GSKILL FLARE achieves highest AMD ram frequency

Quote:

---

Originally Posted by BeepBeep2

I dont understand, what's wrong with 1800 CAS 6?:shrug:

---

That 2300-2400 CAS 6 is possible on Intel :D

Quote:

---

Originally Posted by M.Beier

That 2300-2400 CAS 6 is possible on Intel :D

---

Who cares if the uncore is not capable of using that bandwidth?

Quote:

---

Originally Posted by massman

Who cares if the uncore is not capable of using that bandwidth?

---

You are right, but was simply a reply if 1800 C8 is bad, and fact is Intel has the stronger memory handling at the moment, I dont think anyone questions that :)

Uncore is locked on Lynnfield, so taking it being the i5-750, you have 16x multi for uncore, 5 multi for ram..... 2300 = 230bclk, 2400 = 240bclk.... That means; 230x16 = 3680mhz v 240x16 = 3840mhz, which for dual channel actually is alot for bandwidth, actually not that significant loss my friend.

Quote:

---

Originally Posted by youngpro

OH REALLY? i dont even need them disabled, just disabled them to see if it would help and never re-enabled... thanks dude

thanks all for your kind comments... i dont normally bench this sort of thing, but it is definatly addictive, especially when only 10mhz away from 1200...

---

Cool man, I discovered it while pushing a 940 a while ago. Disabled cores somehow give a rejected validation :)

Quote:

---

Originally Posted by Oj101

Disabled cores somehow give a rejected validation :)

---

Sure, but anything can give you a rejected validation. This is my rejection list for AMD - 2010 only:
http://valid.canardpc.com/show_oc.php?id=1195750
http://valid.canardpc.com/show_oc.php?id=1195722
http://valid.canardpc.com/show_oc.php?id=1179163
http://valid.canardpc.com/show_oc.php?id=1151384
http://valid.canardpc.com/show_oc.php?id=1145535
http://valid.canardpc.com/show_oc.php?id=1145527
http://valid.canardpc.com/show_oc.php?id=1141491
http://valid.canardpc.com/show_oc.php?id=1140714
http://valid.canardpc.com/show_oc.php?id=1140511
http://valid.canardpc.com/show_oc.php?id=943983
http://valid.canardpc.com/show_oc.php?id=943961

46 % of my validations were rejected. It happend with different OSes, mainboards, chipsets, CPUs and core amounts. It´s just random and annoying as hell.

@youngpro: Great results! Did you test how much you can gain from subzero temps? I figured out a DDR3-2200+ DC validation @ air is done easily within a few minutes, so I´m curious what to expect by changing to subzero. Would have tried that weeks ago, but lost confidence about non-rejected results...

Quote:

---

Originally Posted by M.Beier

You are right, but was simply a reply if 1800 C8 is bad, and fact is Intel has the stronger memory handling at the moment, I dont think anyone questions that :)

Uncore is locked on Lynnfield, so taking it being the i5-750, you have 16x multi for uncore, 5 multi for ram..... 2300 = 230bclk, 2400 = 240bclk.... That means; 230x16 = 3680mhz v 240x16 = 3840mhz, which for dual channel actually is alot for bandwidth, actually not that significant loss my friend.

---

I definately said 1800 C6 (as in 6-6-6), which is pretty much the limit of good Elpida Hyper.

As for CAS 7, CAS 8, especially CAS 7-9-7 :mad: on amd @ 2000?...gskill flare is nothing. Just a name. It's PSC. boo hoo, you can get that way cheaper than the 2000 7-9-7 kit.

Quote:

---

Originally Posted by Hyperhorn

Sure, but anything can give you a rejected validation. This is my rejection list for AMD - 2010 only:
http://valid.canardpc.com/show_oc.php?id=1195750
http://valid.canardpc.com/show_oc.php?id=1195722
http://valid.canardpc.com/show_oc.php?id=1179163
http://valid.canardpc.com/show_oc.php?id=1151384
http://valid.canardpc.com/show_oc.php?id=1145535
http://valid.canardpc.com/show_oc.php?id=1145527
http://valid.canardpc.com/show_oc.php?id=1141491
http://valid.canardpc.com/show_oc.php?id=1140714
http://valid.canardpc.com/show_oc.php?id=1140511
http://valid.canardpc.com/show_oc.php?id=943983
http://valid.canardpc.com/show_oc.php?id=943961

46 % of my validations were rejected. It happend with different OSes, mainboards, chipsets, CPUs and core amounts. It´s just random and annoying as hell.

