Phenom (w/L3) versus 775 Quad-core (w/L2)

[VirtualLarry]

Just wondering how much impact the larger L3 cache would have on various DC programs, if one were to use a Phenom CPU, versus an Intel Quad-core with a larger L2, but no L3 cache.

I've heard that there are speed-ups with the Core i7, is that due to the L3, or other architectural aspects of the i7? Just wondering if one would see similar speedups with a Phenom (or Phenom 2, with a bigger L3)?

I'm thinking of purchasing a Phenom II 945 AM3 chip for my HTCP, but when I googled it, it said availability was scheduled for April.

I hope it will be an unlocked "Black Edition", like the 940 is. Both the 940 and 945 are 3.0Ghz stock speeds. 945 is AM3, with a 2Ghz HT bus and 95W TDP, and 940 is AM2+ with a 1.8Ghz HT bus and 125W TDP. I don't think I can use 125W CPUs in my board.

I know that this is overkill for a HTPC, but I plan on running DC on it. Hopefully, with a 300W PSU, that CPU won't be overkill. Planning on using onboard video with a 780G chipset. Foxconn mobo.

 

[Markfw]

I will let you know in 4 days.... Phenom 940 coming, and I have plenty of quads to compare to.

 

[PCTC2]

The i7 speed-ups are both architectural and the IMC. There are some programs that would still benefit from the larger overall quick-response L2, but for DC I am not sure. The difference for having the shared L3 vs a shared L2 for only half of the cores changes power-management schemes (when and what cores can turn off and where the data goes), and for databases the memory hierarchy of the i7 is better.I still haven't answered your questions though.

For DC, I believe the overall "speed" of the processor due to architectural differences at this point in time will determine it's effectiveness. In straight computations like DC, I do not know how the Phenom II fairs against a Core2. Probably similar to daily computing, where the 940 is similar to a Q9400. I do not think it will benefit from having a shared L3 vs a large L2. But Mark, you will have to let us know. I cannot, as I only have a Q6600 and a Core i7 920, and that is like comparing Apples to concrete (not even apples to oranges) because of the architectural differences, the smaller cache of the Q6600 (negating having a large L2 to begin with), and the i7 having an IMC.

 

[dajeepster]

Originally posted by: Markfw900
I will let you know in 4 days.... Phenom 940 coming, and I have plenty of quads to compare to.

same here

 

[Markfw]

I can say my X3350 (Q9450) with the 12 meg cache@3.5 does 2400 ppd on a 2665, and my E8400@4.0 ghz only does 1950. I think cache makes a difference. Both are the smp client. The Q9450 also has 2 GPU units stressing it, and they each take a core, so it should be a fair comparison. Also, my Q6600's also only get 1800 ppd on their smp with 2 GPU clients.

Sp what does an I7 @ 3.5 get on 2 smp clients ? (each)

 

[PCTC2]

Originally posted by: Markfw900
I can say my X3350 (Q9450) with the 12 meg cache@3.5 does 2400 ppd on a 2665, and my E8400@4.0 ghz only does 1950. I think cache makes a difference. Both are the smp client. The Q9450 also has 2 GPU units stressing it, and they each take a core, so it should be a fair comparison. Also, my Q6600's also only get 1800 ppd on their smp with 2 GPU clients.

Sp what does an I7 @ 3.5 get on 2 smp clients ? (each)

Well, with newer drivers, the GPU clients use less of the CPU than before. It would depend on the driver version you are using with the Q9450. And even with older drivers, the GPU doesn't use 100% of a core, so the Q9450 gets a boost over the E8400. But it does look like Cache matters, but the hierarchy also matters. I think the Core2Quads will do better at folding than a Phenom of equal value.

 

[Markfw]

I have tried the 181 drivers. No difference. It seems that Vista is what makes the difference.

 

[Denithor]

How much (if any) effect does HT have on F@H?

/grumble

Waiting for Westmere...still waiting...and waiting...

/twiddle

 

[GLeeM]

Originally posted by: Denithor
How much (if any) effect does HT have on F@H?

I would like to know the same on i7.

For comparison, I can tell how an old P4 Northwood @3.2 single core with HT does on current p385* WUs:

21% better PPD by running two WUs

I know, it slows down the process, but they are sneakernet WUs getting finished at home so they actually get sent in faster this way than if I waited for them to finish on sneakernet computers. I have less idle time this way.

 

[Markfw]

OK, I have had my Phenom 940@3.7 stable for 12 hours now. Dual smp client with the fahsmpaffinitychanger installed like my other boxes. Buy it has a lower PPD than my Q8200@2.8 !!!! 1576 each smp unit. My Q8200 does 1583, I have a Q6600@3420 that only does 1677.

Not impressed.

 

[GLeeM]

Originally posted by: Markfw900
Dual smp client with the fahsmpaffinitychanger installed like my other boxes

By chance does only one SMP running do better?

 

[Neurodog]

Originally posted by: Markfw900

OK, I have had my Phenom 940@3.7 stable for 12 hours now. Dual smp client with the fahsmpaffinitychanger installed like my other boxes. Buy it has a lower PPD than my Q8200@2.8 !!!! 1576 each smp unit. My Q8200 does 1583, I have a Q6600@3420 that only does 1677.

Not impressed.

WoW, that's dissapointing.

 

[Markfw]

Well, I changed the memory to 233x16 and now its up to 1635 on those 2665 units, still short of any C2Q over 3 ghz, but its running 3728 !!!! The only one it can beat is the 2.8 ghz Q8200 (must change sig)