*HOT* | Corsair DOMINATOR GT | 6GB (3x2GB) | DDR3-2000 7-8-7-20 @ 1.65V | *HOT*

[pinto]

What's your settings,board, voltage?

[cstkl1]

on a 920,r2e

if i can run 2000mhz CL8 fine with no errors at qpi 1.48( set 1.53750),dram 1.65v,bclk 200 and also tested the xmp profile also
which was 1.6v for qpi set 143 something for bclk
all dram voltage at 1.65v
pumped dram to 1.67 also no luck.

is there other settings for a latency other than dram voltage??

[cstkl1]

if i was to run 7-8-7-20
end up with 4gb left
and if stressed fails immediately with hci

[Praz]

if i was to run 7-8-7-20
end up with 4gb left
and if stressed fails immediately with hci

Unfortunately this is going to be the outcome for quite a few users. The Elpida Hyper based sticks can provide more performance then a lot of processors will be capable of running at. At least you are able to run at 8-8-8. Some won't even see that.

[cstkl1]

eh
so ure telling me dram timing has to do with IMC's??
dont u think thats kindda thin as so far only uncore speeds are the issue with imc's and qpi voltage limits.
when the ram can run 8-8-8-24.. at 1.65v.. exact.. if i lessen it even by one voltage.. it gets unstable
isnt the problem with most proc is the ability of running uncore 4000 with the ram 2000mhz
i am still heading that this ram wasnt binned at 7-8-7-20.




see the 4gb issue when cl7 ..
i remembered when i was clocking my dominator 1600 CL8.
when i did adventerous timing that doesnt have enough voltage only the issue above happens

[Praz]

eh
so ure telling me dram timing has to do with IMC's??
dont u think thats kindda thin as so far only uncore speeds are the issue with imc's and qpi voltage limits.

Thin? Not the term I would use. The memory controller is part of the CPU. A weak controller will result in weak ram overclockability. Even though the ram is rated at 2000 running at that speed is overclocking the controller. With enough VTT you may find stability at 7-8-7. However, may well be past the voltage limit one is comfortable with.

Edit: Corsair states that up to 1.60V VTT may be required for 1866 memory depending on the quality of the IMC. That kind of puts 2000 speed requirements into perspective.

http://www.houseofhelp.com/forums/sh...ad.php?t=75918

[ChaosAD]

Why were you unable to order them in Greece?

Ok, let me rephrase then. Ordering from US means that i have to pay 19%VAT once for the product when i order from you and 19%VAT once more in Greek customs over the total invoice that comes from you (Price of ram
+ 19%VAT + shipping). Total cost 80-100 euro over 2133X. Not going to happen

I can...
Corsair for the win my friend....

Do you have experience with all the kits and state your opinion?

[pinto]

on a 920,r2e

if i can run 2000mhz CL8 fine with no errors at qpi 1.48( set 1.53750),dram 1.65v,bclk 200 and also tested the xmp profile also
which was 1.6v for qpi set 143 something for bclk
all dram voltage at 1.65v
pumped dram to 1.67 also no luck.

is there other settings for a latency other than dram voltage??

Did you tried 78720 with lower ram frequency? Try 900-950mhz, it didn't stress the IMC too much. I use 1.44v for the qpi/dram for DDR2000 on my 965 and 1.52v for my 920, don't forget to let the TRFC setting to auto.

[eva2000]

see the 4gb issue when cl7 ..
i remembered when i was clocking my dominator 1600 CL8.
when i did adventerous timing that doesnt have enough voltage only the issue above happens

i7 memory quality not only effects frequency head room but ability to run certain timings so you have 10x core i7 cpus where only 5 or 6 of these cpus can do DDR3-2000Mhz at CAS7.

4GB detection usually occurs when vdimm and cpu vtt (uncore) differential balance isn't correctly set to what your core i7 cpu likes.. so play with the vdimm and cpu vtt somewhat. Same improper balance between vdimm and cpu vtt can also cause everest/sandra bandwidth to report lower than it should even if full tri kit modules are detected.

These kits are validated at 1.65v vdimm and 1.6v CPU VTT (uncore) volts.

Right now I'm working backwards for my tests starting with max intel spec'd 1.35v CPU VTT and seeing how my GTs scale as i bump CPU VTT (vdimm is no longer the sole factor determining memory frequency/timing overclocks)!

