I wasn't too much impressed about the DPC latency comparison in the link you provided. There seems something else is going on because DPC latency figure is not truely CPU dependent. It's rather dependent on the motherboard's chipset, drivers (for peripherals and GPUs), and some OS tasks going on. In fact, in that test, none of them used the same board with different CPU (or the same CPU on different boards). So the tester is comparing apples and oranges.

Additionally, my i7 950 with X58 board gives me 40-50 microseconds, and AMD Phenom II with GA880 board gives me even better 35-40 microseconds DPC figures, with no spikes, flat for several hours. These CPUs are a couple of generations old, and much slower than these 3770K or 3930K processors, yet achieve lower DPC latency than the numbers in the test. It's all about how to tweak the BIOS and OS tasks.

Probably the tester in the link didn't spend time to tweak the system, and that's what I guess.
In terms of DPC latency, I suggest to research more around the motherboard/BIOS than the choice of CPU itself. In that sense, you can choose either CPU, they are very powerful/fast CPUs.

I don't believe the difference between these two processor will be reflected in the baseline DPC latency figures, because DPC latency by itself is dependent on how the task is prioritized.

In my hands ASIO performance and CPU performance is a separate issue. More important factor is how ASIO drivers is written for a given interface.

[center]MY DAW AND MY FF800 MOBO IS GA-Z68X-UD3H-B3 GIGABYTE[/center]
[img align=c]http://www.xstreme.it/ltn1.jpg[/img]

DPC Latencies are CPU dependent in that CPU power-saving states affect them and in that each CPU core can serve one interrupt call on its own.

The former means that BIOS power settings, Windows power-profile settings and especially overall CPU load affect DPC readings by DPC Latency Checker. You often can get lower readings by permanently loading the CPU via low priority load (Prime95). And a weaker CPU may run less in power-saving CPU states when faced with the same load compared to a stronger CPU. So DPC readings may be well off, while the stronger CPU usually still handles the load better.

The latter means that the more cores are free when DPC Latency Checker issues a DPC to the OS the quicker the CPU can answer the corresponding interrupt request by the OS.

Personally I do not worry about anything around 100 - 150 us, and if I want I can lower my readings towards 30 us or so. DPC Latency Checker is a tool for indicating possible culprits, nothing more and nothing less.

To make things more complicated, each process that issues DPCs may run at a different priority. And while DPCs all share the same priority (higher than realtime priority 31) the corresponding processes/threads may get less CPU time to even issues a DPC before any high priority audio thread issues its next DPC first (higher priority thread = more CPU time to do stuff like issuing DPCs).

Just to mention it: OS X doesn't work with DPCs (which in essence issue interrupt calls more in bulk), but applications can still suffer from the same CPU power-saving culprits. And while it always sounds as if the CPU is doing all the power-saving work, in reality it's the OS handling all the power-states, core parking and clock-rates via its power-schemes. Windows allows full control over these things once you mess with some Registry entries and then create your own power-profiles.

A CPU core in deep sleep state needs time to wake up before it can handle an interrupt request. Turning off those states completely isn't a realistic option, though, because that will run your CPU hot like an oven and may even force it to downclock in order to get rid of the heat. Not to mention fan noise and power consumption. Instead the better way is to change the power-profile (and some CPU registers) to a less aggressive power-saving methods, like waking CPU cores at lower load already, disabling only certain power-states (C1E) via BIOS/third party software and trying around with different Core Parking settings.

While we are at it: Core Parking again is not so much a CPU feature, but an OS function. What happens is that the OS shifts CPU/core affinity of threads to single CPU cores in order to keep other cores free. Since these other cores have nothing to do then they are allowed to enter deep sleep states. It effectively turns your 8 core CPU into a 4 core CPU in order to save power. Those remaining 4 cores have to work harder, of course, which is when Turbo Boost increases the clock-rate.

I guess the OP's question is, does he get better performance of ASIO buffer settings by upgrading from i7 920 to a newer CPU such as Sandy 2700K, even his CPU usage is 15-20%?