[INQ] Performance RAM will damage your Nehalem

[Donnie27]

What does that have to do with what I asked the other guy? 2.1v for what, DDR2? Now doesn't DDR3 support lower Voltages? What I meant why wouldn't that lower voltage have a lower warning threshold?

One of the reasons for DDR3 is its lower voltage requirement. So why are folks surprised by a lower threshold warning? Also, doesn't dynamic overclocking run up the voltage? If that's the case, it will end up increasing higher anyway.

Anyway, 3 channels of 1333MHz DDR3 will still kill any DDR2 setup out there now and next year Oh man, Nehalem will sucks because I can only get 26GB and not 32GB for example LOL! Oh wait, some of the same folks (running 1.8 to 2.1volts) who came up with SNDS after Intel warned them not to use more than 1.7v on the Northwood, oh brother.

[BrowncoatGR]

1333 is only for the XE. The other variants only go up to 1066 afaik.

[Tony]

http://www.dvhardware.net/article24148.html
http://www.newegg.com/Product/Produc...82E16820146873
http://www.newegg.com/Product/Produc...82E16820231185

Btw, bandwidth matters, not latency.

LOL sure it does...NOT

Access latency, write speeds etc are what counts, read bandwidth adds little to FPS, writes add a lot more...you guys should know this by now

Nahelem may do 1600MHZ with voltages set to 1.6V or so with ram that needs 1.8V on core2....wait and see.

Half the reasons why we run such high voltages are to get around dirty signalling on core2, this was designed out on Nahelem. Low jitter design was mentioned thru out most of the early marketing...I have faith we will see much lower voltages needed on ram than what we see now.

[Eastcoasthandle]

Why haven't we seen DDR3 1600 close to JEDEC of 1.55V or 1.6V (JEDEC of DDR3 is 1.5V)? 2Gigs of DDR3 1600 can come in 1.8V-1.9V. That's a whopping 0.4V difference! JEDEC of DDR2 is 1.8V yet we can get 4Gig kits at 2.0V. Maybe this is another reason why DDR3 is not mainstream!

[T_Flight]

@ Donnie...

Sudden Northwood Death Syndrome! I remember that. Whew, oh man.

Here's a new quote...

"It ain't Xtreme if you kill it."

[SF3D]

I smell some bull s**t

[Rob94hawk]

So instantly rules out 1600Mhz+ DDR3, which kinda defeats the object of using it doesn't it?

We all know theres hardly no gain over DDR2 with low speed DDR3, yes triple channel+on die MC will provide heaps of bandwidth but surely Intel are going backwards here (so much for high end DDR3 prices falling)?

Prices will fall if there is no use for high end DDR3. I'm sure something will be figured out though.

[Astratuner]

guess i'll wiat untill some high speed low Volt memory comes around the corner

[mrcape]

We will know for sure as soon as the samples get around. You can count on the limits to be pushed to the point of no return. Then finally no more of the nehalem news - it's such a bore until the benching starts.

[MpG]

Half the reasons why we run such high voltages are to get around dirty signalling on core2, this was designed out on Nahelem. Low jitter design was mentioned thru out most of the early marketing...I have faith we will see much lower voltages needed on ram than what we see now.

So we're going to see DDR3 come with two different ratings and voltages, depending on what kind of board its going into? The memory forums are going to have a heart attack.

[Shintai]

LOL sure it does...NOT

Access latency, write speeds etc are what counts, read bandwidth adds little to FPS, writes add a lot more...you guys should know this by now

Nahelem may do 1600MHZ with voltages set to 1.6V or so with ram that needs 1.8V on core2....wait and see.

Half the reasons why we run such high voltages are to get around dirty signalling on core2, this was designed out on Nahelem. Low jitter design was mentioned thru out most of the early marketing...I have faith we will see much lower voltages needed on ram than what we see now.

Caches hides most, if not close to all latencies. (Unless you run synthetics)

Even then if you want to measure it. A bloomfield CPU will have alot less latency for memory using some value memory than a Core 2 with super elite edition of some memory that needs to pass the buffers and tracelengths of the MCH and board.

We already see that today with A64 vs C2.

[Blauhung]

1333 is only for the XE. The other variants only go up to 1066 afaik.

that's validated speed that each speed bin has been tested to run at. Please don't confuse that with achievable speed by this community.

All the speed bins have memory and QPI multies unlocked

The main issue as explained by others, is the potential difference between the memory and controller. The voltage applied to each sets the possible potential that the signaling lines see on each end. Higher voltage on the memory means that there will be some direct current flow towards the memory controller. This is destructive, as these pathways on die are not designed to be able to dissipate excess current. As Tony has said, DDR3 voltages have been inflated for the current 45nm line to help with proper signaling to an aging MCH architecture.