@youngpro: Great results! Did you test how much you can gain from subzero temps? I figured out a DDR3-2200+ DC validation @ air is done easily within a few minutes, so I´m curious what to expect by changing to subzero. Would have tried that weeks ago, but lost confidence about non-rejected results...

---

Fair enough, but in my experience all validations with disabled cores were rejected. When I went four cores it was finally accepted.

Quote:

---

Originally Posted by BeepBeep2

As for CAS 7, CAS 8, especially CAS 7-9-7 :mad: on amd @ 2000?...gskill flare is nothing. Just a name. It's PSC. boo hoo, you can get that way cheaper than the 2000 7-9-7 kit.

---

Don't know.

I just ran a couple of scaling results checking the influence of freq, tcl and trcd on the everest read bandwidth (theoretical max difference) and it learned me that the relative importance of the three variables are:

trcd = 1
tcl = 9,1
mhz = 22,5

Of course, at a certain point, the IMC will bottleneck the memory freq.

Quote:

---

Originally Posted by massman

Don't know.

I just ran a couple of scaling results checking the influence of freq, tcl and trcd on the everest read bandwidth (theoretical max difference) and it learned me that the relative importance of the three variables are:

trcd = 1
tcl = 9,1
mhz = 22,5

Of course, at a certain point, the IMC will bottleneck the memory freq.

---

Sorry, my brain doesn't seem to be functioning correctly atm.

What do the 1, 9, 1, 22, and 5 mean? :confused:

How important they are in relation to each other. Importance = have effect on read bandwidth.

if tRCD importance = 1,
then tCL importance = 9.5
and mhz = 22.5

In other words: the effect of MHz is 22.5x the effect of tRCD.

Quote:

---

Originally Posted by massman

How important they are in relation to each other. Importance = have effect on read bandwidth.

if tRCD importance = 1,
then tCL importance = 9.5
and mhz = 22.5

In other words: the effect of MHz is 22.5x the effect of tRCD.

---

Alrighty.

But how many Mhz to = 22.5

IE 1800 6-6-6 vs 2000 7-9-7...

Quote:

---

Originally Posted by Hyperhorn

@youngpro: Great results! Did you test how much you can gain from subzero temps? I figured out a DDR3-2200+ DC validation @ air is done easily within a few minutes, so I´m curious what to expect by changing to subzero. Would have tried that weeks ago, but lost confidence about non-rejected results...

---

well in terms of max validation, i top out on air at 1108 and scale to 1180~ with LN2

for CPU (this is to respond to thomas question also about gains from making IMC cold) not much, maybe 50MHz, but I did not want it to be a problem... before I put memory under LN2 I had no idea if IMC would be a bottleneck, so safer for me to put it under SS than to run it on air..

that being said I still don't know if its a bottleneck and thats something I want to know, so I will be trying to CPU at LN2, maybe -100 instead of -45~ of SS and seeing what impact that has on memory clocking..

to answer a question partially. In the particular board he is using the IMC of CPU will scle till -80 ish....after that at least on asus you will take a huge IMC hit or at least with elpida IC's. 2160 ram is cake on air, drop below -80 on cpu and wall at 2000 ram basically boot on stock cooler than freeze and same settings fail.

1800 is not the limit of good elpida 6-6-6. thats the limit of bad elpida. good does 1900 6-6-6....even 2000 but not 32m stable.

Quote:

---

Originally Posted by chew*

to answer a question partially. In the particular board he is using the IMC of CPU will scle till -80 ish....after that at least on asus you will take a huge IMC hit or at least with elpida IC's. 2160 ram is cake on air, drop below -80 on cpu and wall at 2000 ram basically boot on stock cooler than freeze and same settings fail.

1800 is not the limit of good elpida 6-6-6. thats the limit of bad elpida. good does 1900 6-6-6....even 2000 but not 32m stable.

---

I sorta meant 24/7...nevermind :ROTF:

I've seen a lot of kits of Hyper that fail to do 1800 6-6-6 at reasonable volts...the majority so far. Corsair 1600 C6 kits, Corsair 2000 C8 kits, my kit...I mean, sure GTX2 is nice but...

I don't know.

