[QUOTE=ET_;200625]Note that i5 seems more flexible on overclocking than i7.

Luigi[/QUOTE]

Is this due to 3 memory channels vs 2 memory channels or is it some other thing?

Nobody seems to have noticed that i3's have arrived:
[URL="http://www.newegg.com/Product/Product.aspx?Item=N82E16819115222"]i3-530[/URL] at $125 and [URL="http://www.newegg.com/Product/Product.aspx?Item=N82E16819115221"]i3-540[/URL] at $145 ...and [URL="http://www.newegg.com/Product/Product.aspx?Item=N82E16819115220"]i5-6xx[/URL] stuff.

Anandtech: [URL="http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=3704"]The Clarkdale Review[/URL]

[quote=Batalov;201376]Nobody seems to have noticed that i3's have arrived:
[URL="http://www.newegg.com/Product/Product.aspx?Item=N82E16819115222"]i3-530[/URL] at $125 and [URL="http://www.newegg.com/Product/Product.aspx?Item=N82E16819115221"]i3-540[/URL] at $145 ...and [URL="http://www.newegg.com/Product/Product.aspx?Item=N82E16819115220"]i5-6xx[/URL] stuff.

Anandtech: [URL="http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=3704"]The Clarkdale Review[/URL][/quote]
Boring from a GIMPS point of view - dual core only, no integrated memory controller.
Then again, they might be a good fit for a low-cost HTPC.

[quote=Prime95;198564]No time soon. I've rewritten a few of the FFTs between 2M and 4M. Left to do: rewrite all 2 pass FFTs, optimize for 64-bit, optimize for P4, K8, K10. A long, long, long vacation starting in January will slow development too.

I've changed the way the code is organized so that many different FFT implementations can be supported. The current FFT code remains as one of the FFT implementations, and I've been adding several other variations of radix-4 FFTs to study which ones might be faster.[/quote]
Great work so far and I look forward to the new code. I am using 1K7, 1P4Celeron, 1P4, 1P4 HT (4 threads), 1K8, 2 Core-2Qs, 1Phenom II-X4 (13 threads), and soon a K8-x2 (2 threads).

Are the new processors are all running P4 code currently? Once that changes, there should be dramatic speed increases for K10 and Core-2/i357.

You did not mention Core-2 or Core ix. I wonder if you just need to optimize for 32-bit and 64-bit Core-2/Core-ix and K8/K10. That means 4 new routines (for everything!!!). Of course k8 and k10 largely use the same instructions, and there may be only slight instruction latency differences which may require reordering, so I'm not sure whether you need separate instruction sequences if you don't need the new instructions. Ditto for Core-2 vs Core-i7. [Check whether any new Core i7 and especially K10 specific instructions are relevant to prime factoring or FFT, since K10 does not yet include SSE3 instructions (too bad)].

Given that you have already optimized for Pentium 4 and lower processors, and AMD XP and lower processors, I'm assuming you'll keep in all the existing code for older processors, but I wonder if it is necessary to optimize further for P4 (or any older processor) or to back port new FFT routines for these processors. I'm sure there are many times more P4 and K7 machines currently running than K8, K10, Core 2 and Core i7 (you'll know the exact disposition of systems producing the results), but as time progresses, more of the older systems will retire while the newer systems will be used, each with multi-thread capability. For example, I won't mind 5-10% speed increase in the K7 & P4s, but I doubt more is possible. On the other hand, I'm sure at least 10-30% speed increase is possible on each thread of the Core-2 and Phenom II systems.

BTW, is the code (already re-written) production ready, that is does it produce exactly correct results without further optimization? Does it run faster on any machine especially P4, K10, Core-2? And can the binaries be downloaded by those of us who want in on the leading edge? Please?

One more thing - does the code write regularly (eg every iteration) to one or several memory locations as flags/counters due to register pressure for example, it may help to put those counters/flags in separate 128 byte blocks. In one dual-threaded program I optimized for the P4 HyperThreading, all the processing was done using EAX, EBX, ECX, EDX, ESI, EDI, EBP, and due to register pressure, a separate 32bit integer variable was used for a counter which was decremented every iteration in each thread. On the Pentium 4, there was no problem with both variables being in the same 128 byte cache line, but on Core-2 and Phenom-II I got >31 and 41% speed increase, respectively simply by putting them in separate cache lines.

I also got 10-13% speed increase on the Core-2 and Phenom-II simply by eliminating prefetch.

[QUOTE=vsuite;201383]
Are the new processors are all running P4 code currently? Once that changes, there should be dramatic speed increases[/QUOTE]

Right now it is mostly Core 2 optimized. Version 24 was P4 optimized. Version 26 will let me optimize for both.

Dramatic speed increases? Don't get your hopes too high. Architecture specific optimizations are generally in the 5-10% range.

[QUOTE=vsuite;201384]I also got 10-13% speed increase on the Core-2 and Phenom-II simply by eliminating prefetch.[/QUOTE]

Software prefetching is still very important to large FFT performance.

Well I FINALLY finished building my i5 rig after several complications regarding the wrong cooler being shipped, but eh, it was worth the wait.

The i5 is happily crunching on 4 LL's at 3.2GHz and being fed by 2000MHz of Kingston HyperX DDR3 memory.

Benchmarks [URL="http://www.mersenneforum.org/showthread.php?p=202068#post202068"]here[/URL]

Happy crunching :toot:

I just bought an i5 750 @ 2.66 GHz. It comes with its Intel cooler.

From what I read in this tthread, it should be safe to run GIMPS 100% with it, as Petrw said...

Luigi

[QUOTE=ET_;202183]I just bought an i5 750 @ 2.66 GHz. It comes with its Intel cooler.

From what I read in this tthread, it should be safe to run GIMPS 100% with it, as Petrw said...

Luigi[/QUOTE]

I bought the Noctua cooler that Garo recommended so I can't speak for the Intel Cooler personally. I have second hand opinions that it is not great but I'm sure it will work find with NO any maybe with some(?) overclocking.

Has anyone looked into the newest i5's?

So says newegg.ca............

[CODE]Intel Core i5-650 Clarkdale 3.2GHz
LGA 1156 73W Dual-Core Desktop
L3 Cache: 4MB
Manufacturing Tech: 32 nm
64 bit Support: Yes
Hyper-Threading Support: Yes
Integrated Graphics: Yes
Graphics Base Frequency: 733MHz
Virtualization Technology Support: Yes
Your Price:$199.99 [/CODE]


[CODE]Intel Core i5-661 Clarkdale 3.33GHz
LGA 1156 87W Dual-Core Desktop
L3 Cache: 4MB
Manufacturing Tech: 32 nm
64 bit Support: Yes
Hyper-Threading Support: Yes
Integrated Graphics: Yes
Graphics Base Frequency: 900MHz
Original Price: $214.99[/CODE]

[CODE]Intel Core i5-660 Clarkdale 3.33GHz
LGA 1156 73W Dual-Core Desktop
L3 Cache: 4MB
Manufacturing Tech: 32 nm
64 bit Support: Yes
Hyper-Threading Support: Yes
Integrated Graphics: Yes
Graphics Base Frequency: 733MHz
Original Price: $219.99[/CODE]

[CODE]Intel Core i5-670 Clarkdale 3.46GHz
LGA 1156 73W Dual-Core Desktop
L3 Cache: 4MB
Manufacturing Tech: 32 nm
64 bit Support: Yes
Hyper-Threading Support: Yes
Integrated Graphics: Yes
Graphics Base Frequency: 733MHz
Virtualization Technology Support: Yes
Your Price:$318.99 [/CODE]