Companies can't really bring out their low voltage lines of memory yet, as they are not optimized for the current Penryn platforms. If what Tony says is true though, most current DDR3 kits may just end up stable at lower voltages when operated through the new memory controller architecture on Nehalem, but we will just have to wait for someone to test this for us.

[AliG]

probably same problem the first batch of athlon 64s and phenoms had, the memory controller can't handle the voltage, just up the cpu voltage and you should be fine, otherwise deal with only 1.65v

[XS Janus]

I hope this will bring us closer to JDEC spec 1.5v for faster DDR3 much faster than it would usually take.
This way vendors will have to REALLY make better ram, NOT just overclock it and call it a done deal!

[Shintai]

I hope this will bring us closer to JDEC spec 1.5v for faster DDR3 much faster than it would usually take.
This way vendors will have to REALLY make better ram, NOT just overclock it and call it a done deal!

I couldnt agree more.

[T_Flight]

I like these lower voltages. It means more stability, lower heat, and they will be making good performance RAM. The title of this thread is a misnomer. They already make performance RAM that will not damage a Nehalem CPU. Qimonda RAM has been running 1866 at 1.55...errr, I belive it was 1866. SOmebody correct me if I'm wrong. I'm going off of memory.

For 1.55 that's pretty amazing. These new CPU's are continuously being comapred to old technology and it doesn't equate. You cannot compare mem performance ona system with a MCH on a NBridge to a CPU with an On die IMC. The latencies are FAR lower by design with a IMC.

You will still get lower latencies with looser timings with the newer CPU design than you will with the older MCH and super tight timings with the old way.

We all need rto get past the old way of thinking. This is a completely different kind of running, and to OC it well, we have to listen to those who work with them, and learn the new. As Dr. Who stated, you *must* stick with the physics and electrical properties. That is what OC'ing and performance tweaking is all about. It's not just about the settings. it's what those settings do, and what effect they have.

Higher is not better. Faster is better. There are different ways to get faster now.

[xlink]

hipro5 made some DDR maximizers or something like that for memory overvolting...

[Donnie27]

Sure but there's no problem getting 1333MHz RAM with a 1.5v rating. Yes, they're overclocking slower RAM. That caused Intel, NVidia Motherboards for AMD and other problems. I saw folks get PC 3200 that was really overclocked PC-2700, same goes for 667 RAM sold as 800MHz DDR2 Gskill PK LOL!

[jaredpace]

how fast can a 3200mhz Penryn do SuperPI 1mb? Faster than 12.75 seconds?

http://www.coolaler.com.tw/coolalerc...E965_X58/3.gif

[safan80]

haha how do you like the IMC now

[JumpingJack]

We all know theres hardly no gain over DDR2 with low speed DDR3, yes triple channel+on die MC will provide heaps of bandwidth but surely Intel are going backwards here (so much for high end DDR3 prices falling)?

Surely you understand better the errors of your statement.... I must assume you are joking.

Triple channel DDR3 at 667 will out perform DDR2 800 dual channel, as will triple channel DDR3 800 out perform DDR2 1067 dual channel.... add to that latency will be cut in half as the IMC comes on die....

Don't judge the DDR3 capabilities with respect to the older platform architecture, and take some time to understand memory technologies.

[Drwho?]

The title of this thread is innacurate, so does the article on the inquirer...

[Hornet331]

how fast can a 3200mhz Penryn do SuperPI 1mb? Faster than 12.75 seconds?

http://www.coolaler.com.tw/coolalerc...E965_X58/3.gif

QX9650 @ 3,6ghz need ~13secs so i guess you need 3,65 to 3,7ghz for that, also Kentsfield needs 4ghz for that time.

[JumpingJack]

how fast can a 3200mhz Penryn do SuperPI 1mb? Faster than 12.75 seconds?

http://www.coolaler.com.tw/coolalerc...E965_X58/3.gif

A yorkie at 3.33 Ghz does SP1M at 14.125 seconds.

Jack

EDIT: So to see where this lies on the SP1M curve, here is the SP1M results for a yorkie from 2.4 to 4.0 GHz in steps of 200 MHz, the red dot is 12.750, where the 3.2 GHz Nehalem completes the run based on Coolalers screen shot.



Here is the raw data/screen shots

[systemviper]

Originally Posted by XS Janus
I hope this will bring us closer to JDEC spec 1.5v for faster DDR3 much faster than it would usually take.
This way vendors will have to REALLY make better ram, NOT just overclock it and call it a done deal!

I couldnt agree more.

+27