Are we done here?

[John P. Myers]

No one has posted anything here in 2 days! YAY!!! :D Everyone knows everything now. This is awesome!!! :-bd (ok maybe a little sarcastic. I've actually been waiting 2 days for someone to give me something to talk about. Been pretty boring around here.)

[Maxwell]

You know, you are allowed to post in other threads as well... :p

But since you need something to talk about, tell me the difference between the i5 2500k and the i7 2600k. And the k vs. not k. Why should those things make a difference to me? (And yes, I really don't know these answers...)

[John P. Myers]

You know, you are allowed to post in other threads as well... :p

But since you need something to talk about, tell me the difference between the i5 2500k and the i7 2600k. And the k vs. not k. Why should those things make a difference to me? (And yes, I really don't know these answers...)

Yay! Thank you Maxwell! lol

mkay...

i5 2500k:
4 cores (no HT)
6MB cache
3.3 GHz

i7 2600k:
4 cores with HT (8 threads)
8MB cache
3.4 GHz

If it ends with 'k' you can change the multiplier to overclock far more than if it does not.

[Maxwell]

Ok. So, I understand that bigger is better, but what does the cache of a processor do? And I seem recall that you're not a fan of HT - why was that again? (And yes, I'm enjoying "Maxwell gets to pick JPM's brain" hour...)

[John P. Myers]

Well previously having HT really didn't benefit you. Just made it appear you were getting some substantial benefit. It was originally hyped to make consumers think their computer would run twice as fast since now it could "do 2 things at once" which was a huge joke. But with Sandy Bridge, i gotta say HT has come a long way. I'm impressed with the performance gains for the first time ever :)

More cache = faster processing speeds since the chip can read from it's on-die memory rather than accessing RAM. The chip will store stuff that's repetitive (like alot of what we do) in it's cache. It will also make guesses at what it will do next and store that as well. The larger the cache, the more it can cram in there. When the chip guesses right, it just keeps on crunching, because pulling data from the cache is nearly instant. If it guesses wrong, then your chip stops crunching and waits for the RAM to cough up the data, which can take several nanoseconds (which is nothing, but when it occurs millions of times, you notice).

[YoDude9999]

Ok, I don't know about how things worked in the past with HT when Intel first started the whole thing, BUT I can tell you (as I'm sure most have seen by now) that HT does in fact speed up your computer! Well not really, but it does do twice the work in the same amount of time.

In a comparison, I can run 4 CPU WUs on a Q9650 clocked at 3g in a specific amount of time. On an i7 920 clocked at the same frequency, with HT turned on, it will spit out 8 of the same WUs in the same time as the Q9650 does. Though this isn't making the CPU run any faster between the two CPUs at all, it does do twice the work. This is a great and powerful thing for crunchers.

Serious overclockers turn off HT. Exactly why they do this, I have no idea. I once read an explanation over on the EVGA forum about this but for the life of me I don't seem to recall what was said about it. Perhaps you have some enlightenment on this aspect? I don't understand why anyone would would really want to turn off HT.

[DrPop]

I am sure JPM can enlighten us, bit you're comparing a Core architecture chip to an i7 architecture chip, so that's not an indication of what HT can do; rather it just shows how amazingly good the i7 design has been for Intel. The only way to really compare is to time your i7 with HT on, run 8 WUs of a project. Average the time per WU.Then, run 4 WUs on that same i7 with HT off, and average the time per WU.You will see a % improvement running 8 with HT on vs 4 with HT off, but it won't be that dramatic. -- Unless the code you are crunching sucks.That is because HT works by slipping in some code to crunch when it notices the CPU has idle parts. If any part of the ALU side of the chip is idle, the HT throws in a bit of code to crunch, therefore making the CPU 100% utilized, all the time.To illustrate a simple example, if you are running code that is say, 90% efficient, a *perfect* implementation of HT could give you 10% increase in performance over time. Of course, you will not get this due to some overhead and latency, so maybe you would realize an 8 or 9% performance gain.Is that making any sense? JPM's turn because that's about where my knowledge on the process stops. ;)

[YoDude9999]

Ah yes, well I see your point and that does make good sense. I may have to try that out sometime. THX!

[John P. Myers]

I am sure JPM can enlighten us, bit you're comparing a Core architecture chip to an i7 architecture chip, so that's not an indication of what HT can do; rather it just shows how amazingly good the i7 design has been for Intel. The only way to really compare is to time your i7 with HT on, run 8 WUs of a project. Average the time per WU.Then, run 4 WUs on that same i7 with HT off, and average the time per WU.You will see a % improvement running 8 with HT on vs 4 with HT off, but it won't be that dramatic. -- Unless the code you are crunching sucks.That is because HT works by slipping in some code to crunch when it notices the CPU has idle parts. If any part of the ALU side of the chip is idle, the HT throws in a bit of code to crunch, therefore making the CPU 100% utilized, all the time.To illustrate a simple example, if you are running code that is say, 90% efficient, a *perfect* implementation of HT could give you 10% increase in performance over time. Of course, you will not get this due to some overhead and latency, so maybe you would realize an 8 or 9% performance gain.Is that making any sense? JPM's turn because that's about where my knowledge on the process stops. ;)

That is correct. But another very common way processors use HT is when the physical core had a cache miss (as talked about earlier) and is waiting on data from RAM. Then the logical core (the hyperthread) will crunch something while the physical core (the real core) is waiting. After some digging just now, Intel claimed a max improvement of 30% on the P4 with HT. Would have to be your lucky day to actually achieve that. With Sandy Bridge, the best real world benchmark i've seen showed a 33% improvement. Most were around 10-15%. Some were none. It really depends alot on the specific tasks you're running.

Making a chip HT takes about 5% more die space per core than not. All they do is make an additional bank of registers (per core) within the chip. This group of registers is known as the architectural state. (http://en.wikipedia.org/wiki/Architectural_state) That is how it keeps track of what's going on with multiple programs running on the same hyperthreaded core.

In the end, what all this means is, if you run 8 things at once on a system with only 4 real cores + HT, you will not run them all as fast as you would if you only ran 4. Not ever. HT will never double your processing speed. Ever. But it will take less time to run all 8 with HT than to run those same 8, 4 at a time.

And disabling HT allows you to overclock higher with the same voltage, and your chip will run cooler. As DrPop said, HT just makes your core run more efficiently, using idle parts of it. This increases heat by several degrees. Lowering the temperature by turning off HT makes the chip more stable at higher clocks. And the higher clocks make up for the lost benefit of having HT. This is really only done by gamers though. Doing what we do, it's best to leave it on.