We begin with core voltage again, which climbs to an average of 1.066 V compared to the stock frequency. That's the motherboard's automatic response to elevated demands; we didn’t manually adjust the firmware's voltage setting.

Power Draw
We again compare the values from the VRM sensor to those measured in parallel at the motherboard input, calculating the losses.

An idle measurement of 18 W (that's 22 W, counting losses) and load reading of 108 W (or 121 W with losses added in) at 3.5 GHz is perfectly acceptable for a processor rated at 140 W.
| Power Consumption | Average Idle | Maximum, 100% Load | Average, 100% Load |
|---|---|---|---|
| CPU 12 V In | 22 W | 141 W | 121 W |
| CPU Package | 18 W | 110 W | 108 W |
| VRM Loss | 4 W | 31 W | 13 W |
Temperatures
Naturally, our thermal readings are low at idle. Under load, they look like this:

Let’s take a look at the time-lapse video, too.
| Temperature T | Idle | Maximum, 100% Load | Average, 100% Load (Heated Up) |
|---|---|---|---|
| Core | 27 °C | 53 °C | 45 °C |
| Package | 29 °C | 46 °C | |
| Water (In / Out) | 24 °C / 27 °C | 32 °C | |
| VRM | 34 °C | 47 °C |
Six Cores At 3.5 GHz
Since our Core i7-5930K was in California with Chris, Igor deactivated two cores on his -5960X and adjusted his maximum Turbo Boost frequency to match the second-fastest Haswell-E processor. The CPUs are practically identical apart from the somewhat smaller cache, so the results should be comparable.
Core Voltage
A 1.072 V core voltage is a bit higher than before due to the higher Turbo Boost clock rate.

Power Draw
Once again, the values from the VR sensor are compared to those measured in parallel at the motherboard input, and the losses are calculated.

A reading of 16 W (with voltage regulator losses, 20 W) at idle and 84 W (with VR losses, 94 W) under load, the six-core adaptation uses a bit less power.
| Power Consumption | Average, Idle | Maximum, 100% Load | Average, 100% Load |
|---|---|---|---|
| CPU 12 V In | 20 W | 113 W | 94 W |
| CPU Package | 16 W | 86 W | 84 W |
| VRM Loss | 4 W | 27 W | 10 W |
Temperatures
Our thermal measurements under load yield the following chart:

| Temperature T | Idle | Maximum, 100% Load | Average, 100% Load (Heated Up) |
|---|---|---|---|
| Core | 27 °C | 48 °C | 43 °C |
| Package | 28 °C | 43 °C | |
| Water (In / Out) | 24 °C / 27 °C | 31 °C | |
| VRM | 33 °C | 44 °C |
At 3.5 GHz, both CPUs (but especially the six-core configuration) give us a good impression of an architecture we might not have expected to fare as well. Haswell-E is emerging as a solid foundation for a gaming machine that can be cooled well using air or liquid.
- Three New CPUs For Enthusiasts
- X99, LGA 2011-3 and DDR4: Get Ready For A Big Upgrade
- How We Tested Core i7-5960X, -5930K, And -5820K
- Synthetic Benchmarks
- Real-World Benchmarks
- Battlefield 4, Grid 2, And Metro: Last Light
- Star Swarm, Thief, Tomb Raider, And WoW
- Power, In Depth: Stock Clock Rates
- Power, In Depth: Eight and Six Cores at 3.5 GHz
- Power, In Depth: Eight and Six Cores at 4 GHz
- Power, In Depth: Eight and Six Cores at 4.5 GHz
- Power, In Depth: CPU Health at 4.8 GHz
- Measuring DDR4 Power Consumption
- Power Consumption Through Our Benchmark Suite
- Intel Keeps Enthusiasts On Its Most Modern Design With Haswell-E

Personally, the 3DS and After Effects benchmarks were of most interest, since they are what I spend most of the CPU time on. (3DS in particular, right now I'm logging dozens of CPU hours a day on 3DS alone). It's pretty clear that unless the platform costs of Haswell-E are much higher than IB-E, going with the old won't make sense. The 5930k beats the 4960X. which is at least 50% more expensive.
I've been waiting forever for an upgrade to my i7 930 based workstation, and I didn't feel like jumping on an IB-E a couple months before a brand-new HEDT platform is released.
I had hoped Haswell-E would be a bit more impressive, but OTOH, investing in a DDR4 platform now might be a good idea, given my workstations typically have 3-4 years in them. At the very least, a drop-in upgrade to Broadwell-E would be nice to have as an option.
Now to see how big a pounding I'll take in Denmark for X99/DDR4/Haswell-E...
Therefore anybody who's going to load up on GPUs enough to worry about PCI-E lanes will have sufficient money to drop in a 5960X on principle. Anybody who's adopting X99 for productivity purposes will not skimp on core count and also go 5960X, especially considering they're likely to go at least 32GB RAM and therefore shelling out a lot of money. Those producing on CUDA cards may not even go X99 at all because 1150 Haswell has more than enough power to run the software. Folders and CUDA Miners similarly will want all GPUs running at full tilt so will likely invest in the 5960X to get all the PCI-E lanes.
So really, the only "smart choice" is 5960X or don't go X99 at all.