3.2 Ghz system least affected by pc2700

[RAQ]

Im upgrading from an athlon +2200, with 2x512 mb pc2700

I would like to get one of the 3.2 Ghz rated cpu's and a new motherboard - but id also like to keep my pc2700 RAM.

I know my RAM will slow the system down a bit compared to the faster newer RAM - but what configuration would suffer the least from using the pc2700 RAM?

Thanks

 

[RussianSensation]

P 4 3.2ghz = 3200mhz = 16x200FSB (4x200 = 800FSB effective)

1. Original CPU:RAM ratio = 1:1
2. PC2700 = 166mhz (333mhz effective)

=> 3 possibilities:

1. 1:1 implies running your PC2700 @ 200mhz (400mhz effective). If you can do that and retain equal timings this is the best option.
2. 5:4 implies 200FSB:160memory. You will run your ram @320mhz slightly underclocked
3. 3:2 impies 200FSB:133memory. You will run your ram @266 or PC2100 speeds => worst possibility

Best Alternative if you are set on a P4 C solution:
Now might I ask why you want a 3.2ghz? When 2.8 can be bought for $167 on Pricewatch vs. $277 for P4 3.2
At the very least 2.8@ 3.2ghz+ => Overlcocking.

Perhaps it would be in your best interest to learn the art of overclocking.

Consider P4 2.8s hit 3.5 on default voltage (Check CPU/Processors Overclocking section on this forum!)

Then, 2.8 @ 250FSB => 3500mhz
@3:2 ratio => 250:166mhz memory or running @PC2700 PERFECT!

Now you suffer virtually no loss in performance if your PC2700 runs at least at 2-6-3-3

Also have you considered simply dropping a Barton XP-M 2400 @$77 into yoru current motherboard and getting it up to 2400mhz? That might be your BEST choice depending on what motherboard you have since it is the most cost effective.

 

[Peter]

You should not go P4, since this CPU type is VERY dependent on having fast RAM. That's because the P4's deeply pipelined design has huge penalties for pipeline stalls.

The best performance option for your two (identical?) sticks of RAM is a socket-940 Athlon64 platform. That'll get you a dual channel RAM controller on a CPU that doesn't mind slower RAM anywhere near as much as a P4.

 

[RussianSensation]

Originally posted by: Peter
You should not go P4, since this CPU type is VERY dependent on having fast RAM. That's because the P4's deeply pipelined design has huge penalties for pipeline stalls.

Really? then why is that 2.4@3.0ghz 250/200 (5:4 ratio) with memory @ 400mhz 2-2-2-5 p4 is faster than 250/250 1:1 with memory at 500mhz @ 3-4-4-8?

Look Duvie did testing 1000x times and tightest timings win over memory speed. P4 DOES NOT take a major performance hit using slower memory. In fact SLOWER MEMORY AT TIGHT TIMINGS ALWAYS BEATS faster memory at loose timings.

I do agree that an Athlon solution is a better choice for a CPU. However getting a p4 2.8 to 3.5 @$167 smokes A64 3000 @ $211 overclocked or not.

 

[Peter]

That's because latency matters a lot. 3-4-4-8 at 250 MHz is slower in many applications than 2-2-2-5 at 200. "Slower", in what I said above, includes all aspects of slowness - both lower clock and higher latency.

 

[Tom]

"You should not go P4, since this CPU type is VERY dependent on having fast RAM. "


That is a vast overstatement. P4 with Springdale/Canterwood gives great performance, even when using memory/fsb ratios to accomodate slower ram.

 

[RussianSensation]

Originally posted by: Peter
That's because latency matters a lot. 3-4-4-8 at 250 MHz is slower in many applications than 2-2-2-5 at 200. "Slower", in what I said above, includes all aspects of slowness - both lower clock and higher latency.

Well now that you do realize that latency timings win over speed, it is pretty clear he will not suffer more than 2-3% performance degradation running at 333mhz using ratios, somewhat undermining the VERY dependency of a P4 on Ram speed that you speak of

 

[Algere]

Originally posted by: Peter
You should not go P4, since this CPU type is VERY dependent on having fast RAM. That's because the P4's deeply pipelined design has huge penalties for pipeline stalls.

The best performance option for your two (identical?) sticks of RAM is a socket-940 Athlon64 platform. That'll get you a dual channel RAM controller on a CPU that doesn't mind slower RAM anywhere near as much as a P4.