[cstkl1]

Did you tried 78720 with lower ram frequency? Try 900-950mhz, it didn't stress the IMC too much. I use 1.44v for the qpi/dram for DDR2000 on my 965 and 1.52v for my 920, don't forget to let the TRFC setting to auto.

ok lets assume i tried everything
yes its very thin arguement assuming that vtt voltage has to do with dram timings.

i have used the vtt up to 1.65v and no diff..
the errors still there
the errors lessen only when i up the dram voltage.
i am not yet comfortable to be running my vdimm voltage past 1.7v for stress testing yet on i7. but the errors are less around 4-5
this suggest only that the ram is not binned at CL7.
i have been running the other dominator at 1600 CL7 for nearly two months.
to answer the other question
the vtt voltage is just for the uncore not for the ram timings.
y the possibility that this set is not binned at cl7 is not a acceptable explanation.

i find that odd. latencies has nothing to do with vtt unless somebody here can prove that

[cstkl1]

i7 memory quality not only effects frequency head room but ability to run certain timings so you have 10x core i7 cpus where only 5 or 6 of these cpus can do DDR3-2000Mhz at CAS7.

4GB detection usually occurs when vdimm and cpu vtt (uncore) differential balance isn't correctly set to what your core i7 cpu likes.. so play with the vdimm and cpu vtt somewhat. Same improper balance between vdimm and cpu vtt can also cause everest/sandra bandwidth to report lower than it should even if full tri kit modules are detected.

These kits are validated at 1.65v vdimm and 1.6v CPU VTT (uncore) volts.

Right now I'm working backwards for my tests starting with max intel spec'd 1.35v CPU VTT and seeing how my GTs scale as i bump CPU VTT (vdimm is no longer the sole factor determining memory frequency/timing overclocks)!

so eva ure asking me to change the difference.. ok will try again..
and i get 6gb only if i run 1.5375 set... if i down it to 1.51v set thats when the 4gb issue occurs more commonly.

so i see ure point but up to what voltage should i run on the vdimm for 2000mhz CL7.

[cstkl1]

pinto the tRFC is set to auto.
never saw much of a gain from lowering it in x58 as compared to RE.

[Praz]

Very good explanation George. I'm still wondering how memory manufacturers are going to handle this. Compared to past platforms there seems no sure way for the average user to isolate these types of issues down to the memory itself. At least a method that will convince the user that it is not memory related. It's going to be a headache for the manufacturers for sure.

[cstkl1]

also are u guys so sure that instead of doing the above that
its just not bad luck that my kit cant run CL7

[cstkl1]

testing with lower vtt with memtest86..
hmm interesting 6gb now detected.

[cstkl1]

hmm so far no error in test 5 like above.. hmm
sorry praz for being a idiot.. just was pissed..

[Praz]

hmm so far no error in test 5 like above.. hmm
sorry praz for being a idiot.. just was pissed..

I didn't take your post as anything that needed an apology for. There really needs more info put out concerning memory and processors. Nobody is going to be happy if after spending upward of $500.00 the memory won't run at its rated speed.

If after trying different voltages and voltage relationships the speed still can't be reached the only options left are different processors or memory. Which for most users will require an RMA of the ram.

[eva2000]

so eva ure asking me to change the difference.. ok will try again..
and i get 6gb only if i run 1.5375 set... if i down it to 1.51v set thats when the 4gb issue occurs more commonly.

so i see ure point but up to what voltage should i run on the vdimm for 2000mhz CL7.

vdimm is a secondary voltage to play with really as it's all in the cpu vtt volts first and foremost - unfortunately cpu vtt comes down to the individual cpu.

balance i'm talking about is within that 0.5v difference between vdimm and cpu vtt - elpida 50nm ics seem to like a closer relationship between vdimm and cpu vtt once you pass DDR3-1866-2000Mhz speeds.

My GTs do quad super pi 32M around 20x 188bclk with 10x mem multi @DDR3-1880mhz 7-8-7-20 1T at 1.65v when i restrict CPU VTT to around intel spec's of 1.36v CPU VTT versus Corsair's validated 1.6v. Still testing how it scales in terms of CPU VTT.

Would definitely love to test more 6GB Elpida 50nm Hyper based IC kits later on, to see if it's specific to my i7 920s cpus, board/bios or GT modules itself or a combination of factors.

Very good explanation George. I'm still wondering how memory manufacturers are going to handle this. Compared to past platforms there seems no sure way for the average user to isolate these types of issues down to the memory itself. At least a method that will convince the user that it is not memory related. It's going to be a headache for the manufacturers for sure.

Only one way really = education More of same/similar information getting out to the masses is only way. Raja and Gary at Anandtech also touched on this with CPU VTT differences between Core i7 cpus http://www.anandtech.com/cpuchipsets...oc.aspx?i=3502. CPU and MEM performance/stability on i7 seems to be linked heh

also are u guys so sure that instead of doing the above that
its just not bad luck that my kit cant run CL7

could be, or could just your i7 cpu doesn't like that particular kit. Your same kit on another i7 cpu might perform drastically different from your current i7 cpu.

