Unsolved
This post is more than 5 years old
8 Posts
0
11955
July 20th, 2003 03:00
PowerEdge 400SC Hyperthreading ?
Does 400SC's BIOS support HyperThreading if I pick a P4 3.2GHz CPU ?
If yes, is there any switch in BIOS can turn it on or off ?
Thank you very much for any information you can offer !
Message Edited by hpu on 07-19-2003 11:52 PM
No Events found!



Krumlauf
3 Posts
0
July 21st, 2003 01:00
hpu
8 Posts
0
July 22nd, 2003 17:00
From the Intel's data
http://www.intel.com/products/desktop/processors/pentium4/index.htm?iid=ipp_compare+desk_dsk_p4&
It seems like all 800FSB P-4 CPUs do support HyperThreading.
In this case, does it mean that 400SC will enable hyperthreading if I pick any one of those 800FSB cpu from the configuration purchase page ?
Thanks for any information you can offer !
JoePU
1 Message
0
July 23rd, 2003 13:00
Dell, can you confirm that any P4 running at 800FSB will support Hyperthreading. According to the previous post, Intel says that the 2.8Ghz P4 supports Hyperthreading. Notice this is under the 3Ghz limit as posted on some other threads.
Thanks,
Joe
cft
3 Posts
0
July 30th, 2003 07:00
Even one of our 2.4GHz supports HT but the question is we are getting mixed info from MS & Dell. MS says W2K (all versions) doesn't support HT but Dell says HT boost performance as much as 30%. So, to play it safe, we have it disabled.
maxusa
63 Posts
0
August 21st, 2003 13:00
I can confirm that a 2.8GHz P4 (FSB800) enables Intel Hyper-Threading on the Dell 400SC. I have it working fine in my 400SC.
The claim of 30% performance boost is exaggerated. The symmetric multi-threading (SMT) technology does improve parallel threading under certain conditions. The reality is that most of today's applications are not built to take advantage of that, as multi-threading is harder to code. Plus, as I mentioned, the virtual CPU becomes truly useable only under certain circumstances. When you put together #1 and #2, the result is that the Hyper-Threading actually is about 2% slower because of the system management overhead. It is especially visible in games.
The true kick comes from symmetric multi-processing (SMP), which you can attain with Intel Xeon/XeonMP and/or AMD AthlonMP/Opteron platforms. SMP makes parallel execution truly parallel. If the software was built to use multiple threads for CPU-intensive tasks, then it may theoretically run upto twice as fast.
Hope it helps.
rokicki
3 Posts
0
August 21st, 2003 17:00
That is, running two experiments serially with hyperthreading disabled on a uniprocessor kernel takes 100 seconds (50 seconds each). Running those two same experiments in parallel with hyperthreading enabled on an SMP kernel takes 80 seconds (80 seconds each, but at the same time). Running the two experiments serially with hyperthreading enabled on an SMP kernel took time that was the same as with hyperthreading disabled on a uniprocessor kernel.
YMWV so you have to test the application you care about. Personally this works out well for me; sometimes I want a particular result fast, in which case I run it alone; other times I want a bunch of results overnight in which case I batch them up two at a time in parallel.
theJapester
1 Rookie
•
42 Posts
0
January 7th, 2004 01:00
Like most things, the answer is, "It depends".
Most apps aren't SMP aware, they get no boost at all unless you are running two of them side by side and the OS happens to assign each process on a different virtual processor. This is what Intel is trying to market with the hyperthreaded P4's for desktop PCs: Theoretically, your MP3 or MPEG encoder shouldn't starve the rest of your system of CPU cycles. (YMMV.)
OTOH, an SMP-aware database app like SQL Server running under Windows 2003 will see both virtual processors, and you can see as much as a 20%-25% performance boost under the right type of load. We're talking multiple processor-heavy queries or a parallelized query plan with high processor loads.
The real advantage is that SQL Server and Oracle still license the hyperthreaded P4s as single CPUs, which lets you get a small -- but measurable -- performance boost without paying additional per-processor database licensing fees. Pretty nice for departmental database servers running real DBMSs.
Message Edited by theJapester on 01-06-2004 09:06 PM