^anandtech's VX4000 did 265mhz, but im not sure how much tweaking effort they put into that.
Printable View
^anandtech's VX4000 did 265mhz, but im not sure how much tweaking effort they put into that.
:D O.K. Sascha, 2nd time especially for You:Quote:
Originally Posted by saaya
http://img13.exs.cx/img13/1893/90008ry.jpg
No G.SKILL - A-DATA DDR-600 !!!
:toast:
damn thats some hot stuff, its tccd right? what week?
:D A-DATA is TCCD - don't know anything 'bout the chips, still HS on it.... :(Quote:
Originally Posted by Revv23
:toast:
Charlie great idea for a thread. Threads like these are why this forum is the best. I posted this on the front page.
before you guys get too excited about TCCD doing over DDR600 7-3-3-2.5-1T, let me cool you down....
although doing sandra and SuperPi 1M runs a should not be a problem I had absolutely no luck in running Prime95 for more that 6hrs at that speed. for those that think that Prime95 is BS - D2OL left overnight @ 300x8 7-3-3-2.5-1T resulted in BSOD next morning ;)
AND yes, only 300 7-3-3-2.5-1T can be at least SOME competition for 270 6-2-2-2.0-1T
i can get 29ish stable fot couple of hours but most i can with reasonable timmings is 290 with 2x256MB LA...
I wish I had my benchmarks here to post :(
I have both Geil 3200 Ultra X TCCD and Kingston 3500 BH5.
I tested all my ram sticks on my Gigabyte K8NS Pro and A64 3700 about 2 months ago, and basically I came to this conclusion.
my Geil Ultra X is great for 260 cas 2-2-2-8 but after that, it takes a SHARP drop in cas lateny to overclock with, and the latency kills the performance until I pass about 320. then its great again.
it does all this at just about 3 volts even.
my BH5 however can do up to 283 cas 2-2-2-5 any day I want it too, with an A64 and board that can handle it. but it takes me about 3.8 volts to do that.
BH5 I keep for my Benchmarking, cause my Geil just can't beat it, without ridiculous cas latency, but my Geil Ultra X is great for day to day games and such, cause it does 260 at cas 2's at just alittle under 3 volts and does it easy and stable.
for day to day use, my TCCD is best.
for benchmarking, my BH5 is best.
so my BH5 sits in my Seti Machine, burning away at 3.1 volts everyday, burning in, and earning me alittle seti points at the same time every few months (it's not actually connected to the net but once a month)
I am however interested in seeing other peoples results.
my friend shawn just got some Geil Ultra X 4400 and that stuff did 260 cas 2.5-3-3-8 for him.
that was really cool to see, cause he definatly isn't a hard core overclocker, just a casual overclocker, and to see 260 on a AMD XP from him is pretty cool and showed that TCCD even on old hardware is worth checking out.
charlie, I know what your looking for here and for sure there is a point where it happens. But as far as I can tell the point is somewhere no TCCD can go :DQuote:
There MUST be a point that TCCD at 2.5-3-3 PASSES BH5/VX at 2-2-2...
bachus_anonym, thanks for the tests so far. interesting stuff.
however - as already mentioned - your BH5 timings are not quite where they could and IMO should be. There are still quite a few mem settings in the A64tweaker you could easily lower without loosing any stability or OC MHz (this is the case w/ my UTT/VX sticks and mushky BH5 atleast). And there is a small performance gain to be achieved.
Here's an example (just a quick spi1m test w/ my VX but you'll hopefully get the point):
http://www.akiba-pc.com/K8NF-9/utt_slower.gif
http://www.akiba-pc.com/K8NF-9/utt_faster.gif
:confused: Don't You use image shack ? Easiest way to have worldwide access to Your pics............ :toast:Quote:
Originally Posted by Kunaak
@macci...
i've never said that those timmings for either BH5 or TCCD are tightest possible. you can still tweak more @ both cases. but that's just not the point...
the point is to show how both memory sets perform at exactly same timmings - not at their absolute MAX. TCCD can also be tweaked up, look at the attachment ;), although TRTW of 1 results in a RESET, at least at this speed...
But why would you want to run different memory sticks w/ the same (un-optimal) timings?Quote:
@macci...
i've never said that those timmings for either BH5 or TCCD are tightest possible. you can still tweak more @ both cases. but that's just not the point...
the point is to show how both memory sets perform at exactly same timmings - not at their absolute MAX. TCCD can also be tweaked up, look at the attachment
Wouldn't it be best to run em all as fast as they go (as far as the settings go) and then do the tests?
