Best ram for AMD platform

[mav2000]

Do the 6100 FX have as strong an IMC as the 8120/8150?

[froxic]

Its pretty same as 81XX lineup

[gkiing]

Mushkin 996997
996997.png
llanooc2.pngllanooc1.png

[Daveburt714]

Even though my mem clocking skills are a little weak, the 8Gb kits I've been trying lately have gotten me pretty frustrated, so I've been pulling out some old stuff...

The last few days I've been playing with the Corsair GT 2000/CL8 kit. It did pretty well on Deneb, but I had trouble with it on Thuban and the FX.
Anyway tonight I gave up on it's 2000 8-8-8 rating (and the 1866 divider) and tried tightening timings @ 1600.

The results were actually half way decent. Either the board/mem or chip seems to like the 1600 setting much better!

It's only a 15 minute Blend stable shot (Custom/3Gb usage), but I didn't go easy on it (surfing the web and listening to tunes while running) and it seems pretty darn good.

1950 7-7-7



Thanks for the advice Zeus (and Beep)...
Finally got the Hypers stable on the 8120 with decent clocks & sweet timings!
I can live with this for awhile...

[Spcial]

DDR2 706 MHz 5-5-5-18 1M with Team Xtreem 1300C6

[Barr3l Rid3r]

Even though my mem clocking skills are a little weak, the 8Gb kits I've been trying lately have gotten me pretty frustrated, so I've been pulling out some old stuff...

The last few days I've been playing with the Corsair GT 2000/CL8 kit. It did pretty well on Deneb, but I had trouble with it on Thuban and the FX.
Anyway tonight I gave up on it's 2000 8-8-8 rating (and the 1866 divider) and tried tightening timings @ 1600.

The results were actually half way decent. Either the board/mem or chip seems to like the 1600 setting much better!

It's only a 15 minute Blend stable shot (Custom/3Gb usage), but I didn't go easy on it (surfing the web and listening to tunes while running) and it seems pretty darn good.

1950 7-7-7



Thanks for the advice Zeus (and Beep)...
Finally got the Hypers stable on the 8120 with decent clocks & sweet timings!
I can live with this for awhile...

what advices? 1t or 2T?

[Mafio]

2300 MHz 7-8-7-22 1T super-pi 32m stable, not bad

[Daveburt714]

what advices? 1t or 2T?

The advice was actually more encouragement.
Don't mess with too many low level sub timings, keep v's below 1.75 for 24/7 use and bump chipset NB if needed (didn't need to).
Command rate is 1T. I had tried 2T and it didn't offer any real advantage (@ 1866 anyway... Haven't messed with it @ 1600).

What really seemed to open them up for me was switching to the 1600 divider.
I just set them to 1600 7-7-7 and then brought the HTRef up until they started failing.

They just play alot better @ 1600 compared to 1866.

[Barr3l Rid3r]

The advice was actually more encouragement.
Don't mess with too many low level sub timings, keep v's below 1.75 for 24/7 use and bump chipset NB if needed (didn't need to).
Command rate is 1T. I had tried 2T and it didn't offer any real advantage (@ 1866 anyway... Haven't messed with it @ 1600).

What really seemed to open them up for me was switching to the 1600 divider.
I just set them to 1600 7-7-7 and then brought the HTRef up until they started failing.

They just play alot better @ 1600 compared to 1866.

yes that's true I have noticed before and forgot it, anyway gigabyte doesn't have too many subtimmings options just the basic ones, wich I used to set them at 4-4-4-8 or 5-5-4-10 trfc at 90ns or 110ns none of them seems to change stability by much

[Spcial]

AMD FX-4100
ASUS Crosshair V
G.Skill PI 2000 6-9-6 @ 1.79v

1156 MHz @ 6-10-7-22 @ SuperPI 32M

[Cecil]

I have PSCs, HCF0s, and Crucial 1.5V 4GB Dimms that Im compairing to each other with a Sabertooth and 8120.

Havent done anything with the HCF0s, but the Crucials do 2000 8-8-8-24 and 2200 9-9-9-27 stable with 1.6375V or less. The PSCs I couldnt get to post at 2200Mhz with 1.65V and even 8-10-8 timings. At 2000Mhz, the best I can get with 1.65V is 7-9-7-24-1t. I dont have time for 32m testing of each kit, so Im just running 1m and 4m. These pics compare the PSCs at 2000Mhz 7-9-7-24-1t and 2t. There is barely any advantage with 2t. Obviously if you can run 1t at the same speed you are limited to, then do so, but if 2t allows more speed or tighter timings, then its worth it. Going to go back to the Crucials and run them, then try the HCF0s.

(for some reason CPU-Z is not showing 1t/2t)

PSCs 2000 797 1T Spi1m.JPGPSCs 2000 797 1T Spi4m.JPGPSCs 2000 797 2T Spi1m.JPGPSCs 2000 797 2T Spi4m.JPG


Crucial Smart Tracer 1866 999 1.5V kit at 2000Mhz 8-8-8-24-1t, Spi1m is a little slower then above, but 4m is a bit faster.

