Embodiments of the invention describe a system main memory comprising two levels of memory that include cached subsets of system disk level storage. This main memory includes “near memory” comprising memory made of volatile memory, and “far memory” comprising volatile or nonvolatile memory storage that is larger and slower than the near memory.

The far memory is presented as “main memory” to the host OS while the near memory is a cache for the far memory that is transparent to the OS, thus appearing to the OS the same as prior art main memory solutions. The management of the two-level memory may be done by a combination of logic and modules executed via the host CPU. Near memory may be coupled to the host system CPU via high bandwidth, low latency means for efficient processing. Far memory may be coupled to the CPU via low bandwidth, high latency means.

Embodiments of the invention describe a system main memory comprising two levels of memory that include cached subsets of system disk level storage. This main memory includes “near memory” comprising memory made of volatile memory, and “far memory” comprising volatile or nonvolatile memory storage that is larger and slower than the near memory.

The far memory is presented as “main memory” to the host OS while the near memory is a cache for the far memory that is transparent to the OS, thus appearing to the OS the same as prior art main memory solutions. The management of the two-level memory may be done by a combination of logic and modules executed via the host CPU. Near memory may be coupled to the host system CPU via high bandwidth, low latency means for efficient processing. Far memory may be coupled to the CPU via low bandwidth, high latency means.

Provided are an apparatus and method for a non-power-of-2 size cache in a first level memory device to cache data present in a second level memory device having a 2.sup.n cache size. A request is to a target address having n bits directed to the second level memory device. A determination is made whether a target index, comprising m bits of the n bits of the target address, is within an index set of the first level memory device. A determination is made of a modified target index in the index set of the first level memory device having at least one index bit that differs from a corresponding at least one index bit in the target index. The request is processed with respect to data in a cache line at the modified target index in the first level memory device.

Provided are an apparatus and method for a non-power-of-2 size cache in a first level memory device to cache data present in a second level memory device having a 2.sup.n cache size. A request is to a target address having n bits directed to the second level memory device. A determination is made whether a target index, comprising m bits of the n bits of the target address, is within an index set of the first level memory device. A determination is made of a modified target index in the index set of the first level memory device having at least one index bit that differs from a corresponding at least one index bit in the target index. The request is processed with respect to data in a cache line at the modified target index in the first level memory device.

An information processing apparatus includes a memory, and a processor coupled to the memory and configured to count first number indicating storing a plurality of arrays of data to each of cash lines, the data being accessed in accordance with execution of a program, and count second number indicating cache thrashing to the cache lines when the first number exceeds number of ways of cache.

Described herein is a graphics processing unit (GPU) comprising a first processing cluster to perform parallel processing operations, the parallel processing operations including a ray tracing operation and a matrix multiply operation; and a second processing cluster coupled to the first processing cluster, wherein the first processing cluster includes a floating-point unit to perform floating point operations, the floating-point unit is configured to process an instruction using a bfloat16 (BF16) format with a multiplier to multiply second and third source operands while an accumulator adds a first source operand with output from the multiplier.

Described herein is a graphics processing unit (GPU) comprising a first processing cluster to perform parallel processing operations, the parallel processing operations including a ray tracing operation and a matrix multiply operation; and a second processing cluster coupled to the first processing cluster, wherein the first processing cluster includes a floating-point unit to perform floating point operations, the floating-point unit is configured to process an instruction using a bfloat16 (BF16) format with a multiplier to multiply second and third source operands while an accumulator adds a first source operand with output from the multiplier.

Methods, systems, and devices for cache block budgeting techniques are described. In some memory systems, a controller may configure a memory device with a cache. The cache may include a first subset of blocks configured to statically operate in a first mode and a second subset of blocks configured to dynamically switch between operating in the first mode and a second mode. A block operating in the second mode may be configured to store relatively more bits per memory cell than a block operating in the first mode. The controller may track and store, for each block of the second subset of blocks, a respective ratio of cycles performed in the first mode to cycles performed in the second mode. The controller may select a block from the second subset of blocks to switch between modes responsive to a trigger and based on the respective ratio for the block.

Methods, systems, and devices for cache block budgeting techniques are described. In some memory systems, a controller may configure a memory device with a cache. The cache may include a first subset of blocks configured to statically operate in a first mode and a second subset of blocks configured to dynamically switch between operating in the first mode and a second mode. A block operating in the second mode may be configured to store relatively more bits per memory cell than a block operating in the first mode. The controller may track and store, for each block of the second subset of blocks, a respective ratio of cycles performed in the first mode to cycles performed in the second mode. The controller may select a block from the second subset of blocks to switch between modes responsive to a trigger and based on the respective ratio for the block.

An embodiment of a memory controller device includes technology to control access to a multi-level memory including at least a first level memory and a second level memory, provide direct access to the first level memory based on a system memory address, cache accesses to the second level memory in a second portion of the first level memory, and address a memory space with a total memory capacity which includes a first capacity of the first portion of the first level memory plus a second capacity of the second level memory. Other embodiments are disclosed and claimed.