1 |
B Vance posted <1138205784.11529.1.camel@×××××××××××.ShadowRealm>, |
2 |
excerpted below, on Wed, 25 Jan 2006 11:16:24 -0500: |
3 |
|
4 |
>> virtual cpus? |
5 |
>> |
6 |
>> intel has hyperthreading, if you mean that, not amd. |
7 |
>> |
8 |
>> the rule of thumb is cores+1 |
9 |
>> |
10 |
>> but it is totally personal taste, what you set for jX. |
11 |
>> 1,2,3,4,5 just try it and choose the number where the result is most to your |
12 |
>> liking. |
13 |
> |
14 |
> Cool, thanks. Must of been thinking of something else that AMD had |
15 |
> licenced from Intel for core CPU use. Could have sworn that AMD had a |
16 |
> version of Hyperthreading. |
17 |
|
18 |
Maybe you are thinking about (the unrelated) HyperTransport? That's an |
19 |
all-AMD thing, the reason they seem to have such good performance compared |
20 |
to Intel, which is stuck with legacy front-side-bus architecture for the |
21 |
moment, with it throwing a /serious/ wrench into their upgrade plans ATM. |
22 |
|
23 |
As for Hyperthreading, AMD hasn't used it and really doesn't need or want |
24 |
to. Hyperthreading works its magic by working around the extremely deep |
25 |
pipeline that Intel had used, in ordered to get their clock rates so high. |
26 |
Such a deep pipeline forces a *HUGE* penalty when there's a branch |
27 |
miss-prediction and the CPU has to throw away several cycles worth of work |
28 |
and let the bottom end of the pipeline idle while it loads the correct |
29 |
instructions from the top of the pipeline again. Hyperthreading is a |
30 |
rather neat solution that avoids the severe penalty by having a second |
31 |
task ready and waiting to execute while the first one gets loaded into the |
32 |
pipeline once again. |
33 |
|
34 |
AMD has a somewhat shorter pipeline, and uses some other technologies to |
35 |
improve the branch prediction and lessen the severity of mis-prediction |
36 |
hits, both. It isn't /that/ much shorter a pipeline, the branch |
37 |
prediction isn't /that/ much better, and the miss-prediction penalty not |
38 |
/that/ much less, but put all three together, and add in the effect of the |
39 |
front-side-bus vs hypertransport on memory latency, and the individual |
40 |
single-digit improvements begin to add-up to something rather bigger. |
41 |
Hyperthreading only yields an approximate 30% increase in performance, |
42 |
best-case, anyway, compared to a theoretical 100% increase if it were a |
43 |
real CPU. Thus, AMD doesn't have to cover as much ground with their |
44 |
individual "minor" improvements as one might think, and because |
45 |
hyperthreading has certain downsides (effectively halving the available |
46 |
cache per thread, and certain thread-cache-access security issues where |
47 |
the hyperthread has access to the memory of the other thread), they |
48 |
believe they do better without hyperthreading. |
49 |
|
50 |
I believe AMD does better without hyperthreading as well, from a technical |
51 |
viewpoint, altho Intel has the PR game won with hyperthreading, as it's |
52 |
simply not possible to explain in two sentences why AMD doesn't need |
53 |
hyperthreading, as opposed to the relatively simple concept Intel has to |
54 |
get across, of pretending there are two CPUs so the second task can get |
55 |
some work done when the first task is waiting for something. |
56 |
|
57 |
-- |
58 |
Duncan - List replies preferred. No HTML msgs. |
59 |
"Every nonfree program has a lord, a master -- |
60 |
and if you use the program, he is your master." Richard Stallman in |
61 |
http://www.linuxdevcenter.com/pub/a/linux/2004/12/22/rms_interview.html |
62 |
|
63 |
|
64 |
-- |
65 |
gentoo-amd64@g.o mailing list |