:D O.K. guys - show down on high noon, sword or gun ? :p:
:toast:
well, you right... but wouldn't it also be OK if you run them both at SAFE but still fast settings ? the comparison wouldn't be fair if I run them both at e.g. totally slack timmings for BH-5 vs FAST timmings for TCCD...Quote:
Originally Posted by macci
as long as I don't show some un-fair comparison i think I can consider it valuable information...
EDIT: this is just not a contest of what lowest timmings each memory can go... it's just a comparison of what both can do at EXACTLY same settings as this is what we're looking for. both sets can still be tweaked...
Everything we have seen in this thread so far is valuable info. No doubt about that :)Quote:
as long as I don't show some un-fair comparison i think I can consider it valuable information...
But the original question was 'Low latency, HIGH HTT, who wins??' - right?
So lets use some real low latency settings w/ the low latency ram and lets push the highMHz ram as far it can go and find out which one stays on top as far as performance goes :)
talking about SAFE settings.. the pic w/ a bit faster spi1m time shows the settings I'm useing w/ VX for everyday operation. Has been stable and I'd consider em safe.
I dont have experience w/ TCCD tweaking but if it can run those high speeds w/ faster settings then why wouldn't you use em?
I'll be doing another comparison very soon, maybe even tonight... unfortunately, I'm kind of limited to non-3D benchmarks at this moment...
anyway, I'll be most likely doing 300 7-3-3-2.5-1T vs 270 6-2-2-2.0-1T. both will be called "benchable" not "rock-stable" (there's no way in the world i can get them Prime95 stable for even couple hours, especially with 300 7-3-3-2.5-1T...) and done at MAX POSSIBLE tight timmings.
I think, that results wil be at least VERY interesting ;)
µnrealneo², yea I know, showing the gap was the point :)
I'll try w/ 2x256s next
and sold my EBs already, but thats pretty impressive unbuffered you got there!
Again, back to the original question.... HTTvs blah blah blah
HTT is irrelevent. This is the bus that connects the A64 to the IOcontroller chip. The MEMORY controller is inside the A64. The 800Mhz HTT bus stock is capable of... 2 BYTES (16bits) x2 (rising+falling clock transfer) x800,000,000 transfers per second. Thats 3.2GB/s transfer rate or the max theoretical output rate of TEN SataII disks in Raid0. Nothing your system is doing is going to tax even STOCK HTT.
I know its nice "numbersmanship" to pump up whatever is pumpable, but it really cant meaningfully impact anything but stability from pushing the limit of fiberglass PCboard technology with a higher HTT frequency. Our frankly CHEAPO motherboards are mass market materials, nothing like military level PC materials and you just ask for needless trouble bothering.
Inside the A64 the MEMORY controller is connected to the core over an internal 8 BYTE (64bit wide) HTT bus for a theoretical MAXIMUM of 25.6GB/s transfer rate at the stock HTT setting. There is no dram memory in the world that will ever clock 1,1,1,1.0 @ 800Mhz so you cant ever even THINK about taxing the stock 800Mhz HTT transfer pipe.
What MATTERS is how fast the MEMORY bus is.... the bus between the memory controller and the sticks of ram. Here the 250-350Mhz range is where the realm of possibility with current technology places us.
So the Question is.... Is 250Mhz 5,2,2,2.0 best, low-bus-speed low-latency; or is 325Mhz 7,3,3,2.5 best, high-bus-speed high-latency.
Never thought i'd be thinking 250Mhz is LOW bus speed :D
The answer is.... the most MB/s is what is best, how you get there is IRRELEVENT.
If the internal HTT speed is WAY beyond the theoretical maximum the memory could ever deliver @ 1,1,1,1.0 then it will not make any measurable difference whether it is a little over what's needed or ALOT over what's needed. it is still OVER what is needed to completely meet the maximum datarate the memory can deliver.
You guys with A64 systems and good VX and TCCD sticks can answer the question by simply measuring the MB/s the various configurations result in.
If you are asking me, from a stability and reliability standpoint I find 250Mhz @tight is better than 300+Mhz @un-tight if they both result in the same MB/s.
Purely from a reliability and stability point of view. you are likely to encounter less trouble system wide at 250mhz than at 300+mhz. Since these are NOT differential signals like HTT/PCI-E, we are approaching the signal integrity threshold of average fiberglass PCboard materials. All of a sudden passing 300mhz we need to pay close attention to the path's data lines take between controller and DIMMs, etc etc.