Smart Tracer 2000 888 1t Spi1m.JPGSmart Tracer 2000 888 1t Spi4m.JPG


With tRC set to the SPD rating of 47, 1m was a bit faster (and tied the PSCs result above). From the thread on these, and this tests, it seems they are more stable with a tRC of 47.

Smart Tracer 2000 888 1t trc47 Spi1m.JPG


Smart Tracers @ 2000Mhz 8-8-8-24-2t tRC 47, 2t seems to show wider results for these then the PSCs above.

Smart Tracer 2000 888 2t trc47 Spi1m.JPGSmart Tracer 2000 888 2t trc47 Spi4m.JPG


Smart Tracers @ 2250Mhz 9-9-9-27-2t tRC 47, pretty disapointing how it just tied the above. CPU speed is a tad slower though, tried to get the CPU and NB/HTT speeds as close as possible.

Smart Tracer 2250 999 2t Spi1m.JPGSmart Tracer 2250 999 2t Spi4m.JPG

[mav2000]

Great 8GB kit for Bulldozer on the CH V Formula:



2,250 Mhz CL9 1T, with 1.65V



2,333 Mhz CL10 1T, with 1.675V

More here:

http://www.erodov.com/forums/review-...8gb/50354.html

Working on a 16GB kit now and its looking good as well.

[Cecil]

Mav2000, those pics are way too small to see (at least for my screen anyway). What are the rest of the timings?

[mav2000]

Funny, they dont seem to expand on clicking.

The timings are : First one 2,250 Mhz CL 9-11-10-28-45 1T

Second one: 2,333 CL 10-12-10-28-45 1T

Have not played with any of the timings except first 3.

[Cecil]

Seems Crucial 1.5V 4GB dimms are still best.

[Hyperhorn]

Just something I want to share with you guys as I had a few minutes to test yesterday:
2 x 1 GiB D9GTR (Cellshock CS3222580) + FX-8150 + Crosshair V Formula (UEFI 1102) was really disappointing. I didn't expect any miracles, but the best I could pull of at Prime 95 Blend (custom) was ~ 700 MHz (DDR3-1400) 6-5-5-15. No boot at DDR3-1600/1866 no matter which Vdimm (tried from 1.65 V up to 2.1 V), vCPU-NB (tried from 1.2 up to 1.35 V) or main timings (tried up to 10-10-10-30, 2T). Is Bulldozer really that bad for good old D9GTR or is it me to blame? Have some of you used D9GTx + Bulldozer and what's your experience with it?

btw when going from 2.0 to 2.1 V (no mercy with the IMC ) I noticed memory detected from CPU-Z decreased from 2048 to 2012 MiB and it wasn't possible to change the reference clock with AI Suite anymore (changes had no effect). At 2.0 V and lower values everything was normal again.

I know this Vdimm range is not recommended for FX-CPUs, but it didn't work properly with 1.65 V before, so I was curious what would happen at usual voltages for D9GTx-ICs.

[Mafio]

IMHO it's more likely because of poor BIOS support.
Micron D9-KPT works quite well but not as good as on intel 1155 or 1366, anyway i managed to pull 2700 or 2750 MHz with 2.2 volt or something like that on C5F and FX-4100.
on the other hand Micron D9-JNM doesn't post even at default settings, i've 2 or 3 kits of Crucial value PC3-8500 c7 and they refuse to post with this setup.

hypers are still my first choice with this platform, 2300 MHz 7-8-7 is quite easy.

[BeepBeep2]

Hmm, I'll give it a try with my 2x1GB D9GTS, both are Crucial Ballistix donated to me by chispy.

@Mafio
2300 7-8-7 with what Hypers?...my SuperTalent ProjectX dont scale with tRCD +1 but I can run tRP -1. I can't pull off 2180 7-7-7-21 or 7-8-7-22 at 1.8v but I can easily do 2133 7-7-6-20 @ 1.75v which seems to be my sweet spot for example. I'm sure if I went cold with DI or LN2 I wouldn't have issues pulling 2200+ 7-7-6.

I run 2172 8-8-7-24 1.64v with my Hypers 24/7 @ 2560 CPU-NB (unstable). 2600 CPU-NB is way too unstable on this CPU to be usable. My PSC do 7-10-7 at the same frequency and voltage.

[Mafio]

wait wait, i'm talking about benching, not daily use.
2300 7-8-7 for benching is easy with both g.skill 2133 and kingston hyperx 2000 c8 i have here.

[BeepBeep2]

wait wait, i'm talking about benching, not daily use.
2300 7-8-7 for benching is easy with both g.skill 2133 and kingston hyperx 2000 c8 i have here.

What voltage then?
Maybe my sticks aren't as strong as yours or I'm not as insane when it comes to voltage.

Like I said, my STT ProjectX don't scale with tRCD +1 and I can run tRP -1 about the same as I can tRP = CAS however without (I'm guessing) 1.85v I won't be able to do 2200 7-7-6.

