Originally Posted by mdocod
Originally Posted by Mega Man
yea no in so many ways. lets take this apart,
added numbers for ease (bold,underlined and Italics )
1, mostly true, apm means the cpu will throttle @ core temp of 40c and socket temp of 70c HPC takes away the throttle @ core temp, however leaves in place 72c on socket temp throttling
I've thoroughly tested APM with a wide variety of speeds and voltages. There is certainly a TDP "calculation" involved similar to the way that many GPUs impose absolute TDP limits on stock BIOS even when cooled more aggressively. There is ALSO temperature based throttling going on with APM. There is ALSO some form of core-under-load-count based throttling going on. This can result in conditions where APM throttles where neither TDP nore temp limits are in play. At the end of the day, the "final" word is the TDP calculation. Doesn't matter if you're holding 30C under load APM will hold to a calculated TDP based on voltage and clocks.
2 some ram may not be able to oc, and there is an EXTREMELY low chance that 1600 ram @9-9-9-24-cl1 can match 2400 10-12-12-31-cl10 in benchmarks although most uses will not notice a difference in day to day some may
1600-7-8-8-20-1T was only beat by 2133-10-12-12-30-2T in my testing by 1-4%. That's a very different comparison than the one you are alleging I have made here.
3 this is awesome because tbh i suck at it however i would love to know where you get this grey wall from
Different benchmarks generate different results. In the software I use to benchmark memory performance, I'm seeing a "wall" around 21GB/s "copy" speeds that I'm close to regardless of memory speeds as long as they are above 1600MT/s with the timings nailed down as tight as they will go. I've seen the same poor bandwidth scaling above 1600MT/s show up in a number of popular "review site" articles when they actually take the time to perform a test with the timings tightened up at 1600 speeds. Perhaps coincidentally, this is the rated bandwidth of the IMC by AMD, and when we actually look at the memory support page, most FX-AM3+ configurations actually only officially "support" 1600MT/s speeds. It's not particularly surprising to me to see this "grey area" wall show up here, with relatively poor scaling above it.
My testing indicates, that there is indeed a decent amount of memory performance to be found beyond "default/generic" speeds/timings (like 1600-9-9-9), however, I have not personally been able to generate a test result that pointed to a clear advantage to going for higher clocks instead
of tighter timings. I run linux so the benchmarks I can run are likely going to lead to different conclusions. AIDA appears to be the sort of benchmark that is very synthetic, hardware level, almost "theoretical" as it just hammers the IMC with all CPU cores generating calls as fast as they can. In "real world" workloads calls to the IMC aren't apt to ever have that sort of "attention" level.
4 while this maybe true being the fact that i have quadfire and i have done 1x 2x 3x and 4x gpus this is simply not true,
Like I said, I have seen a lot of DIFFERENT results and claims on the issue. I think it is worth exploring in each individual case because different workloads with different hardware configurations will respond differently.
while you may think this works this way, ht has little to no impact on physics, as a matter of fact the only thing i remember ht increasing is graphics score, i may add that i see very little improvement over 3000 ( considering i run 3900 on my ht i feel i have a leg to stand on, unless you actually run anything more then that, or as close ? and can prove it is stable )
Approaching 4000MHZ HT link speed would be perfect to "almost entirely alleviate" the HT link bottleneck when all 32 PCIE lanes are occupied by "big" GPU/s. The leg you are standing on supports
the philosophy that faster HT speeds may be beneficial. I recommended
attempting 3200MHZ and seeing if it helps. What is your
recommendation? You're sort of hinting at "don't bother mdocod is wrong" on one hand while on the other hand gloating about near-8GT/s HT link speeds? Do as you say, imply, or do here?
In the same way that my memory benchmark produces different results than AIDA, we have to concede that some game engines may have a lot more traffic between the CPU, system memory, and GPUs than a particular benchmark. I'm not sure what context you are referring to here with "physics" as that is just one piece of the puzzle for communications on the HT link while gaming.
also to note have you ever tryed to get 3000+ht on most chips, while the newer chips may have an easier time i can say it is relatively rare to be able to get to 3000 let alone above it, and very risky to do so as you will have a corrupted os if you do not verify that it is stable.
FX chip comes off the same assembly line as an Opteron, it's just packaged differently. The PD Opterons support 6.4GT/s HT link speeds on 4 HT link controllers. The AM3+ packaging effectively disables 3 of those links. Most desktop (FX) PD chips should have no trouble running at these speeds. Any holdup to stability is more likely to be in the HT controller in the chipset-NB, not the CPU-NB.