Low latency, HIGH HTT, who wins??

[Holst]

From my point of view its pretty much a draw.

"most" BH5 seems to max out at 265 with sensible voltage (around 3.2v)

This is similar in performance as TCCD at around 300 with lesser timings (as bachus shows above)

The TCCD is avalable in larger modules, and does not require as high a voltage (both +'s in my book) it is also more easily avalable.

I think for most people with good Bh5 there is litte point in getting TCCD, there may be slight advantages for some people, but its not guaranteed.

Bh5 has the advantage of not requiring such a high HTT, which may benefit people who's systems dont like higher HTT. It may also help tune your OC as you can run more deviders (higher HTT and memory running at under 1:1) which might help to max your CPU clock better. TCCD must pretty much run 1:1 unless your system can do insane HTT speeds.
Also if you can get good Bh5 that will do over 270 then you probably will beat TCCD anyway... but not all Bh5 will go that quick. In my experience most wont unless you go over 3.5v... which isnt much use for anything other than a bench.

[charlie]

No offense but I think you need to stay on the ground little bit harder
You mean, DDR660 7-3-3-2.5-1T ??? SuperPi 8M ??? I'll let you try that first With reasonable timmings this is gonna be VERY HARD...

ok, ok the cans of AMP are catching up to me
how bout 310@2.5-3-3 (do yu have 256mB sticks?)

[EnJoY]

Bachus, your BH5 timings are far from optimal, was this intentional or necessary to acheive the speed you've got, or did you just think those were the tighest timings?

[saaya]

interesting....
but I think P95 is useless. If a rig can do everything that it's required to do, it's STABLE. If you run a network of college 'puters researching the DNA sequence, that's ONE definition of stability, if yuo play doom3 that's another definition of stability.
From dictionary el-Charlie:
Stability= The ability to run reliably during a required task.
In fact, if all you do is stare at the desktop, a suicide screenshot IS 100% stable. If you're benching sPi, a completed run IS 100% stable.

And as far as 512mB vs. 1gB... for benching (racing) you don't need 1gB, except for Bank interleave. So it's a valid comparo to pit 512mB tccd against 1gB VX. And my G.Skill go 310ish at 2.5-3-3

thats 100% not whats stable for me! if its not prime stable then theres a chance that my system will crash eventually wich i dont want and cant afford to happen...

[bachus_anonym]

Bachus, your BH5 timings are far from optimal, was this intentional or necessary to acheive the speed you've got, or did you just think those were the tighest timings?

I tried to show tight TCCD vs tight BH5... BH5 can be tweaked little bit more comparing to TCCD, but what looks great (lower values in A64 Tweaker, like CAS 1.5 or TRAS 0 or TWTR 1 or lower refresh rates) doesn't necessarly give much higher performance besides slightly better numbers in Sandra Buffered.
..

BH-5 can be tweaked even more but higher on MHz you go harder it is to maintain stability...

[Major]

From my point of view its pretty much a draw.

"most" BH5 seems to max out at 265 with sensible voltage (around 3.2v)

This is similar in performance as TCCD at around 300 with lesser timings (as bachus shows above)

The TCCD is avalable in larger modules, and does not require as high a voltage (both +'s in my book) it is also more easily avalable.

I think for most people with good Bh5 there is litte point in getting TCCD, there may be slight advantages for some people, but its not guaranteed.

Bh5 has the advantage of not requiring such a high HTT, which may benefit people who's systems dont like higher HTT. It may also help tune your OC as you can run more deviders (higher HTT and memory running at under 1:1) which might help to max your CPU clock better. TCCD must pretty much run 1:1 unless your system can do insane HTT speeds.
Also if you can get good Bh5 that will do over 270 then you probably will beat TCCD anyway... but not all Bh5 will go that quick. In my experience most wont unless you go over 3.5v... which isnt much use for anything other than a bench.

All that ! Plus I like having 300mhz worth of play room to maximize your OC

[saaya]

very sweet michal! very nice results!

some here are talking about the htt speed, wich is almost completely irrelevant, but i think you mean the fsb, not the htt speed.
i know its notactually the fsb, but some mobo makers started calling it fsb and it was adopted as its a known word and pretty much resembles what we mean when we say fsb.

we used to have only the fsb, now on a64s we have the bus between cpu and the memory and the bus between the cpu and the system wich are now two independant busses. both are beeing referred to as the HTT bus by different people though, can we all agree to call the bus between the cpu and the memory the fsb and the bus between the cpu and the system the htt bus?
i think this will save us from many understandings

michal, all we need now to get an idear of the speed at wich bh5 and tccd meet performence wise in the different benchmarks is at least one more meassurement.

it can be done even at 200mhz speeds as long as the cpu speed is the same for the bh5 and the tccd system again if you can make the same comparison you made at 2 more settings we can make a nice table wich should show what speed bh5 has to run at to match tccd at XXXmhz or vice versa

[mdzcpa]

Great thread! Subscribed!

