Printable View
I recently found out that there is speculation that 8 cores of the Core i7 may actually be hurting game performance and that turning hyperthreading off actually helps performance in some games.
Link:
http://vr-zone.com/articles/does-cor....html?doc=6160
Are games just not optimised to run on more than 4 cores? Is it the OS (Vista 64)?
Anyone with knowledge on this subject help out?
It is not surprise, coz many of games are optimized still on 2 cores and you have there 4X2 :D
I'm a pretty serious gamer lol, and I've used HT on since I built this i7 beast. Not once have I noticed a deficit in performance. This is just a few games that I play though. Crysis Warhead, FC2, COD5, COD4 and a bit of HL2 and it's episodes...but they all run butter smooth. Some games might not like it, but I have yet to find one.
I noticed some odd behaviour in TF2, when i use HT and have 8 core enabled it theres sometimes a bit stuttering, if all the 8 cores are loaded in the background (boinc) it becomes more noticable.Quote:
Originally Posted by 1337633k
BUT if i set affinity for TF2/HL2 only to 4 core or less its smooth as a baby butt. :D
Imho thats either a problem of the OS (using vista ultimate 64) or TF2 itself, that has problems. I dont notices such behaviour in crysis or L4D, but that dont surprises me much, cause L4D uses a new HL2 engien which supports multithreading much better then the "older" TF2 engien.
Btw. i wouldn't be surprised if you would get the same performance, when you set the affinity for games to 4 cores (even with HT enabled) and you would get the same score as with HT disabled. :p:
Shaun White Snowboarding will not run with hyperthreading enabled.
This is the only game I've found so far. Granted Shaun White is a poor console port, but it is a new game.
IMO, crysis bench they are within the margin of change, and the way far cry 2 is so damn wonky with x2s I would need to see more. Maybe some minimum numbers included as well which the HT ON in crysis is higher ;)
I can't notice anything weird in TF2, L4D or GTA IV.
L4D loads all the threads nicely when hosting a local server (given that the dedicated servers are useless this is actually a good solution). Who said we had no use for quads.. :p:
is it possible that windows is allocating the cores wrong?
IE its optimized for 2 cores, however it can't tell the difference between a HT core, and a physical core? so it uses a physical and a logical core instead of 2 physical cores?
It should be true also, coz X58 hardware released recently and developpers are still working on softs optimization on itQuote:
is it possible that windows is allocating the cores wrong?
we have to wait a bit, think...
There is no such thing as "wrong" core (in Nehalem's case). All the four cores in there are symmetrical in capabilities, a.k.a. SMP.
What the OS kernel can do, for instance, is to spread active threads in a way so the physical resources being utilized at maximum, e.g. if there are four active threads, the OS will look at the CPU# index table (eight logical cores, w/ HT on) and assign each one thread to an index# in a way, that each physical core runs a single thread at a time:
# [0-1] [2-3] [4-5] [6-7] -- optimal performance in a case of four active threads, where each thread is assigned to a CPU core and HT is generally not involved.
# [0-1] [2-3] [4-5] [6-7] -- not optimal performance, where all the active threads now rely on two physical cores, virtualized through HT into four logical ones.
Of course, the more active threads are in the pool, the less is possible to optimize the situation.
^ this
check this clock to clock comparison (@3,7ghz)
between
Deneb, Agena, Bloomfield, Kenstfield, Yorkfield
in the games benchmark section it does report weird fps drops when HT is on
the thread is in Greek
but besides the "google translate to English" icon on top of the page,
the diagrams speak for themselves
I just finished my bevy of gaming tests. Numbers available upon request, but to sum it up:
All things being equal, there is a less than 1% performance difference on average over a field of about 14 tests. The 1% performance sway is, however, consistent and in favor of HT OFF.
Thanks for testing!Quote:
Originally Posted by Truckchase!
oh wow... this is good info.. I am going to disable HT and clock my 920 up higher, just because of the temp benefit of having HT off. :)
I wonder, if this is still the issue with W7's who know how to adress / load cores compared to Vista , which Did it the wrong way , meaning Vista did load core 1 2 3 4 , which were 2 virtual and 2 real compared to W7 that knows 2 4 6 & 8 are Virtual.
Any heads up on this ?