[Daveburt714]

I've been accumulating alot of mem lately trying to find the best ram for FX chips.
So... I decided to compile my results into one post showing the best I've been able to get from each kit while maxing out all components of the system (including IMC, clocks and timings).
At the same time maintaining overall stability (at least 30 minutes Blend max mem stable), and good for 24/7 use.

To be honest, I'm still trying to figure out what all these numbers mean myself, but, I have a decent collection (using various IC's).

FWIW, I'll list them by order (of my preference) and let you folks decide for yourselves.

#1
G.Skill RipJawX (Hynix BFR?) 2133 9-11-10 8Gb (2x4) @ 2222 9-11-10 1.6875v
This will be the kit I stick with for now and runs very solid at these clocks without quirks!
It also gives me good balance and the best core clocks (plus is 8Gb ):



#2
Mushkin Black (PSC ?) 1600 7-8-7 4Gb (2x2) @ 2038 7-10-7 1.675v
This kit has been a staple for me and is still one of my favorite kits... Honestly, it only loses to the RipJaws because it's only 4Gb.


#3
Corsair GT (Hyper) 2000 8-8-8 4GB (2x2) @ 2000 7-7-7 1.70v
This kit offers up some really good numbers (especially in WinRar), but even though it will Blend at these clocks it still has some quirkiness to it.
Sometimes it hangs on post and crashes when encoding video...
This was a great kit on Deneb, but honestly, I feel I've given it enough chances on Thuban and Zambezi and its just too unreliable IMHO...



One thing I've noticed while pushing memory is a serious drop in L2 read speed when you go too far...
Does anyone know if this is just a quirk with AIDA64, or is it a legitimate problem with FX chips?

I haven't really tested this, but it seems like it may be legit and somethng is trashing L2 performance.

[Cecil]

Just so you know, wprime has nothing to do with memory. In fact, most the time I bench it, its better to have slower mem speed and looser timings for wprime.

[Daveburt714]

Thanks Cecil.... I only ran the 32m Wprime to show some amout of stability.
All those clocks were >30 minutes Blend (max mem) stable, I just didn't have SS's to show it, and I didn't have hours to waste.

[BeepBeep2]

I've been accumulating alot of mem lately trying to find the best ram for FX chips.
So... I decided to compile my results into one post showing the best I've been able to get from each kit while maxing out all components of the system (including IMC, clocks and timings).
At the same time maintaining overall stability (at least 30 minutes Blend max mem stable), and good for 24/7 use.

To be honest, I'm still trying to figure out what all these numbers mean myself, but, I have a decent collection (using various IC's).

FWIW, I'll list them by order (of my preference) and let you folks decide for yourselves.

#1
G.Skill RipJawX (Hynix BFR?) 2133 9-11-10 8Gb (2x4) @ 2222 9-11-10 1.6875v
This will be the kit I stick with for now and runs very solid at these clocks without quirks!
It also gives me good balance and the best core clocks (plus is 8Gb ):



#2
Mushkin Black (PSC ?) 1600 7-8-7 4Gb (2x2) @ 2038 7-10-7 1.675v
This kit has been a staple for me and is still one of my favorite kits... Honestly, it only loses to the RipJaws because it's only 4Gb.


#3
Corsair GT (Hyper) 2000 8-8-8 4GB (2x2) @ 2000 7-7-7 1.70v
This kit offers up some really good numbers (especially in WinRar), but even though it will Blend at these clocks it still has some quirkiness to it.
Sometimes it hangs on post and crashes when encoding video...
This was a great kit on Deneb, but honestly, I feel I've given it enough chances on Thuban and Zambezi and its just too unreliable IMHO...



One thing I've noticed while pushing memory is a serious drop in L2 read speed when you go too far...
Does anyone know if this is just a quirk with AIDA64, or is it a legitimate problem with FX chips?

I haven't really tested this, but it seems like it may be legit and somethng is trashing L2 performance.

1. Great!...awfully slow but a lot of ram. Middle of the line with 8GB now that the Crucial/Micron kits have been released.

2. Decent, but not great. Most PSC can do 2133+ 7-10-7 under 1.7v and a lot do 2200 under 1.7v

3. Again decent, but not great. Corsair binned their Hyper tight and unfortunately that's about the worst out of them. GTX2 however, does 2133+ 7-7-7 at the same voltage. I don't think its a matter of being unreliable, but you are trying to push the envelope too much on a kit that's not up to snuff. You should try CL8 or at the very least 7-8-7 IMO. Hypers IMO are the best benching ram for Zambezi. I guess they have some trouble with Llano.

As far as not having hours to waste, set the CPU to something that won't catch fire and "set and forget" as you sleep or go to work.

[Hyperhorn]

Hmm, I'll give it a try with my 2x1GB D9GTS, both are Crucial Ballistix donated to me by chispy.

That should run fine, because some minutes ago I've put 2 x 1 GiB Crucial BL12864BA2009-8SFB into the setup and have no issues at all. I could boot DDR3-1600/1866 at the very first try and now I'm running Prime 95 Blend (custom) at 1,000 MHz (DDR3-2000). So it seems to be an issue with the Cellshock sticks - related to the SPD-EEPROM maybe.