m3 ultra flops estimates based on some quick profiling

50b6800a61830f540b1f77622c11876a74891844 · 2025-03-11 22:51:23 -0700 · Alex Cheema

Files touched

M exo/topology/device_capabilities.py

Diff

commit 50b6800a61830f540b1f77622c11876a74891844
Author: Alex Cheema <alexcheema123@gmail.com>
Date:   Tue Mar 11 22:51:23 2025 -0700

    m3 ultra flops estimates based on some quick profiling
---
 exo/topology/device_capabilities.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/exo/topology/device_capabilities.py b/exo/topology/device_capabilities.py
index e60003f2..4794c43a 100644
--- a/exo/topology/device_capabilities.py
+++ b/exo/topology/device_capabilities.py
@@ -56,6 +56,7 @@ CHIP_FLOPS = {
   "Apple M3": DeviceFlops(fp32=3.55*TFLOPS, fp16=7.10*TFLOPS, int8=14.20*TFLOPS),
   "Apple M3 Pro": DeviceFlops(fp32=4.97*TFLOPS, fp16=9.94*TFLOPS, int8=19.88*TFLOPS),
   "Apple M3 Max": DeviceFlops(fp32=14.20*TFLOPS, fp16=28.40*TFLOPS, int8=56.80*TFLOPS),
+  "Apple M3 Ultra": DeviceFlops(fp32=54.26*TFLOPS, fp16=108.52*TFLOPS, int8=217.04*TFLOPS),
   "Apple M4": DeviceFlops(fp32=4.26*TFLOPS, fp16=8.52*TFLOPS, int8=17.04*TFLOPS),
   "Apple M4 Pro": DeviceFlops(fp32=5.72*TFLOPS, fp16=11.44*TFLOPS, int8=22.88*TFLOPS),
   "Apple M4 Max": DeviceFlops(fp32=18.03*TFLOPS, fp16=36.07*TFLOPS, int8=72.14*TFLOPS),

← 2857975b upgrade grpcio and grpcio-tools to 1.71.0 · back to Exo · downgrade grpcio, grpcio-tools to 1.70.0 ad3bc6ce →