1. Adaptive Blocks for Eliminating Waste in the Memory Hierarchy
Amoeba-Cache
Adaptive Blocks for Eliminating Waste in the Memory Hierarchy
Snehasish Kumar
Arrvindh Shriraman
Eric Matthews
Lesley Shannon
Hongzhou Zhao
Sandhya Dwarkadas
I am going to be talking about the Amoeba-Cache and how we eliminate waste in the memory hierarchy by using variable granularity cache blocks
Amoeba Cache : Adaptive blocks for Eliminating Waste in the Memory Hierarchy

2. Cache data utilization
Tag Array
Data Array
Here is a look at a fixed granularity cache where the data accessed by the application during the lifetime of the blocks is highlighted. We can see there exists waste in the form of untouched data in the cache blocks.
Untouched
Data
Tags
Data
Amoeba Cache : Adaptive blocks for Eliminating Waste in the Memory Hierarchy

3. Variable Granularity Blocks
Tag Array
Data Array
How to support variable # of blocks / set ?
How to support variable granularity for each block?
Now let us imagine for a minute that we were able to eliminate the waste from the cache blocks.
We now have a non uniform structure as you can see on the right. In order to maintain precise information about the need extra meta data in the form of more tags. Also note the sets are no longer equal in size
2 Key questions are
Amoeba Cache : Adaptive blocks for Eliminating Waste in the Memory Hierarchy

4. Our Approach : Amoeba Cache
Unified SRAM Array
Amoeba Cache : Adaptive blocks for Eliminating Waste in the Memory Hierarchy

5. Target Metrics Bandwidth Amoeba Cache Space Miss Utilisation Rate
We can visualise the 3 issues in this fashion, there has been previous work which target certain aspects but only the amoeba cache targets all of them at the same time.
Space
Utilisation
Miss
Rate
Amoeba Cache : Adaptive blocks for Eliminating Waste in the Memory Hierarchy

6. Results Reduces L1 and L2 miss rate by 18%
Reduces on-chip bandwidth by 46%
Reduces off-chip bandwidth by 38%
Reduces energy by 11%
Improves performance by 10%