Furture dominance in descending order?

[andy5174]

There are three futures, Frequency, cache and FSB, which are the most dominant to CPU performance. However, I want to know which one is the most dominant. Could you guys please place them in descending order as to of their dominance regarding the CPU performance? Thanks in advance.

 

[JAYDEEJOHN]

Actually, youre omitting a fourth and fifth, and raising fsb too high as it now pertains to DT.
IPC, or general arch effects, is one which is harder to market on, but is most important , and of course, number of cores.
Since both the former and the latter effect freq, its this combination you should really be concerned with. To me,
IPC
Core count

Itll be how the overall arch is balanced as to how good the freq are. Larger multi cored di wont clock as high, while wider, shorter piped arch's hinders clocks as well.

Since both AMD, and Intel, now with their i7 arch use IMC, the front side bus has been pretty much eliminated as something seen as a top priority, tho speeds can be still argued, the BW is no longer a priority.

Cache plays a vital role in apps, some more than others, especially in gaming, where more has greater effect, but on some apps, it matters little. Itll be interesting with W7, and DX11, and how it turns the needs of games over to cpus, while also, freeing up cpus, and having the gpus do more, how those changes effect gaming and perf overall, and the needs/rankings of cpus.
For each app, these (freq,cache,fsb) changes in order of effectiveness, and to pin down a certain order of relevance depends on said app

 

[andy5174]

Hi, I do know that IPC and no. of cores are more important than those features I listed. However, these two features are far more than easy to understand their effect on the CPU. Sorry for the confusion, what I am actually asking is that the order of dominance of the three features regarding the CPU performance based on the same IPC and no. of cores. By the way, I didn't actually consider the higher IPC of i7s as they are too expensive.

 

[JAYDEEJOHN]

Well, again , theyre very app dependent. Id say
Clocks
FSB
Cache
With clocks way ahead of either FSB and cache, which can trade dependent again, on app

 

[andy5174]

Thank you very much for your answer.

 

[Reynod]

I would rate:

IPC / Core frequency
QP / HT Frequency / FSB
L1 size / L2 cache latency / size / L3 Cache size (latency here being less important than size.

A slower dual core (AMD64) would eat a much faster dual core Intel (pentium 4) due to much higher IPC - well actually the IPC increase AMD had was often due to the bug the Pentiums had with their pipes. Still ... a fair example of how a 2Ghz cpu would beat a 3Ghz cpu.

And to play fair ... a 2Ghz core2 would flog a 3Ghz AMD64.

 

[JAYDEEJOHN]

Just remember, as you asked, future, is all important, as we see which direction Intel and AMD are heading. Getting those monstrous multi cores up to speed will be the biggest issue, with the highest payoff, at least seen immediately, until MT truly takes off, but theres slowdowns there as well, that wont go away ever.
The speeds on the FSB will still be important, especially in a MT scenario, and with shared cache seen more n more, it starts to take a back seat.
Heres a nice link that pertains to much of this
http://www.wepc.com/discussions/view/6612/Two_Important_Trends_in_the_CPU_Industry

 

[andy5174]

Wouldn't the performance of CPU be bottle necked by FSB and cache when the speed reaches certain level? If so, can I say that the FSB and cache is more important when the "certain level of speed" is reached?

e.g. Assume that the max number of processes can be performed on a 800MHz FSB and 2MB L2 Cache CPU is only 30 which is EQUAL to the max no. of processes can be done on single core 4GHz CPU. Will the 3GHz dual core CPU which can do 1.5 times processes of the single one be bottle necked by 50%?

 

[andy5174]

Also, in this article "http://www.tomshardware.com/reviews/multi-core-cpu,2280-11.html", if I didn't misunderstand it, it claims that the performance between single/dual/tri/quad core CPUs will be normalized when resolution goes up. Hence, a single core CPU with a good graphic for normal user and gamer will be good enough, will it?

