || Brette, Rudolph et  al.; Traub, Contreras et al. || 14 cell types,  conductance and compartment model || 3,560 || 3,500 gap junctions,  1,122,520 synapses || Cray XT3 2.4 Ghz, 800 CPUs || 200? || Event driven || 5954-8516 equations/CPU Gives # compartments
|-
| 2005 || Traub, Contreras et al. || 14 cell types, conductance and compartment model, 11 active conductances || 3,560 || 3,500 gap junctions, 1,122,520 synapses || 14 cpu Linux cluster, an IBM e1350, dual-processor Intel P4 Xeon, 2.4-GHz, 512 Kbytes of L2 cache; or, e1350 Blade 1, at 2.8 GHz. || Unknown || 2μs || None
|-
| 2001 || Goodman, Courtenay Wilson et al. || 3 layer cortical column, 25% GABAergic, conductance models, synapse reversal pot., conductance, abs. str., mean prob. release, time const. recover depress facilitation, || 5,150 per node, max 20*5,150=103, 000 || In columns 192,585 per node, max 20 = 3.85x10<sup>6</sup>, intercolumn 1.4x10<sup>6</sup> || NCS, 30 dual-Pentium III 1GHz processor nodes, with 4 GB of RAM per node || 165,000 || Unknown || None
|-
| 2004 || Frye || Four cell types. K<sub>m</sub>, K<sub>a</sub>, K<sub>ahp</sub> channels || 1,000/cells column, 2,501 columns = 2.5x10<sup>6</sup> || 250K per column, total 6.763x10<sup>8</sup> || BlueGene, 1024 CPUs || 3000? || Unknown || 2.338x10<sup>10</sup> spikes per processor. Memory requirement 110 Mb per processor
|-
| 2000 || Aberdeen, Baxter et al. || ANN feedforward network trained with iterative gradient descent || 4,083 || 1.73x10<sup>6</sup> || Cluster 196 Pentium III processors, 550 MHz, 384Mb || Unknown || Unknown || None
|-
| 1996 || Kondo, Koshiba et al. || Hopfield network || 1,536 || 2.4x10<sup>13</sup> || Special purpose chips || Unknown || Unknown || None
|-
| 2003 || Ccortex || Layered distribution of neural nets, detailed synaptic interconnections, spiking dynamics || 2x10<sup>10</sup> || 2x10<sup>13</sup> || 500 nodes, 1,000 processors, 1 Tb RAM. Theoretical peak of 4,800 Gflops. || Unknown || Unknown || Claimed performance, not documented.
|-
| 2002 || Harris, Baurick et al. || Point neurons, spikes, AHP, A & M potassium channels, synaptic adaptation, Spike shape and postsynaptic conductance specified by templates. || 35,000 cells per node || 6.1x10<sup>6</sup> synapses per node || Cluster, 128 Xeon 2.2 GHz processors with 256GB, 1Tb disk || Unknown || Unknown || None
|-
| 2003 || Mehrtash, Jung et al. || Integrate and fire, event driven pulse connections with STDP plasticity. 1-8 dendritic segments. || 256,000 || 10<sup>6</sup> || Special purpose chip, connected UltraSparc processor 500 Mhz, 640 Mb || 1021.21 timesteps/ms || Unknown || None
|-
| 2002 || Shaojuan and Hammarstrom || Palm binary network || 256,000 || Order of 3.2x10<sup>10</sup> || 512 processors, 192 GB, 250 Mhz mips 12K processors || Execution time 4 min 7 sec, -13% MPI calls for retrieving 180K training vectors, ~O(10) iterations. || Unknown || None
|-
| 2007 || Johansson and Lansner || BCPNN spiking, minicolumns hypercolumns || 1.6x10<sup>6</sup> || 2.0x10<sup>11</sup> || 256 node, 512 processors, Dell Xeon cluster, 3.4GHz 8 Gb memory, peak performance 13.6 Gflop || 9% real-time = 11.1 || Unknown || Weight updates 47 ms, update activities 59 ms
|-
|| 2006 || Markram || Morphologically complex compartment models || 10,000 || 10<sup>8</sup> || BlueGene/L, 4096 nodes, 8192 cpus, peak performance 22.4 TFLOPS || Unknown || Unknown || Not documented?
|-
|| 1982 || Traub and Wong || Unknown || 100 || Unknown || Unknown || Unknown || Unknown || None
|-
|| 1992 || Traub, Miles et al. || 19-compartment cells, ionic currents. || 1,200 || 70,000 || IBM3090 computer || 325 min/1.5s bio time = 13,000 || 0.05 ms || None
|-
|| 1997 || Moll and Miikkulainen || Binary units with binary weights || 79,500 || 7.82x10<sup>8</sup> || Cray Y-MP 8/864 || 0.006s (retrieval of 550,000 patterns; assuming human retrieval speed = 1 s) || Unknown || None
|}