Is single-channel and 2T for A64 really that bad?

[duragezic]

So I got my Opty 146 which I had to pay a killing for, but its stable at 2.6ghz right now, and I'm pretty sure 2.8ghz or higher will be do-able once I get the stock heatpipe hsf. To go with this I got a Epox 9NPA3 Ultra and a 7800 GT.

I'm using 2x512mb of some un-marked RAM (from a cyberpowerpc system), 1x512mb of some OCZ Value ram, and 1x256mb of some Buffalo. All PC3200. I have to run my memory divier at PC133 which currently gives me an actual memory freq of 178mhz. Anything more than 1 or 2 over 200mhz mem freq makes the system not POST.

Now..

1) For 1T command rate, I would need 2 sticks of RAM correct? So I'd either have to drop down to 1GB, or pay out the ass for 2x1GB and have these current 4 sticks be garbage pretty much.

2) For dual channel, I'd need 2 matching sticks, or 2 pairs of matching size sticks, correct? But having 4x512mb (easily attainable, I could sell my 1x256 for $15, then pick up 1x512 of some value ram for $40) would give me dual channel, but 2T, correct?

I'm not interested in timings, but I think dual-channel and 1T/2T are the biggest factors in performance. Can I stick with this odd config of 1792mb ram in 64-bit, slow timings, 2T or does this greatly hurt my performance in games? It'd really cost a bundle just to have dual-channel 1T, and I don't wanna do it if its only gonna give me a couple of extra FPS in games.

 

[Leper Messiah]

Single channel on A64 is maybe 10% tops. And thats only in a very few cases. Normally its maybe 5%. 2T is less than that. Zebo and Duvie and other members of this forum debunked the whole 1T/2T thing back in May 05.

 

[Jotho]

Well, my s754 mobo and Kingston ValueRAM force me into the single-channel and 2T scenario. While my machine is not necessarily using the most high-end parts today, I've been quite happy with how it has performed in the games that I play.

 

[Blooz]

Here's some info on dividers and the 1T/2T thing done at DFI-Street;

http://www.dfi-street.com/forum/showthread.php?t=40178

 

[cubeless]

i put 2 more 512 sticks in my kid's ultra-d... had to go to 2t... the increase in memory overcame any loss of performance from 2t...

one interesting side effect (not absouloutely sure it's related, haven't pulled the memory out to see...) is that the chipset temp seems to have gone up... but i also went from 240 to 280 htt...

 

[duragezic]

Oh ok well I guess I won't worry about it for the time being. Maybe in a month or two I'll get a 512mb stick so I can at least be running dual-channel, but my timings will be quite slow cause the slowest of the RAM in there is only rated at CAS 3-4-4-8, and I'll be in 2T, but ah well I guess. Maybe mess around a little with the voltage too.

And I was under the impression that Zebo claimed 2T greatly hurt A64 performance. THat's what the overclocking thread says, but this isn't true?

 

[Furen]

2T hurts performance much more than any other timing and decreases Sandra/Everest bandwidth tremendously. The performance hit in real-world apps, however, is probably less than 3%. Single channel is another 5% or so and the other timings are probably something like 1-2% each (overall, some apps will take huge performance hits from any of these changes but others will take no hits). None of these changes is insanely horrible but once you start stacking them it can hurt your overall performance quite a bit. Another thing is that sometimes the memory controller on the CPU can't take high clockspeeds while driving lots of banks (4 double-banked Dimms being the worst possible).

 

[Makaveli]

I agree, the performance hit shouldn't be to bad, but once u add all those up your looking at over 10% loss in performance.

 

[duragezic]

Very true, after all them and 10% or more, it is getting a little significant. Still, I think I'll look at my budget in a couple of months, and if I can only spend a little if at all, I'll add another 512mb for 4x512mb in dual channel, or if I have $$ at that time I'll get 2x1GB for the whole works (DC, 1T, tighter timings).

 

[5t3v0]

When I did some 3DMark05 benchmarks on command rate & timings with my previous GeIL memory (2x512 in dual channel), I found that the causes of cpu/memory performance decreased in the order of frequency > command rate > timings. What that means is lowering memory frequency by 39MHz (i.e. going from 166 to 133 divider) had a bigger effect on performance than dropping from 1T to 2T, which in turn had a bigger effect than loosening timings from 2-3-3-6 to 3-4-4-8. Game scores were virtually unaffected with only 27 marks between the fastest & slowest. Of course once you start combining these sacrifices, the scores drop still further e.g. when I tested loose timings with 2T I got lower scores in both game & cpu tests. I didnt test the effect of single channel but I remember the effect on my old nforce2 system was very noticable.

Here's the scores in case your interested. Apologies for duplicating info from my own thread.

MHz..Timings..CPC..3DMarks..CPUMarks
229...2-3-3-6...1T.........6478.........5297
180...2-3-3-6...1T.........6453.........4869
229...2-3-3-6...2T.........6451.........4939
229...3-4-4-8...1T.........6453.........4975
229...3-4-4-8...2T.........6437.........4849

 

[Makaveli]

CAn u do some real testing, 3dcrap scores mean nothing.

 

[5t3v0]

Sorry no can do. Took them out to make way for my 2Gb G.Skills & have them on auction ending tomorrow. 3Dcrap scores hardly mean nothing. They are a better indicator of real-world (games) performance than sandra or everest. The reason I ran these tests was to try different scenarios to help me decide on an upgrade to 2Gb. At first I was considering adding 2 more sticks of 512 & taking the hit with 2T, then I was considering getting 2x1Gb of value ram with slow timings, in the end I opted for 2x1Gb of PC4000 with slow timings but underclocked the frequency & tightened the timings. I think what my tests show nicely is that games performance is hardly affected by these "compromises" which is probably what our budget concious friend here wants to hear.