 

[JAYDEEJOHN]

Since the intro of IMC on both makers (Intel and AMD) BW isnt a problem anymore, at least on DT, which is why buying a i7 for DT because it has tri channel is even more redundant
I mentioned cache speeds and sharing as well. Theyre all somewhat dependent, but if the BW cant really be saturated, like we saw on older Intel products, speed on FSB isnt as needed, tho, again can help.
Sharing cache into the future will be creating pools thatll be fine. Its more how current and future gen cpus actually use the cache, as to amount as well.
Its why scaling is so important, as if the arch isnt truly done correctly, it wont scale as we see clocks increase. Some has to do with the core directly, but also, some of it can be caused by slow cache, or even low cache, like you say.
Thats why we see differing levels of cpus, as some (usually deffective) cpus have lower cache amounts, and dont scale properly, and even show deffeciencies at stock clocks, compared to their full, bigger brothers

 

[JAYDEEJOHN]

"As far as games go, we see a huge 60% performance jump from going single-core to dual-core, and a further 25% leap from dual- to triple-core. Quad cores offer no benefits in the sampling of games we tested. While more games might change the landscape a little, we think the triple-core Phenom II X3s are looking good as a low-cost gaming option. It's also important to note here that as you start shifting to higher resolutions and adding visual detail, the picture gets a lot murkier as graphics muscle becomes the prevalent determinant of frame rates."

Somehwat self explanitive. As your res goes higher, the slowdown, or weakest point of the config becomes the gpu. So, if a quad get only 10fps as well as a single core, its something you cant base it on at all, since, the honus shifts the workload to the gpu, and not whether you have enough cores or not

 

[Reynod]

Yeah but that guy is now part of Intel's propaganda machine - Anand will be working for Intel officially sometime soon ... not just a Shill as he is now.

They already have one Anand ... maybe 2 are better ??

"Intel’s vision for the future of microprocessors. I’m starting to believe it’s right." ... just made me puke.

I must admit he is probably one of the best tech writers around but bends all of his will to the blue team now.

Only those German dills that do hardware reviews (if thats what you call them here) for THG are more insanely pro-Intel.

Check out the language in these classics of Pro-Intel (we are getting our asses handed to us by AMD but we are paying others to say otherwise !!).

http://www.tomshardware.com/reviews/dual-core-stress-test,1049-20.html
http://www.tomshardware.com/reviews/cpu-stress-test,942.html

Guess what ... nothing's changed.

 

[JAYDEEJOHN]

Like in the other thread, why isnt Intel expected to do more now? I know, I see it also. Its condescending, elitest attitudes

 

[JAYDEEJOHN]

The problem I have with that is, wheres the questioning for the "other" side? its as if they cant wont question them. And, at the very least, their inability to think outside the box, the one with ding dong ding inside

 

[JAYDEEJOHN]

After reading some of that old article, the writer failed tagan and tagan to get to the real source of the failures, which we know now by hindsight. yea, it was something elses fault, sure. Not that the old P4s needed tons of power or anything, and were hot, and spread that heat throut the rest of the components in the case.
And of course, they had nothing to say about the AMD setup. Unless if it had to be reset, which when done wasnt elaboated on at all, leaving the reader with what?
Same ol same ol is right

 

[andy5174]

IMC? BW? DT?

 

[JAYDEEJOHN]

Integrated Memory Controller
Band Width
Desk Top

 

[andy5174]

thnx

 

[andy5174]

BTW, is it best for the Cavier Black 1TB to be partitioned into THREE 334GB area as it is constructed with three 334GB platters?

 

[andy5174]

Is this CPU "http://processorfinder.intel.com/details.aspx?sSpec=SL8HZ" in your opinion a good scaling one? I actually find this CPU with 9600GT to have better performance in SOLIDWORKS compared to E8500 with 9500GT PCs in the engineering department of my university.

 

[ainarssems]

BTW, is it best for the Cavier Black 1TB to be partitioned into THREE 334GB area as it is constructed with three 334GB platters?

No. Because partitions will be made from all platters using equal amount from each platter and on one platter. Select size depending on other requirements. Usally one smaller partition on the beginning of the drive where it is fastest( it will be made on outside of of all three platters) for OS and programs and other partition on the rest of space for storage.

Is this CPU "http://processorfinder.intel.com/details.aspx?sSpec=SL8HZ" in your opinion a good scaling one? I actually find this CPU with 9600GT to have better performance in SOLIDWORKS compared to E8500 with 9500GT PCs in the engineering department of my university.

It is no wonder You see better performance with better video card as it is 3D application and more dependant on video card then CPU. But that P4 cpu is not very good by todays standards and You would see even better perfomance with E8500 and 9600GT, and even better with more powerfull video card

 

[andy5174]

Usally one smaller partition on the beginning of the drive where it is fastest( it will be made on outside of of all three platters) for OS

Do you mean that the smaller the better for the system drive?