Now PEOPLE's decisions about this are clouded by AMD, since they UPPERlocked the multipliers on the chips. So if you cant afford a x11 x12 x13 chip and have to settle for x9 x10 then getting the internal core clock speed to its MAXIMUM requires higher and higher external bus speeds.
So if you are stuck with a 3000+ x9 multi, then 300Mhz TCCD makes more sense, but if you have a 3500+ x11 chip 250Mhz will let you still reach the chips internal limit. If 250Mhz 5,2,2,2.0 yeilds the same result as 300Mhz 7,3,3,2.5 then this really is all the decision boils down to.
uwackme... tho people keep using the term HTT incorrectly it's not what is meant. The point of this thread is to find what is best between Low Latency maxed, and high 1:1 "HTT" at average timings...
whenever you see HTT from now on walk over to the wall, bang your head a few times and then return to the PC and pretend it said "RAM MHz" instead hehehe
aaah yeah.. what do you mean by MB/s? How would you go bout measuring that?
Cos Sisoft Sandra sure as heill dont tell me jack bout any systems performance.... it's great for overclocking competitions and pretty screen shots, but for anything "real" it means diddly squat
Is it typical that Geil ultra X can run around 250-260mhz with 2-2-2-x timings at around 3v?Quote:
Originally Posted by Kunaak
You may be stuck w/ 9x multiplier but your RAM divider options are still open.. ;)Quote:
So if you are stuck with a 3000+ x9 multi, then 300Mhz TCCD makes more sense, but if you have a 3500+ x11 chip 250Mhz will let you still reach the chips internal limit. If 250Mhz 5,2,2,2.0 yeilds the same result as 300Mhz 7,3,3,2.5 then this really is all the decision boils down to.
Quote:
lets say one is useing a 3000+ and it hits lets say 2600MHz, so what options do you have?
9x289 w/ RAM @289MHz w/ divider 9, 2.5-3-3-6
9x289 w/ RAM @260Mhz w/ divider 10, 2-2-2-5
9x289 w/ RAM @236MHz w/ divider 11, 2-2-2-5
Ok, Im on the same wavelength now ;-)
on the mB/s, the Sandra's is a guesstimate. you could use a Oscilliscope and memtest to actually SEE and measure for real what the MB/s transfer rate really is. On paper, you have to take the A64 databook, draw out the ram timing diagrams, and use the various bios settings for the timing values.
You can write the theoretical MAXIMUM the A64 can perform, simply use the min possible timing parameters in the diagram. But the thoery isnt achievable in real life. But using the #'s from A64 tweaker/bios you can actually get damn close to what your real MB/s is off the timing diagram. You want to calculate burst read, and random write, assume 95% read 5% write in real world use. Then add it up.
All those little #'s... cas 2.0, Trfc 16...etc, etc all represent clock cycles so at 250Mhz thats 4ns. cas 2.0 means 8ns MINIMUM between each successive 128bit access. "On page", in a burst the A64 can suck in a bunch of data in repeated CAS accesses, where all the other signals stay static and CAS plus the lowest address lines change per access. Each taking 8ns, 8ns, 8ns..... at 300Mhz cas 2.5, its (3.3+3.3+1.75) 8.35ns per access. So you see you need to actually get nearer 315mhz just to achieve an 8ns cas cycle to equal cas 2.0 @ 250Mhz.
if Im reading this right the difference from cas2 and cas3 is 85mhz ?Quote:
Originally Posted by uwackme
Thanks
Dan
In theory it would seem so but in reality ('real world performance') CAS latency doesn't really make any difference at all (on modern platforms).Quote:
if Im reading this right the difference from cas2 and cas3 is 85mhz ?
--- uwackme ---
FYI, CAS latency has no bearing on interword access timing during bursts. CAS only comes into play for the first access of a burst when the starting address is loaded. After that, data is clocked out on every edge of the clock with the memory chip itself internally incrementing the address (for bursts up to 8 words).
Note also that the CAS latency can be effectively hidden during longer bursts because the address change (and associated CAS and latency) can occur while a burst is in progress.
When you throw bank interleaving and DIMM interleaving into the picture, the rate at which CAS access time causes a "stall" in the every-clock-edge-stream becomes fairly small for localized accesses of the same type (reads or writes).