Of course it'll have to be registered DDR memory for it to work

 

[Peter]

Originally posted by: RussianSensation

Originally posted by: Peter
That's because latency matters a lot. 3-4-4-8 at 250 MHz is slower in many applications than 2-2-2-5 at 200. "Slower", in what I said above, includes all aspects of slowness - both lower clock and higher latency.

Well now that you do realize that latency timings win over speed, it is pretty clear he will not suffer more than 2-3% performance degradation running at 333mhz using ratios, somewhat undermining the VERY dependency of a P4 on Ram speed that you speak of

With your fact twisting skills, you should be a politician.

Of course there is a quantification to everything, and of course, DDR500 at extremely slow timings is slower than DDR400 at absolutely maxed out timings. Saying generically that "timings win over speed" is laughable enough that I won't bother talk about it.

But you're distracted away from the point anyhow, completely so. No matter what kind of RAM performance delta we're talking about, P4 will be noticeably more affected from it than an Athlon-64. And that's the question here. Got anything to say about THIS?

Algere, not really. Socket-939 is just around the corner, and some socket-940 boards also accept non-registered RAM as long as you keep it to one DIMM per channel.

 

[RussianSensation]

Thats the whole point here.....For him running P4 using ratios with his PC2700 will drop performance LESS than running A64 using PC2700 ram in comparison to 1:1 ratios of stock systems for each. We know A64 runs slower with PC2700 than with PC3200 by a lot more than 1-2%, when a P4 runs lose like 1-2% using a ratio. What you were talking about is the effect of latencies for both systems. A64 is not affected by latencies as much as a p4 system is because it has an onboard memory controller, BUT A64 is affected MORE by pure memory speed than P4 is (remember not running XP with 1:1 is detremental to performance as opposed to not running p4 in 1:1, which still holds true for A64 platform). So in this case slower memory with tight latencies suits a P4 solution better in that sense.

Look Here
A64 FX51 running at 200, 266, 333 and 400mhz memory speeds. The difference in performance is from 86.3 all the way up to 104.6 by just changing the RAM speed. There is no way you'll get that type of performance degradation on a P4 solution when using ratios. Just compare P4 2.8B running at 266mhz vs. P4 2.8C running at 400mhz and 92.8 vs. 87.2. But you must remember FSB performance is what gives the C advantage here 800FSB vs. 533FSB and memory probably only makes liek 1% difference and it is STILL only a difference of 5FPS. Now compare that to FX running at 266mhz which gives 95.5frames and running at 400mhz which gives 104.6 frames a difference of almost 9FPS.

Now the harder question is which system will be faster overall in the end even with the slower ram? Probably A64 and that is where I agree with you.

 

[Tom]

"Algere, not really. Socket-939 is just around the corner,"

How are you basing your recommendations, if the product you recommend isn't even available ?

How do you know how well pc2700 ram is going to work in boards that haven't been released ? How can you compare non-released products to well established products with a straight face ?

I could understand if you counciled waiting because there's a strong likelihood that it could turn out that way, but you talk as though you already KNOW pc2700 will work fantastically better in these unreleased products..

 

[Peter]

This is because we already know all the technology involved - remember that with AMD64 architecture, it's all inside the CPU. There have been no technology changes inside, just a reshuffled pinout to ease the layout toward the DIMM sockets. If anything, that'll improve performance over socket-940. (It does anyway, since not using Registered DIMMs already cuts one cycle off the latency.)

When someone's asking for an upgrade path I rather recommend to wait a bit longer and get a brandnew platform, rather than place their bets right now on a dead end like socket-478 or -A.

 

[Tom]

Originally posted by: Peter
This is because we already know all the technology involved - remember that with AMD64 architecture, it's all inside the CPU. There have been no technology changes inside, just a reshuffled pinout to ease the layout toward the DIMM sockets. If anything, that'll improve performance over socket-940. (It does anyway, since not using Registered DIMMs already cuts one cycle off the latency.)

When someone's asking for an upgrade path I rather recommend to wait a bit longer and get a brandnew platform, rather than place their bets right now on a dead end like socket-478 or -A.

I'm sure you're aware that "the best laid plans of mice and men.." in other words there are many many details that cannot be known beofre the end product is actually in consumer's hands, that is the reason for my skepticism about you're making such an absolute statement about a very specific set of ram that the OP already owns.

 

[Peter]

Can't comment further on this, but there are hands stuff goes through before it reaches the consumer's