[saaya]

sticky

[bachus_anonym]

@saaya,

well, I think that little table shows pretty much one thing...

TCCD @ 7-3-3-2.5-1T needs about 30-35MHz to catch up with BH-5

TCCD @ 7-3-4-2.5-1T needs more than 60Mhz to keep up with BH-5

It also comes down to A64 CPU one is using. Say, you have a 3000+ Winnie with x9 multi... You know you can pull 2.7GHz on it but your BH-5 sucks and can't go above 260. But you found some TCCD that does 300 @ 7-3-3-2.5-1T...
What you do ??? Of course you grab TCCD - it will VERY likely outperform that BH-5 at everything since you just can't maximize your PC by running max CPU and max BH-5 clocks at the same time...

EDIT: I wish I had either decent PCIe video card or one more Winnie. My Neo2 is collecting dust as well as X800XT PE at this moment. And I just don't want to extract my Winnie from DFI and put it into Neo2 to do some serious 3D benches and game benchmarks to show even more comparisons...

[saaya]

@saaya,

well, I think that little table shows pretty much one thing...

TCCD @ 7-3-3-2.5-1T needs about 30-35MHz to catch up with BH-5

TCCD @ 7-3-4-2.5-1T needs more than 60Mhz to keep up with BH-5

It also comes down to A64 CPU one is using. Say, you have a 3000+ Winnie with x9 multi... You know you can pull 2.7GHz on it but your BH-5 sucks and can't go above 260. But you found some TCCD that does 300 @ 7-3-3-2.5-1T...
What you do ??? Of course you grab TCCD - it will VERY likely outperform that BH-5 at everything since you just can't maximize your PC by running max CPU and max BH-5 clocks at the same time...

EDIT: I wish I had either decent PCIe video card or one more Winnie. My Neo2 is collecting dust as well as X800XT PE at this moment. And I just don't want to extract my Winnie from DFI and put it into Neo2 to do some serious 3D benches and game benchmarks to show even more comparisons...

cant you please run two more comparison at a different speed? please? i really would like to make a scale that shows where tccd and bh5/dh5 meet!

and about your example with a 3000+ that does 2700mhz and your bh5 only does 260... wouldnt 8x337 with the 200/180 memory divider work? if oskars memory divider table for the a64 is correct then this should result in 259mhz memory speed
i dont know if 337fsb could be a problem though... but i dont think it would be with a htt multi of 3x or 2.5x or 2x , right? randi (mricee) managed to get 500fsb and dominic (evilspork) got 400fsb+ our of the box on his dfi...

[bachus_anonym]

cant you please run two more comparison at a different speed? please? i really would like to make a scale that shows where tccd and bh5/dh5 meet!

just write exactly what speeds you want me to run in at...

and about your example with a 3000+ that does 2700mhz and your bh5 only does 260... wouldnt 8x337 with the 200/180 memory divider work? if oskars memory divider table for the a64 is correct then this should result in 259mhz memory speed

Doesn't it go like that ?

337x8 = div180 8/9 (DDR600) div166 8/10 (DDR538) div150 8/11 (DDR490) ?

[charlie]

cant you please run two more comparison at a different speed? please? i really would like to make a scale that shows where tccd and bh5/dh5 meet!

and about your example with a 3000+ that does 2700mhz and your bh5 only does 260... wouldnt 8x337 with the 200/180 memory divider work? if oskars memory divider table for the a64 is correct then this should result in 259mhz memory speed
i dont know if 337fsb could be a problem though... but i dont think it would be with a htt multi of 3x or 2.5x or 2x , right? randi (mricee) managed to get 500fsb and dominic (evilspork) got 400fsb+ our of the box on his dfi...

yeah, good idea saaya... a graphing chart showing different timings crossing at different performance levels.


C

[EMC2]

just write exactly what speeds you want me to run in at...

290Mhz, 2-2-2-5, 1T and 375Mhz, 2.5,3,3,6

<re-engages lurking device and runs for the hills>

[Holst]

Yep, if you could make a graph like that it would be very interesting.

I think the difficulty in comparing one to the other is allways in the timings you choose and trying to make the CPU speed the same.

I suppose you can get a reasonable comparison without tweaking all the timings to max for both sticks.

Also another thing to consider.....

I run a 754 system with single chanel memory.
I think that TCCD with higher throughput "might" do better on 754 than on dual channel 939 when compared to Bh5 at lower latency.
Unfortunately I dont have TCCD to test out.
Anybody have an oppinion on this?

[saaya]

just write exactly what speeds you want me to run in at...


Doesn't it go like that ?

