Anyone tried a Core i7-4770 v Core i7-4770K yet?

Patrick

Administrator
Staff member
Dec 21, 2010
12,353
5,488
113
Been playing with this in Windows a bit. The Core i7-4770 and Core i7-4770K are performing almost identically with the K part sometimes being a bit faster. Maybe transaction memory isn't working in live code yet?

TrueCrypt AES7-ZipCinebench R11.5Power Consumption
TrueCrypt 7-Zip BenchmarkCinebench R11.5Min PowerMax PowerMin-Max Power
CPU(1GB) AES SpeedMIPSCPU TestwattswattsDelta in watts
Core i7-4770K4.3GB/s220068.15199677
Core i7-47704.4GB/s218328.14199576


Averages over 15 runs each.

To be expected given the 3.9GHz base clock speed, but good validation that for a desktop, unless you are overclocking the Core i7-4770's extra enabled Haswell features do not mean it is much faster than the Core i7-4770K.
 

Salami

New Member
Oct 12, 2012
31
1
0
The benchmark program would likely have to be reprogrammed and recompiled for these types of new features of the CPU. The CPU gives you new "instructions" you can use, so either the compiler or the programmer has to use them.

Also, it is my understanding that the transaction memory logic is designed to help applications that are multi-threaded and operate on the same set of data, like a database server or a scientific application. The benchmark programs may be multi-threaded but each thread is independent, so no transactions.
 

Patrick

Administrator
Staff member
Dec 21, 2010
12,353
5,488
113
The benchmark program would likely have to be reprogrammed and recompiled for these types of new features of the CPU. The CPU gives you new "instructions" you can use, so either the compiler or the programmer has to use them.

Also, it is my understanding that the transaction memory logic is designed to help applications that are multi-threaded and operate on the same set of data, like a database server or a scientific application. The benchmark programs may be multi-threaded but each thread is independent, so no transactions.
Very true. I have just been hearing people opine that the K model was going to be slightly slower clock-for-clock than the non-K model due to missing features even in current applications. Seems as though this is not going to be the case.