Brief P45 Bandwidth Testing

[CryptiK]

I thought I'd investigate the bandwidth and latency achievable using the P45 chipset. The P45 is getting a reputation for low clock for clock efficiency compared to its predecessor, the P35, and the performance designed X38/X48 chipset. I tested a few combinations of FSB, ram speed and tRD to see what the P45 could do, and did a brief comparison with an X38 chipset (Thanks to Grnfinger for running the X38 tests ).

Board: Maximus II Formula, Bios 1307 (604 in first SS)
CPU: e8400
Ram: TeamXtreem 8500 (GMH) / CellShock 8000 CL4 (GKX)

445 FSB, 4GHz, 556.5 MHz CL 5, tRD 8 (my 24/7 settings)


450 FSB, 3.6GHz, 600 MHz CL 5, tRD 8


500 FSB, 4GHz, 600 MHz CL 5, tRD 8


505 FSB, 4.04GHz, 606 MHz CL 5, tRD 8


530 FSB, 3.97GHz, 635.7 MHz CL 5, tRD 8


550 FSB, 3.85GHz, 660 MHz CL 5, tRD 8 (best bandwidth and latency I have achieved so far, not bad for a P45 )


434 FSB, 3.9GHz, 651 MHz, CL 5, tRD 6


439 FSB, 3.95GHz, 658.6 MHz, CL 5, tRD 7


539 FSB, 3.9GHz, 539 MHz, CL 4, tRD 10


570 FSB, 3.9GHz, 570 MHz, CL 5, tRD 10


Comparison to Grnfinger's X38 chipset (not using same memory modules and some slight difference in secondary timings but same Primary timings & FSB). I used lowest bootable tRD for each test.



The P45 can pull some decent numbers, I was pretty happy to get 11K across the board in everest.

In comparison to the X38, the P45 has a lower read bandwidth, but higher write and higher copy bandwidth. The X38 also has a lower latency at any given FSB/ram speed combination. Would these differences be evident in day to day usage, like gaming etc? I doubt it, and from my testing and personal experience I think the P45 is a decent mid range chipset, and if there is no need to step up to the X48 chipset (ie: no requirement for two 16x PCI-E slots when both slots are populated), the P45 looks to be a good balance between performance and price.

[TheKarmakazi]

nice write up! Thanks for taking the time to do a comparison. P45 is also balanced by the higher FSB clocking capabilities, where as X38 (in my experience) hasnt done as well.

Also maybe a future update to this would be to apply some strap tweaks via pmem and check out the differences.

[Jimbo Mahoney]

Very helpful and informative, thanks!

[RPGWiZaRD]

Yep thanks for sharing. That P45 would be slow clock for clock is exaggerated.

At 500x8 600 config it's even marginally faster than my Gigabyte P35 DS4 board. But then again I have pretty good latency if running 1:1 at 500 fsb and cl4 timings, ~49ns and ~9950MB/s read. Don't know why using the 1.2x ram multi the latency shoots up to 51ns tho but read B/W improves to around 10200MB/s.

[Shiranui Gen-An]

How much influence does the CPU clock and cache size have on memory bandwidth with the same FSB and RAM speed? I set my E6300 to 400FSB on my new Maximus II Formula and Corsair XMS2 (non-Dominator) 8500 to 600MHz C5, Performance Level 8, and my scores are quite a bit lower.

[CryptiK]

I just replicated your cpu clock, PL, FSB and ram speed and primary timings. Here:



Perhaps the cache size makes a significant difference. I don't have any 65nm CPU's to test it.

EDIT - also just noticed you are using a different version of Everest to me, perhaps that makes a difference too.

[Ket]

Write and Copy bandwidth is far more important than Read bandwidth. Also, Cryptik, shame on you! I can break 9k Read @ a lowly 3.4GHz, you should be able to do much better

[CryptiK]

Hey

It's not very dependent on CPU speed at all, much more so on FSB. However the P45 doesn't scale so well with FSB increases. 11K across the board is not too bad, but it took 550 FSB and 660 MHz ram to hit that.

