Early Benchmarks Of The Intel Core i7 7700K On Linux

[waxhead]

It light of the discussion here I wonder if Phoronix will ever do a follow-up story on this https://www.phoronix.com/scan.php?pa...-Scheduler-Bad

[Michael]

Well, HT can only help maximize the utilization of a core, meaning a core with HT may perform closer to the theoretical maximum for a single core, never above it.

-----

I would love to see a comparison with i7-6800K and i7-6900K, especially considering how pricy i7-7700K is over i5-7600K.

6800k will be included in the big comparison as I had bought that, but I have no 6900k.

[wizard69]

Well, HT can only help maximize the utilization of a core, meaning a core with HT may perform closer to the theoretical maximum for a single core, never above it.

-----

I would love to see a comparison with i7-6800K and i7-6900K, especially considering how pricy i7-7700K is over i5-7600K.

Actually this isn't true either. Well not completely true as you can have weird cases where turning on HT actually lowers overall performance. The usefulness of HT is entirely dependent upon the tasks being executed and the interaction of software and hardware. Sometimes HT is a bad idea if you are after performance.

[wizard69]

It light of the discussion here I wonder if Phoronix will ever do a follow-up story on this https://www.phoronix.com/scan.php?pa...-Scheduler-Bad

Nothing Linux is written in stone. Parts of the kernel get updated as people take interest.

[BillBroadley]

Actually this isn't true either. Well not completely true as you can have weird cases where turning on HT actually lowers overall performance. The usefulness of HT is entirely dependent upon the tasks being executed and the interaction of software and hardware. Sometimes HT is a bad idea if you are after performance.

Agreed. Although generally for code that's well multithreaded it's generally a win. In the P4 era it was often slower, and it was pretty common practice to disable it if you were doing CPU intensive stuff in production. Both the linux kernel and the implementation by intel have improved since then. Generally (but not always) it's a win to turn HT on.

[caligula]

Agreed. Although generally for code that's well multithreaded it's generally a win. In the P4 era it was often slower, and it was pretty common practice to disable it if you were doing CPU intensive stuff in production. Both the linux kernel and the implementation by intel have improved since then. Generally (but not always) it's a win to turn HT on.

HT on P4 is different from the one on more recent architectures. There's not only a single brand of HT. Take a look at POWER CPUs for more exotic configurations. It is really hard to give a simple answer if HT is faster or not. It depends on the task type, the best solution is to do benchmarks.

[efikkan]

Actually this isn't true either. Well not completely true as you can have weird cases where turning on HT actually lowers overall performance. The usefulness of HT is entirely dependent upon the tasks being executed and the interaction of software and hardware. Sometimes HT is a bad idea if you are after performance.

Not quite.
HT lets another thread execute during stalls in the CPU. But as with any feature, this is not without drawbacks, I've never claimed otherwise. HT increases the throughput of a core, not a specific thread or application. This means that HT can let other threads "disturb" another thread by occupying the CPU. This is why HT may cause stutter in gaming if you have a web browser running in the background or similar. The two threads sharing a core might also impact the cache hit rate in extreme cases. But most HT related problems are caused by the user overloading the CPU, so the solution is then more cores.

[lsatenstein]

You got the money, Honey, I got the CPU.

What is the cost/performance ratio? Should we be looking at different measures such as a balance between doing DMA/IO transfer and CPU performance? My feeling is that the 7700 is geared for gamers or for encryption code hackers. The average corporate database system, to my knowledge, already has a balance between CPU cycles and database I/O