Computer Specs

Computer 1 (Brownie)

Here you can see the specs for Brownie, one of the computers used in the benchmarking tests.

Processor

Key	Value
Manufacturer	Intel
Product	Core i5-8300H

Relevant details viewable on WikiChip.

Of specific note are the 4 available cores (8 threads via hyperthreading). The processor runs at a base frequency of 2.3 GHz with boost scaling as such:

Turbo Freqency [GHz]	Cores
4.0	1
4.0	2
3.9	3
3.9	4

The L1D (D for data) cache is 32 KiB which is shared by 2 threads per core and the unified L2 cache has 256 KiB available per core. Lastly, the L3 has up to 2 MiB available per core wich can be shared between all cores.

Below you can see the latency pyramid for fastest load-to-use (simple pointer accesses):

Memory	Latency [cycles]
L1D	4
L2	12
L3	42

Compared to RAM, data is loaded from L1D via 64 byte cache lines. This leads to the following theoretical L1D bandwidths:

Memory	Bandwidth Base Frequency [GB/s]	Bandwidth Turbo Frequency [GB/s]	#Active Cores
L1D	147.2	256	1
L1D	294.4	512	2
L1D	441.6	748.8	3
L1D	588	998	4

While the AVX2 instruction set is supported, AVX512 is not. The chip integrates Intel's UHD Graphics 630 but in our benchmarks we usually forego this integrated GPU.

Graphics

Key	Value
Manufacturer	NVIDIA
Product	GeForce GTX 1050 Ti with Max-Q Design

It features 4 GB of dedicated GPU memory. The memory modules have a frequency of 1752 MHz. Each GDDR5 cycle transfers 4 bytes which leads to an effective datarate of 7.008 GB/s per pin. As the memory bus is 128 bit wide, we have therefore a theoretical bandwidth of 112.128 GB/s. But this theoretical bandwidth is only useful if the memory already is on the device. For host-device transfers data goes through a PCI-Express 3.0 x16 interface, which has a theoretical bandwidth of 15.5 GB/s.

Running NVIDIA's publicly available bandwidthtest.cu bandwidth measuring program, yields the following results:

Device 0: NVIDIA GeForce GTX 1050 Ti with Max-Q Design
Quick Mode

Host to Device Bandwidth, 1 Device(s)
PINNED Memory Transfers
Transfer Size (Bytes)        Bandwidth(GB/s)
32000000                     12.6

Device to Host Bandwidth, 1 Device(s)
PINNED Memory Transfers
Transfer Size (Bytes)        Bandwidth(GB/s)
32000000                     11.8

Device to Device Bandwidth, 1 Device(s)
PINNED Memory Transfers
Transfer Size (Bytes)        Bandwidth(GB/s)
32000000                     94.9

These values are reasonably close to the aforementioned limits. The theoretical FP64:FP32 FLOP rate is 1:32 for this consumer graphics card.

Memory

Key	Value
Manufacturer	Micron
Product	8ATF1G64HZ-2G6E1

Obtained via:

$ wmic memorychip list full

The system is outfitted with 2x 8 GB RAM sticks for a joint memory of 16 GB. Given a clock speed of 1333MHz (which translates to 2666 MT/s of 64 bit values) we get a theoretical bandwidth of 21.328 GB/s per RAM stick. This is the highest bandwidth memory the Core i5-8300H supports. The CPU supports two of these, so the overall theoretical maximum bandwidth is twice the amount, topping at 42.656 GB/s.

Storage

Slot 0

Key	Value
Manufacturer	Western Digital
Product	SN720
Size (Gross)	512 GB

Slot 1

Key	Value
Manufacturer	Crucial
Product	P3
Size (Gross)	1 TB