Hyperthreading Hype predates Intel

[sje]

Hyperthreading Hype (or Distortion or Untruth or whatever) predates Intel by years.

Back in the early 1960s Seymour Cray, then working for Control Data, designed a family of 60 bit mainframes. The CPU design had as its main goal the need to be the world's fastest floating point calculator. To assist with overall floating point throughput, Cray had the CPU's memory dual ported to be available as an I/O space that appeared to a set of PPUs (Peripheral Processing Units), and had all physical I/O (disks, tapes, etc.) be handled by the PPUs. Each PPU was a twelve bit machine somewhat similar to a DEC pdp-8. In one CDC mainframe that I worked with, there were ten of these PPUs.

Well, not really. The ten PPUs were really only one set of processor registers that accessed ten different banks of memory. Back then the memory (magnetic core) was very slow relative to arithmetic function register speed. So slow that Cray had a ten way uniform interval timeslice multiplexor on the PPU memory and data buses. CDC advertised the scheme as ten different processors.

I wonder if Intel has ever given Cray any credit for inventing hyperthreading hype.

[Matthias Gemuh]

Hyperthreading Hype ?
Hyperthreading is not hype .
It gave me a second processor when I had only one ! Great stuff.

Best,
Matthias.

[Spock]

Indeed, hyperthreading was very useful. I recall my P4 machine being much more responsive than my Athlon at the time. Although not a true dual core by any means, it DID help make a more responsive machine for multitasking and general use. As we know the P4 wasn't quick at chess, but for everything else it was streets ahead of my AMD boxes at the time

[bob]

Hyperthreading Hype ?
Hyperthreading is not hype .
It gave me a second processor when I had only one ! Great stuff.

Best,
Matthias.

It gave you a very weak second processor and slowed the primary processor by almost 50% to boot...

[Matthias Gemuh]

Hyperthreading Hype ?
Hyperthreading is not hype .
It gave me a second processor when I had only one ! Great stuff.

Best,
Matthias.

It gave you a very weak second processor and slowed the primary processor by almost 50% to boot...

There's always a price tag indeed, but the much more responsive machine is a net gain for a machine with high user interactivity.

Matthias.

[lmader]

It gave you a very weak second processor and slowed the primary processor by almost 50% to boot...

This is a blanket statement and isn't accurate. Obviously it depends on the application. Try 7-Zip on a long archive creation or extraction. Waaay faster. Try Bob's chess engine... slower. Compare system responsiveness - better, etc., etc.

[bob]

Hyperthreading Hype ?
Hyperthreading is not hype .
It gave me a second processor when I had only one ! Great stuff.

Best,
Matthias.

It gave you a very weak second processor and slowed the primary processor by almost 50% to boot...

There's always a price tag indeed, but the much more responsive machine is a net gain for a machine with high user interactivity.

Matthias.

I've got one and I can't feel the "more responsiveness". I have a dual PIV xeon that supports hyper-threading (I have it disabled) and use two physical processors only. If you turn SMT on, you end up with two logical processors for each physical processor, but each logical processor is around 1/2 as fast as a normal physical processor. I'm not sure how that would make it feel "more responsive" and when I enable it on mine, I can't tell the difference and I run all kinds of stuff all the time on it...

[bob]

It gave you a very weak second processor and slowed the primary processor by almost 50% to boot...

This is a blanket statement and isn't accurate. Obviously it depends on the application. Try 7-Zip on a long archive creation or extraction. Waaay faster. Try Bob's chess engine... slower. Compare system responsiveness - better, etc., etc.

I have a PIV that supports SMT. I don't see any "more responsiveness" at all, although I am running Linux which has an extremely good process scheduler...

[Matthias Gemuh]

I've got one and I can't feel the "more responsiveness". I have a dual PIV xeon that supports hyper-threading (I have it disabled) and use two physical processors only. If you turn SMT on, you end up with two logical processors for each physical processor, but each logical processor is around 1/2 as fast as a normal physical processor. I'm not sure how that would make it feel "more responsive" and when I enable it on mine, I can't tell the difference and I run all kinds of stuff all the time on it...

Because you have a "dual", you cannot know how annoying a "single" can be.
On a "single" without HT, if a chess tournament is running, simple tasks like unpacking an archive can either take very long or even freeze the system.

Matthias.

[bob]

I've got one and I can't feel the "more responsiveness". I have a dual PIV xeon that supports hyper-threading (I have it disabled) and use two physical processors only. If you turn SMT on, you end up with two logical processors for each physical processor, but each logical processor is around 1/2 as fast as a normal physical processor. I'm not sure how that would make it feel "more responsive" and when I enable it on mine, I can't tell the difference and I run all kinds of stuff all the time on it...

Because you have a "dual", you cannot know how annoying a "single" can be.
On a "single" without HT, if a chess tournament is running, simple tasks like unpacking an archive can either take very long or even freeze the system.

Matthias.

First, get a better O/S. Linux _never_ "freezes".

But whenever you run more tasks than processors, things are going to get slow. Hyper-threading just makes them slower...