337x8 = div180 8/9 (DDR600) div166 8/10 (DDR538) div150 8/11 (DDR490) ?

huh? i dont know what it shows in bios, but as i said, if oskars frequency table is correct it should be 259mhz (or are the memory dividers not fixed and the memory speeds completely change again if you increase the fsb from 200 to 201?)

what speeds i want you to run?
whoohoo i took over control on michals pc! hahahaha
let me check
will post in 5 mins

and holst, the difference on a 754 plattform will be minimal i think, the throughoutput isnt that much bigger to really such a big difference... maybe i can get a stick of tccd on cebit though, then i could run some tests as i have my dfi 754 with a 3000+ and my dfi 939 with a 3200+ and a lot of bh5 and other memory to test (its all old 256mb sticks, but that shouldnt make a big difference ,right?)

[saaya]

michal, can you please do

330x7 2310mhz
289x8 2312mhz
257x9 2313mhz

330x6 1980mhz
283x7 1981mhz
247x8 1976mhz

for us to make the table with the results?
thx!

[SynGroW]

What kind of voltage are you throwing at that BH-5 compared to the TCCD?

About 3.7v to 3.8v under load (booster drops a little, IDLE it gets 3.8/3.85v)

These voltages are only usefull for benching ofcourse. So for gaming TCCD is the way to go I think.

SynGroW, Bh-5 seems to win there indeed and I have a feeling that the difference would have been slightly bigger if you had used A64Tweaker and ran both at the very optimal settings.

When you look at both of my Pifast runs which ofcourse are fully A64 Tweaker tweaked the difference is bigger.

39.55 @ 3008MHz vs 39.84 @ 3020MHz

The 12MHz advantage of CPU speed for the TCCD will help about 0.07s I think so the difference in pifast is about 0.35s!
BH-5 wins ofcourse.

btw, I also think TCCD can't beat these lobby scores in 01
http://service.futuremark.com/compare?2k1=8423012

[saaya]

These voltages are only usefull for benching ofcourse. So for gaming TCCD is the way to go I think.

hmmm i dont know about that, michal isnt the only one who said that he has to drop his max tccd oc around 20-30mhz before its really stable... dont know if that drop is requiered to reach prime stability only and if you can play games at higher speeds than prime stability just fine...
my expirience has always been that if an oc isnt prime stable then games will crash sooner or later, wich is why i test a lot with prime to avoid game crashes or reboots... so far successfully

[Shroomalistic]

Ive got a 1gig kit of GSKILL 4800LA and a 512mb kit of GSKILL 4800LA coming probly next week. Ill do some test for you guys too. I also got a Gig of TwinMOS speed premium coming at the end of this week and I already have 2x256mb of the stuff. This should be a great combo of ram for testing and I will for sure let you guys know how it goes.

[SynGroW]

hmmm i dont know about that, michal isnt the only one who said that he has to drop his max tccd oc around 20-30mhz before its really stable... dont know if that drop is requiered to reach prime stability only and if you can play games at higher speeds than prime stability just fine...
my expirience has always been that if an oc isnt prime stable then games will crash sooner or later, wich is why i test a lot with prime to avoid game crashes or reboots... so far successfully

I'll test the gskill for primestability in a few weeks probably.
My cpu is dead and maybe my mobo too so I can't test it today.

[HARDCORECLOCKER]

Maybe best is a mix of still good timings and highest possible HTT with these - like here:

http://www.xtremesystems.org/forums/...ad.php?t=55387

[esdee]

my personal opinion is that the best OCing RAM wins.
since BH5-UTT can do 270@2-2-2-5
and TCCD can do, lets say, 320@2.5-3-3-7 to catch up with the above winbond chips

it's all about how high your ram can get...

[iddqd]

I sure hope you mean "high bandwidth" or "high frequency" (the two are most directly related) instead of "high HTT". Having high HTT = 400Mhz, and memory at 1:2 (=200MHz) is pretty damn useless.

You could try pulling tRcd=0 out of your memory, at 200Mhz or so... although I don't think CH-5 UTT will do that, although BH chips might, with enough voltage. That shoud pull out some fairly marginal performance gains... Otherwise, high bandwidth gains very little (unless we're talking about Intel here, but then again TCCD won't do 320MHz on Intel's silly non-integrated memory controllers, so it's an empty discussion then..). Low latency (<25ns) won't pull out that much either, unless you can kill tRcd of course, which would actually help a lot more. Well, there's only one way to find out .

[saaya]

Ive got a 1gig kit of GSKILL 4800LA and a 512mb kit of GSKILL 4800LA coming probly next week. Ill do some test for you guys too. I also got a Gig of TwinMOS speed premium coming at the end of this week and I already have 2x256mb of the stuff. This should be a great combo of ram for testing and I will for sure let you guys know how it goes.

cool! thx a lot josh!

esdee, ive only seen 2 or 3 people with 320mhz 2.5-3-3+ and 2 of those are gsill reps with special sticks ... i havent seen many people hitting 300mhz 2.5-3-3 and only very few hit 320+
not all vx does 270 though... or does it? the 3200vx did an average of 260 iirc... ocz said the yileds improved, thats why they released ocz vx 4000 hmmm so maybe those do 270 on average... if they do, then its a no brainer imo...