Discussion Forums > Technology
Future Computer Parts / General Computer Discussions
kitamesume:
i'm gonna bump this question and add a few more notes.
--- Quote from: kitamesume on May 01, 2014, 04:31:22 AM ---if anyone could answer this question of mine though i'd greatly appreciate it.
--- Quote from: kitamesume on April 30, 2014, 11:09:13 AM ---on a side note hearing some further news with the haswell refresh, seems like they'll be improving overclockability to an extent.
http://www.guru3d.com/news_story/intel_devils_canyon_and_pentium_k_launch_june_2nd_2014.html
this makes me wonder though, how would a dual-core haswell pentium @ 5Ghz compare to a hawell i3 @ 3.5Ghz.
even if you factor in HT its still barely slower in multi-threaded workloads, and thats without considering the part that HT doesn't scale well enough like real cores.
but coming from my i3-2100 (sandybridge @ 3.1Ghz) i wonder if its a downgrade, its worth considering for the massive single-threaded performance boost.
[100% = base sandy @ 3.1Ghz]
100% x 1.1(ivy) x 1.1 (haswell) = 121% x ( 5.0Ghz / 3.1Ghz ) = 195% or 95% faster in single-threaded performance.
if taking out HT means decreasing the overall multi-threaded performance by 33%(1/3) its still gonna be 30% faster.
but for a worst case, its getting little to no performance improvements in multi-thread workloads.
--- End quote ---
edit: adding one more question, if a dual-core has twice the single-threaded performance of a quad-core, wouldn't it perform identical in multi-threaded as well?
e.g. lets say a "dual-core haswell @ 5Ghz [vs] quad-core haswell @ 2.5Ghz"
its actually similar to "intel i5 4core [vs] AMD FX 8core" in a sense of "fast 2core [vs] slow 4core"
--- End quote ---
theoretically a 5Ghz dual core should perform equal to or faster than 2.5Ghz quad core on multi-threaded workloads, simply because Ghz scales better than more cores.
the threads on the faster core can be executed twice as fast than those on the slower quad core, even if you parallelize the apps the threads still takes more time to finish than the faster cores.
e.g. 5Ghz dual core finishes two threads in 10seconds, while 2.5Ghz quad core finishes four threads in 20seconds, the time it takes to finish is still long even though it did twice more work than the dual core.
now if you factor in multi-threading overhead, which is the main deficiency of core scaling, a quad core might even do worse because threads doesn't finish evenly.
but this is not always the case, if you think of the cores as pipes, threads would form a queue line which means apps takes turns to get processed.
if you put it in direct relation, a single pipe would have twice as long line than having two pipes, which means the stall-rate is higher for the single pipe.
even if the single pipe can execute the queue twice faster than having two slower pipes, if one app stalls or the system interrupts it affects the whole line.
now, imagine each thread taking up 1ms on the line's execution time, if the pipe had 100threads in line(including system), it'd take 100ms for the last thread to get executed.
if you think about it from a different perspective, that 100ms would've felt like a "hang" or "stutter", this is without factoring in stalls and interrupts.
but if you had twice as many pipes, you can split the line into two groups of 50ms instead, which minimizes the perceived hangs or stutter.
this is the reason why multi-tasking benefits a lot more from having more cores than having faster cores.
notes: stalls happens when CPU needs to fetch data from ram,ssd,hdd or anywhere else before starting thread execution.
another cause would be when cache trashes it's current fetched data and collects fresh new data from ram.
while system interrupts... i'm not sure what they're exactly for but i remember system I/O and user interaction is one of the main cause of system interrupts.
now for the main question though, since an i3 isn't exactly a quad-core, i'd expect that a dual-core running at a much faster rate should outperform it in all cases, unless hyper threading does have that much of an impact in performance.
Tatsujin:
Yeah but I still gotta fork out about 130 to 200 USD. I mean its going to sit there as a NAS and nothing else. If you guys are saying its not going to work at all, even during playback and transferring files in and out then how much money would I be saving monthly on electricity versus a haswell DC Pentium or i3 haswell hyperthreaded.
kitamesume:
objectively i estimate a 10drive+Q9550 to consume about 270watts during operation, while a 10drive+i3haswell around 200watts, petium at 170watts.
the drives themselves would take about 8watts each during idle spin, 10~15watts during write but i doubt all 10drives will do write at the same time.
that makes the drives consume about 80~120watts by themselves. the Q9550 at 150watts, i3 at 80watts and pentium at 50watts.
PS: why not build the NAS first on the Q9550? you could easily swap the board and cpu later on if you find it lacking.
edit: speaking of file accesses, you could gang two NICs onto the router.
although the overall throughput will still be limited by the one accessing the NAS.
the overall bandwidth will make it so that two people can access it at the same time without causing a bottleneck congestion.
Tatsujin:
Lol, what? Are you saying that one person can access NAS at a time? So you can't have three devices access NAS at the same time? Please clarify and keep it in a way of where it makes sense to dummy people like myself!
kitamesume:
no you misunderstood, any number of person can access the NAS at a time, the problem is how would you split 1gbps amongst those users?
doubling the 1gbps would at least reduce the perceived congestion by half.
i mean, imagine if someone was streaming movies, and another writing large libraries of series, then a 3rd user copying some movies off the NAS.
Navigation
[0] Message Index
[#] Next page
[*] Previous page
Go to full version