Both my twin i7 920 3836A756 cpus perform nearly identical for memory just one cpu needs way less CPU VTT to run same clocks > http://i4memory.com/105269-post4.html

testing with lower vtt with memtest86..
hmm interesting 6gb now detected.

it can be a fine balance so far come across 3 types of core i7 cpus from my rounds on forums and my own experience

Type 1 - core i7 cpus that need high to very high cpu vtt to overclock for memory and bclk.

Type 2 - core i7 cpus that like low-medium low cpu vtt to overclock for memory and bclk and scale with cpu vtt and memory voltage

Type 3 - core i7 cpus that only like low-medium low cpu vtt to overclock for memory and bclk BUT DO NOT scale for either cpu vtt and memory voltage or both

low cpu vtt = 1.3-1.35v
medium cpu vtt = 1.35-1.5v
high cpu vtt = 1.5-1.55v
very high cpu vtt = >1.55-1.65v

hmm so far no error in test 5 like above.. hmm
sorry praz for being a idiot.. just was pissed..

memtest is only the first stage of testing - next is windows

[cstkl1]

now i can pass memtest86+
but will fail memtest hci
difference is i increase the vdimm to 1.67
reduce the vtt to 1.5187set
6gb detected..

first thought must be the ioh/ich pair again

[cstkl1]

memtest hci prooved only stable when vdimm at 1.72v

testing now with a 965xe.
i am convinced that this ram is not binned right no offence folks.

[Praz]

memtest hci prooved only stable when vdimm at 1.72v

testing now with a 965xe.
i am convinced that this ram is not binned right no offence folks.

It may well be the ram. I think all we are trying to say is without components that one knows well it's difficult to isolate the issue to a single part.

[bachus_anonym]

memtest hci prooved only stable when vdimm at 1.72v

testing now with a 965xe.
i am convinced that this ram is not binned right no offence folks.

What's your motherboard? Do you have the ability to measure actual VDIMM with a multimeter in a DIMM socket? The reason I ask is that e.g. Gigabyte X58 boards as well as DFI X58 deliver almost spot on VDIMM voltage (in reference to BIOS setting) under full load, but we observed that e.g. Asus P6T or R2E would undervolt by up to 40mv under full load, with 3 DIMMs installed. Believe it or not, 40mv on Elpida Hyper may make it or brake it @ 7-8-7...

To be honest, the very kit you received was went through our 7-8-7 production test and qualification as well as was retested under Vista64 with HCI Memtest and SMT enabled before we shipped it, to be on a sure side, since your shipment was already delayed. I'd like to assure you that we put super-super-extra care into manufacturing and testing Dominator 2000C7GTs.

What Praz and George said regarding perfectly matched CPU/mobos is very accurate, and anticipating that we put together some notes included with every 2000C7GT your kit as to what to expect when DDR3-2000+ clocking at 7-8-7.

Dominator 2000C7GT is a bleeding edge product that we target towards very particular audience, who understands Core i7 memory overclocking well, and is ready to spend some extra time tuning their setups. I'm not trying to lay blame, far from it, but only to point out that with 2000C7GT on Core i7 there may be more to it than just plugging it in and have it work, as we approach the limits of what Core i7 can handle without going sub-ambient on CPU on it to improve memory overclockability.

I'd like you to be satisfied with 2000C7GT... But if there's a doubt in your mind, we will take it back - replace it or refund it.

Thank you for understanding.

[cstkl1]

i am quoting from a dmm on voltages on load.
testing now with a 965xe. just took it from the office. since my friends 965xe also prooved the same.
i really trying not to rma it bachus.. as it took heck of a long time to get here and sending it back is just another cost of shipping for me.

hmm will get the dfi tommorow since the stock just came yesterday for the ut and dk.

[Yellowbeard]

Ok, let me rephrase then. Ordering from US means that i have to pay 19%VAT once for the product when i order from you and 19%VAT once more in Greek customs over the total invoice that comes from you (Price of ram
+ 19%VAT + shipping). Total cost 80-100 euro over 2133X. Not going to happen
:

I see what you mean and I am sorry to hear that. Perhaps in the future we can better serve our customers online. Our store is very new to us so we have plenty of room to improve.

[andyr(uk)]

I see what you mean and I am sorry to hear that. Perhaps in the future we can better serve our customers online. Our store is very new to us so we have plenty of room to improve.

Same here tbh, payed tax in US then again in UK , got the c8 kit but payed more than the c7 kit, could the product be linked through the Countries local Corsair shop ? be less expensive for us.

Also to you guys with i7 experience which is the best x58 board to go with as i have my ram and cpu and i am about to buy the board , was going to go for the asus RE2 but not too sure now