Quote:

---

Originally Posted by BeepBeep2

I sorta meant 24/7...nevermind :ROTF:

I've seen a lot of kits of Hyper that fail to do 1800 6-6-6 at reasonable volts...the majority so far. Corsair 1600 C6 kits, Corsair 2000 C8 kits, my kit...I mean, sure GTX2 is nice but...

I don't know.

---

I have non GTX2 kits that do 1800 fine ;) in fact all my sets do 1800 fine.

It's 1900+ that only 3 sets can do on ambient others scale to that speed cold fine.

You have personally used the ram listed to make that statement?

The variance with hypers is trcd. And tbh.....you get what you pay for, buy cheap get a lower bin, buy higher get a higher bin.

To make an accurate analysis you must test on all platforms to make any "claim" ;)

Quote:

---

Originally Posted by chew*

I have non GTX2 kits that do 1800 fine ;) in fact all my sets do 1800 fine.

It's 1900+ that only 3 sets can do on ambient.

---

*gazes in suprise*

Quote:

---

Originally Posted by BeepBeep2

I definately said 1800 C6 (as in 6-6-6), which is pretty much the limit of good Elpida Hyper.

---

The STT Speed 6GB 2000C7 kit I got is 1.71v 1800 666 spi 32m and 1800 666 stable:confused:, which embrassed me a lot:(:
the "best" Hyper kit I have does 1.71v 1860 666 spi 32m but only 1760 666 stable:shakes:.

Quote:

---

Originally Posted by chew*

I have non GTX2 kits that do 1800 fine ;) in fact all my sets do 1800 fine.

It's 1900+ that only 3 sets can do on ambient others scale to that speed cold fine.

---

Do you mean full stablity (e.g. LinX and Hyper PI) or spi 32m?
How much voltage did you push on ambient?

Quote:

---

Originally Posted by BeepBeep2

Alrighty.

But how many Mhz to = 22.5

IE 1800 6-6-6 vs 2000 7-9-7...

---

To give you an example:

1800 CL7-8 = 10905 MB/s
1800 CL7-12 = 10825 MB/s

1800 CL7-8 = 10905 MB/s
1800 CL8-8 = 10690 MB/s

1800 CL7-9 = 10858 MB/s
1500 CL7-9 = 10091 MB/s

It s what my kit of Flare can do with CPU/RAM on Stock cooling.

I didnt use C4F to test it but Asus M4A89TD PRO

First screenshot are just here to show the bandwidth evolution :p

1066
http://wd.ch-img.com/1188511-flare-200x20-1067cl7.png
1333
http://wd.ch-img.com/1188521-flare-200x20-1333cl7.png
1600
http://wd.ch-img.com/1188531-flare-200x20-1600cl7.png
2000
http://wd.ch-img.com/1188541-flare-250x16-2000cl7.png
2140
http://wd.ch-img.com/1188551-flare-268x15-2142cl7.png
2200
http://wd.ch-img.com/1188561-flare-275x14-5-2206cl7.png

IC's
http://wd.ch-img.com/1188581-flares-psc-xee.jpg

That latency is awesome for 7-10-7 :eek: You still owe me a PM ;)

Quote:

---

Originally Posted by shadewither

The STT Speed 6GB 2000C7 kit I got is 1.71v 1800 666 spi 32m and 1800 666 stable:confused:, which embrassed me a lot:(:
the "best" Hyper kit I have does 1.71v 1860 666 spi 32m but only 1760 666 stable:shakes:.


Do you mean full stablity (e.g. LinX and Hyper PI) or spi 32m?
How much voltage did you push on ambient?

---

I guess my credibility only carries me so far eh?
http://www.xtremesystems.org/forums/...&postcount=420
http://hwbot.org/community/submissio...in_43sec_406ms


Lets not derail Pro's thread anymore and stick to gskill flare.

Thanks for sharing. what voltage did you give for 2200MHz?? My Flare can not get this timing stable.

on air? i dont need to use more than 1.7v and if i do it actually hurts my results for the most part...

please make sure you play with your subtimings because these are key to achieving anything over 2100MHz on AMD platform.. you might also want to look at drive strengths and driving the ram very mildly.. try the lower drive strengths

:)Thanks. I will try some more timing.

I wonder Pt1t the user if it is difficult to disassemble the heat sink of memory gskill flares?
I am to receive one of these memory kit, and will use dissipation by watercooler on it and for that I have to disassemble the same! already had an experience, not pleasant, as my old corsair dominator ddr2!

thanks