Intel's new processors how come they don't do hyperthreading

[Glavinsolo]

It might have been 5 years ago that I actually could have said I knew everything about the pc industry back when I made a living off of selling pcs, now that I work in IT I haven't followed as closely.

I am trying to imagine why Intel gave hyperthreading the back seat for their new line of processors. I'd imagine dual core that supports hyperthreading would produce better results as a virtual cpu for each core allowing 4 cores.

I know that when I built servers we used to disable hyperthreading since we needed to pay a license fee for each processor identified under windows, so maybe this is the case, for their business xeon builds maybe?

Anyhow anyone else have a clue or fact as to why intel no longer has their procs support hyperthreading?

 

[myocardia]

The only reason Intel came up with HyperThreading is because the P4 had such a long pipeline, it was actually being under-utilized with a single thread. Their new processors have a much wider, much shorter pipeline, and don't need HT anymore.

 

[Viditor]

Originally posted by: Glavinsolo

I know that when I built servers we used to disable hyperthreading since we needed to pay a license fee for each processor identified under windows

Ummm...windows (and every software company I know of) charges per socket, not per processer.

 

[sandorski]

Dual Core>>>>HT. There were other reasons, but Dual Core basically makes HT obsolete.

 

[bunnyfubbles]

Because adding hyperthreading capabilities doesn't magically increase processor performance. Even on the crappy Netburst architecture that was actually pretty good for hyperthreading due to the long pipeline, hyperthreading still wasn't that great as it offered minimal performance boosts at best and was only notably useful in only a few niche scenarios. Hyperthreading on current Core 2 CPUs would most likely be useless no matter the scenario.

Although Nehalem looks like it will bring a return of HT, we'll have to see what Intel has done (if anything) to make it worthwhile.

 

[zsdersw]

Originally posted by: Viditor
Ummm...windows (and every software company I know of) charges per socket, not per processer.

Windows, in this context, isn't a company.

 

[fleshconsumed]

For all you naysayers, Intel is bringing hyperthreading back in Nehalem.

As far as licensing costs all microsoft products charge per socket, not per core. However, if say, you're running Oracle SQL they will charge you per core with discounts for more cores, something like 1x per first core, 0.5x per second core, etc. Creative at one point had a really stupid EULA which basically said you couldn't run their hardware legally on SMP machines. But back to topic as long as you're running Microsoft OS and MS SQL you do need to disable hyperthreading or second core in C2D because Microsoft only charges per socket.

 

[Viditor]

Originally posted by: fleshconsumed
For all you naysayers, Intel is bringing hyperthreading back in Nehalem.

As far as licensing costs all microsoft products charge per socket, not per core. However, if say, you're running Oracle SQL they will charge you per core with discounts for more cores, something like 1x per first core, 0.5x per second core, etc. Creative at one point had a really stupid EULA which basically said you couldn't run their hardware legally on SMP machines. But back to topic as long as you're running Microsoft OS and MS SQL you do need to disable hyperthreading or second core in C2D because Microsoft only charges per socket.

Actually, even Oracle stopped per core licensing a year ago, except on their >4S Enterprise systems. As far as I know, most all software is now per socket...

 

[IntelUser2000]

The only reason Intel came up with HyperThreading is because the P4 had such a long pipeline, it was actually being under-utilized with a single thread. Their new processors have a much wider, much shorter pipeline, and don't need HT anymore.

Ironically, the Netburst architecture was the reason enabling Hyperthreading resulted in significant performance penalty in certain cases. It was being under-utilized, but the flaws on the architecture meant that the benefits were all vanished.

Here is the explanation on Replay:
"The scheduler in a Pentium 4 processor is so aggressive that it will send operations for execution without a guarantee that they can be successfully executed. (Among other things, the scheduler assumes all data is in level 1 cache.) The most common reason execution fails is that the requisite data is not available, which itself is most likely due to a cache miss. When this happens, the replay system springs into action. The replay system signals the scheduler to stop, and then repeatedly executes the failed string of dependent operations until they have completed successfully."

The problem was that when there was 2 threads execution like in case of Hyperthreading, the CPU might be in "replay" and is hogging up execution resources. If the CPU is already being used up, adding another thread means what?? It meant loss in performance.

One reason that the Core microarchitecture forgoed Hyperthreading was because the design teams' philosophy was different from the designers of Netburst. The Core microarchitecture team went for a more proven approach rather than a radical, all new approach that Netburst went for.

I bet with Nehalem, Hyperthreading will be seen in a much better light than it ever was with Pentium 4. Even with Pentium 4, HT seemed very good until Prescott and Pentium D's came and completely trashed it.

 

[Modelworks]

Hyperthreading is far from useless.
Or a better name for what it really is smt
http://en.wikipedia.org/wiki/S...taneous_multithreading

Hyperthreading is just intels buzzword.