Just wait till you get a wolfie, then I can laugh at you too

[Don_Dan]

CPU frequency has a great influence on write and copy bandwidth, while read bandwidth and latency are more dependent on FSB and RAM frequency.
Just look at my results that I posted some days ago in the Micron D9 club thread, I'm at a lower FSB but same RAM frequency and CL.
Read bandwidth and latency is quite good taking into account all sub-timings were on auto, but I have a very low read and copy bandwidth compared to your scores. zsamz_ said they are very low, but after looking for comparable scores from nVIDIA chipsets I think it's mainly because of my low CPU frequency, and I think the 680i chipset is not that good at write and copy bandwidth ( chipsatz latency maybe ).

[Don_Dan]

It's not very dependent on CPU speed at all, much more so on FSB. However the P45 doesn't scale so well with FSB increases. 11K across the board is not too bad, but it took 550 FSB and 660 MHz ram to hit that.

You posted while I was writing my post, well, I think it is dependent on CPU frequency, maybe you could also replicate my settings as close as possible on your P45 and run Everest? And please update to a new version.

[CryptiK]

I think it is really chipset dependent too, you're on 680i, I'm on P45.

Compare the screen I posted at 450 FSB, 600 MHz ram, PL8 @ 3.15 GHz to this one at 450 FSB, 600 MHz ram, PL8 @ 3.6 GHz. The only real difference is the read and latency. Comparing different chipsets at a given FSB, CPU clock and ram freq is comparing apples to oranges. Each chipset has characteristic behaviour.

[Ket]

CPU frequency has a great influence on write and copy bandwidth, while read bandwidth and latency are more dependent on FSB and RAM frequency.
Just look at my results that I posted some days ago in the Micron D9 club thread, I'm at a lower FSB but same RAM frequency and CL.
Read bandwidth and latency is quite good taking into account all sub-timings were on auto, but I have a very low read and copy bandwidth compared to your scores. zsamz_ said they are very low, but after looking for comparable scores from nVIDIA chipsets I think it's mainly because of my low CPU frequency, and I think the 680i chipset is not that good at write and copy bandwidth ( chipsatz latency maybe ).

Yu got that a bit back-to-front Write and Copy bandwidth are influenced by FSB speed while Read bandwidth is heavily influenced on CPU speed. To see for yourself set something like a 6x multi, then increase FSB until you hit around stock CPU speed, then do a everest test Of course you can't rule out how different chipsets behave, but you should get an idea.

[WaterFlex]

Awsome Everest results! Good job!

[Shiranui Gen-An]

I can't come close to those scores even with my Rampage'd Maximus Formula + Q6600 + OCZ Reaper 8500 kit:

[Shiranui Gen-An]

500 FSB, 4GHz, 600 MHz CL 5, tRD 8

I got close with my new E8400 on the DFI P35 board, but my read score is still a bit lower. Did you tweak the subtimings?

[CryptiK]

No, all the subtimings were left on auto for those runs. You would need to run tRD 7 on the P35 to be equivalent to the P45, so if you replicated my settings you would score a little lower in read BW. Clock for clock, the P35 has a higher read BW and lower latency, but a lower write & copy BW when compared to the P45. Probably just need to do some more tweaking.

[Shiranui Gen-An]

No, all the subtimings were left on auto for those runs. You would need to run tRD 7 on the P35 to be equivalent to the P45, so if you replicated my settings you would score a little lower in read BW. Clock for clock, the P35 has a higher read BW and lower latency, but a lower write & copy BW when compared to the P45. Probably just need to do some more tweaking.

This is what I get with my Maximus II Formula and the same E8400 with my CellShock 8000 C5 kit. The read score is lower than the P35 and your P45, but I guess I'll just have to chalk it up to individual differences in the components.

[WaterFlex]

It seems, you have PL=10! Set it to 8 and you will get 10250-10300 Read speed.