git mv cpu-memory compute/cpu-memory

Author: stas00

README.md

@@ -27,7 +27,7 @@ My apologies if the layout is a bit unstable while I'm writing new chapters and
 
 1. **[CPU](./cpu/)** - cpus, affinities (WIP)
 
-1. **[CPU Memory](./cpu-memory/)** - how much CPU memory is enough - the shortest chapter ever.
+1. **[CPU Memory](compute/cpu-memory)** - how much CPU memory is enough - the shortest chapter ever.
 
 
 **Part 3. Performance**

chapters-md.txt

@@ -11,7 +11,7 @@
 ./storage/results/hope-2023-12-20-14-37-02-331702-summary.md
 
 ./cpu/README.md
-./cpu-memory/README.md
+./compute/cpu-memory/README.md
 
 ./performance/README.md
 ./performance/hardware.md

compute/accelerator/README.md

@@ -64,7 +64,7 @@ The two metrics for a high ROI for ML training are:
 1. the speed at which the training will finish, because if the training takes 2-3x longer than planned, your model could become irrelevant before it was released - time is everything in the current super-competitive ML market.
 2. the total $$ spent to train the model, because if the training takes 2-3x longer than planned, you will end up spending 2-3x times more.
 
-Unless the rest of the purchased/rented hardware isn't chosen carefully to match the required workload chances are very high that the accelerators will idle a lot and both time and $$ will be lost. The most critical component is [network](../network), then [storage](../storage/), and the least critical ones are ([CPU](../cpu) and [CPU memory](../cpu-memory)).
+Unless the rest of the purchased/rented hardware isn't chosen carefully to match the required workload chances are very high that the accelerators will idle a lot and both time and $$ will be lost. The most critical component is [network](../network), then [storage](../storage/), and the least critical ones are ([CPU](../cpu) and [CPU memory](../compute/cpu-memory)).
 
 If the compute is rented one usually doesn't have the freedom to choose - the hardware is either set in stone or some components might be replaceable but with not too many choices. Thus there are times when the chosen cloud provider doesn't provide a sufficiently well matched hardware, in which case it's best to seek out a different provider.