Kshat, on 21 January 2017 - 08:52 AM, said:
That benchmark is limited to one thread, which devalues it by ludicrous amounts. There is no real world scenario of significance which is limited to one thread.
You can almost any CPU overclock by a much wider margin when you limit the stresstest to a single thread.
On top, the values given in these charts representing other CPUs are more of a worst case-scenario. Usually even adding "decent" RAM- not even OC'd one - will boost these values significantly.
The impact of a CPU on gaming is a more complicated matter than it might seem. First of all, your CPU doesn't have to be the fastest, it just needs to be fast enough to not limit your GPU in the games you play.
Second, there is some odd behaviour when it comes to realworld-performance. On most games/modern engines, you can reach high average FPS with a heavily OC'd Quadcore. Hell, even a G4560 or heavily OC'd i3 (with HT) will do perfectly fine.
But when you look at framelatency and framedrops, usually you get a much smoother experience when using more threads, so an i7 or i7(2011)[Ryzen] will deliver less framedrops. The reasoning behind this behaviour got something to do with load distribution and pipeline length and is closely tied to the reasoning why Skylake architecture delivers smoother framepacing than an overclocked Devils Canyon - maybe with the sole exception of certain broadwell-specimen.
Tl;Dr: benchmarks are only a fragment of the truth and not representative if they aren't specifically tailored to the real world scenario you're running.
1st point I should make is that I originated the 6,8 thread amd cnfg files that floated around for a while, with the help of several forums members, Goose, WiredX and myself as well as others i cannot recall we spend some good time trying to determine the best individual tast threads to designate to which threads, Wired was the first to strik gold with his 6core I7 5820? correct me if im wrong he got 12 threads working about 40-50% each, that was enough to eliminate the CPU cnfg -systaskthread_0 taking 80% of all information MWO was trying to communicate. we spread that areound and voila no more dips below 60fps, a few patches after MWO had implemented some form of this is the code, it worked better much better lately for more threads.
2nd off sometimes you don't get exact returns on multi thread instances, its not always scaling on a 1:1 ratio, some cpus will put out 1.xx in return for adding 1 core. others like AMD FLOPx vishera you get .8 per core or thread you add to the scenerio. it chokes for instructions and pre fetch traffic.
Im not sure how to respond to this, If you go back a year or two in the hardware forums here, you will find all about these things your are trying to enlighten me upon....we have years of data logging and frametime readouts here in MWO forums, I was on the hunt for 60+ FPS at any given time no matter what with MWO and my 8350 which i ended up running a max core speed of 5.2ghz which I could not cool, settled for 5017mhz and have posted all this information way back. My newest rig, which is I5 3550 @ 4172mhz walks all over my 8350 @ 5017mhz same ram same GPU but the intel memory controller flows about 30% more information.
20,000 mb/s roughly with 8350 2133ddr3 gskill sniper @ 2020mhz 10-10-10-30 2T...........max stable ramspeed with tight timings.
27,000mb/s roughly with I5 3550 same ram @ 9-10-9-27 1T.
A perfect example, firestrike combined score............
8350 @ 5.0 score 11500 physics
I5 3550 @4172 score 7900 physics.
GPU 1528/8001 both tests.
the combined scores reveal all,
I5 3550 (7900physics +13647gpu) score = 10710 THE KEY IS combined score: 5146
8350 @ 5.0 (11,500physics +13647gpu) score = 9505 combined score : 4797
Edited by Smokeyjedi, 21 January 2017 - 01:13 PM.
2
