| Author |
Message |
Guest
|
 Posted:
Thu Nov 24, 2005 1:15 am Post subject:
Sun's Ultrasparc T1 threads vs. Intel Hyperthreading? |
|
|
Despite the hype, hyperthreading on Intel hasn't lived up to it's
grandiose promises.
Now Sun is announcing its new Ultrasparc T1 chip: 8 cores with 4
threads each. Sun's marketing literature would have you believe that
this is in effect a 32 processor SMP system.
1) How is Sun's threads different from Intel's hyperthreading?
2) Just what is a thread in the Sun and Intel processors? How much
functional unit duplication is there?
Thanks,
Joe
Joe McGuckin |
|
| Back to top |
|
 |
David Kanter
Guest
|
| Posted:
Thu Nov 24, 2005 1:15 am Post subject:
Re: Sun's Ultrasparc T1 threads vs. Intel Hyperthreading? |
|
|
Intel's Hyper-threading is SMT, Sun's Niagara uses a combination of
time-slice and switch on event multithreading.
The difference is that the P4 is a 3-issue/execute/retire CPU, while
Niagara is upto 8 single issue CPUs. Each CPU hosts 4 threads, which
interleave across time, until one experiences a long latency event
(i.e. fetch to memory), then it removes itself from the rotation until
the data is ready.
In SMT, functional units are shared dynamically between two threads.
In SoEMT, the chip is assigned to a given thread each cycle.
Use google to find out more.
DK |
|
| Back to top |
|
|
Jason Hu
Guest
|
| Posted:
Thu Nov 24, 2005 1:15 am Post subject:
Re: Sun's Ultrasparc T1 threads vs. Intel Hyperthreading? |
|
|
| Quote: | From my understanding, Hyperthreading is just let another thread use
idle functional unit resource. No functional unit duplication. |
In Sun's Ultrasparc T1, each thread has its own functional units and
architecture register. |
|
| Back to top |
|
|
Guest
|
| Posted:
Thu Nov 24, 2005 12:10 pm Post subject:
Re: Sun's Ultrasparc T1 threads vs. Intel Hyperthreading? |
|
|
joe@via.net wrote:
| Quote: | Despite the hype, hyperthreading on Intel hasn't lived up to it's
grandiose promises.
|
Intel's implementation suffered from adding SMT but not making all the other
changes needed such as increasing cache sizes, making the caches aware of
the SMT and balancing usage between virtual processors etc.
The performance degredation associated with Hyper Threading is specific to
Intel's implemenation and isn't a characteristic of SMT in general. If done
right it may not have any advantage in many cases but shouldn't have any
pitfalls.
-p
--
"What goes up must come down, ask any system administrator"
-------------------------------------------------------------------- |
|
| Back to top |
|
|
Oliver S.
Guest
|
| Posted:
Tue Nov 29, 2005 9:15 am Post subject:
Re: Sun's Ultrasparc T1 threads vs. Intel Hyperthreading? |
|
|
| Quote: | In Sun's Ultrasparc T1, each thread has its own functional units and
architecture register.
|
That's not right. Of course each thread has its own register-file,
but the functional-unit(s) are mapped per core to four threads. |
|
| Back to top |
|
|
Sander Vesik
Guest
|
| Posted:
Tue Nov 29, 2005 10:25 pm Post subject:
Re: Sun's Ultrasparc T1 threads vs. Intel Hyperthreading? |
|
|
Oliver S. <Follow.Me@gmx.net> wrote:
| Quote: | In Sun's Ultrasparc T1, each thread has its own functional units and
architecture register.
That's not right. Of course each thread has its own register-file,
but the functional-unit(s) are mapped per core to four threads.
|
It depends... There are 8 threads at a time that all have their own
functional units. The "other" 24 don't though.
--
Sander
+++ Out of cheese error +++ |
|
| Back to top |
|
|
|
|
|
|
FAQ
Memberlist
Register
Profile
Log in to check your private messages
Log in