I originally went into this testing specifically trying to determine whether or not overclocking would increase the strain enough on Haswell?s memory controller to justify higher speed memory. In testing, I discovered fairly conclusively that DDR3-1600 essentially leaves performance on the table even at stock clocks.
For testing I ran Intel?s Core i7-4770K at stock speeds and overclocked to 4.5GHz. A 32GB (4x8GB) kit of our Dominator Platinum DDR3-2400 was used to scale from DDR3-1600 CAS 9 to DDR3-2400 CAS 10. Test system specs are as follows:
Intel Core i7-4770K CPU
Stock Speed (3.5GHz nominal, turbo to 3.7GHz on four cores or 3.9GHz on one core)
Overclocked (4.5GHz, 45x100 BClk, 4GHz Northbridge
Gigabyte G1.Sniper 5 Motherboard
4x8GB Corsair Dominator Platinum DDR3-2400
DDR3-1600 (9-9-9-24 CR2)
DDR3-1866 (9-9-9-24 CR2)
DDR3-2133 (10-11-11-31 CR2)
DDR3-2400 (10-12-12-32 CR2)
NVIDIA GeForce GTX 780 Overclocked (980MHz nominal, boost to 1150MHz, 7GHz GDDR5)
240GB & 480GB Neutron GTX SSDs (for Adobe testing)
I very deliberately chose a mixture of synthetic and real world benchmarks. Cherry picked synthetics can admittedly overstate the importance of higher speed memory; I wanted tangible, demonstrable, practical benefits.
Overclocking the CPU itself had virtually no effect on memory bandwidth, producing results essentially within the margin of error. We?re just going to measure the raw amount of bandwidth made available to the i7 as memory clocks increase.
The Hail Mary in the group is StarCraft II: Heart of the Swarm. Testing was done at 1080p with all of the settings maxed out, so it at least looks the same way it might on your home system
Comment