Core i7 - Is High VDimm really a Problem?
by Gary Key on October 8, 2008 12:00 PM EST- Posted in
- Memory
Numerous articles and forum posts have been popping up recently about the potential of high VDimm settings damaging or destroying the upcoming i7 processor series. Will high VDimm cause damage? The answer to that question is not so simple actually. Unfortunately, due to the current NDA status, we cannot go into detail about this subject matter but can provide a general brief on it.
Our answer at this time is Yes and No. It sounds like we are straddling the fence but in actuality the correct answer depends on the available BIOS options, BIOS settings, memory selection, and final voltage settings. Intel’s stance is clear on this subject, run VDimm higher than their 1.50V~1.65V guidelines and you will affect the life span of the processor.
Exactly what the impact to the processor will be is dependent upon several factors. Put simply, if you go crazy with VDimm, let’s say around 2.0V~2.2V without additional tuning, then expect to greatly reduce the processor lifespan to a few weeks or maybe days. We have already witnessed several CPUs being damaged or destroyed at the motherboard partners with high VDimm settings, especially those that ran at 2.0V or higher with base settings. By base settings, we mean configuring an i7/X58 platform in the same manner a typical user now sets up a Penryn/X48 DDR3 platform. The rules have changed completely for Intel, just we cannot discuss the playbook at this time (hey, it is frustrating for us also).
Likewise, we have seen high VCore/VDimm test beds operate without a problem for benchmarking purposes (yet still fail with long-term bench testing) provided a multitude of BIOS settings for the core, DIMM, IMC, Uncore, and QPI selections were properly set. The base secret (there are more) is maintaining correct amplitude levels, something we will discuss at product launch. For now, high VDimm is not necessarily the true problem here, but it is the quickest way to damage/destroy an i7 if the rest of the system is not properly tuned.
However, we highly recommend keeping VDimm at or below Intel’s recommendations along with proper BIOS settings for the long-term health of the i7. Upcoming DDR3 products from the major memory suppliers will all support low voltage DDR3-1066~DDR3-1600 operation with fairly aggressive timings. This is key as the retail 920/940 processors will not be memory multiplier locked. In addition, the vast majority of JDEC spec DDR3 memory currently on the market will operate fine on the X58, albeit at slightly higher memory timings in some cases.
The extreme performance modules will also operate correctly, as we have found in testing to date, just at higher timings in order to meet the voltage guideline requirements. Certain performance modules that require 1.8V or higher upon POST will probably not work correctly (POST) unless the board manufacturer steps outside of Intel’s guidelines. In that case, you can replace the memory, reflash the SPD if the supplier allows, or toss in a pair of JDEC spec modules (current 1066 is fine), correctly set the voltages and timings that you require in the BIOS, shutdown, and reboot with the performance modules. Realize, this does place you outside of the guidelines but most users that fall in this category are already outside the lines anyway.
Personally, with the right board, cooling, and BIOS settings, 1.7V~1.8V should be fine (no promises yet) and will allow the upcoming low-voltage, high clock speed DDR3 DIMMS to reach the 1866~2200MHz level. This should satisfy most performance enthusiasts, but probably not the extreme clockers who will try for more. For the rest of us, this platform offers simply amazing bandwidth and latency numbers with tri-channel DDR3 1066 or DDR3 1333. In fact, we think tri-channel DDR3-1333 at 5-5-5-12 timings or DDR3-1500~DDR3-1600 6-6-5-15 settings (1.65V) will provide optimal memory bandwidth, write speeds, and latencies for 95% of the 920/940 users at this point. So, unlike the P45/X38/X48 platforms, having low-speed rated DDR3 is not going to be a hindrance to extracting fantastic performance from a i7/X58 setup.
36 Comments
View All Comments
araczynski - Thursday, October 9, 2008 - link
the sky is falling!! maybe...trust us...D3SI - Thursday, October 9, 2008 - link
Upgrading is pricey. Going to run the e8400 till it dies hehe :)
Calin - Thursday, October 9, 2008 - link
e8400? That's very recent upgrade, you could probably use it for a couple more years.Darkness Flame - Wednesday, October 8, 2008 - link
Think about it; even if you had a QX9750 processor with it's 1600 MHz FSB, you wouldn't notice any significant bandwidth increase over DDR2/3 1066. Even when pushed up to 2000 MHz FSB, DDR2 1200 Mz - DDR3 1333 MHz dual channel sufficed. Going up to DDR3 1600 could get you tighter latencies, and a few more decimals, but not much. The number of people who went higher than that was very low in percentage to the rest.Now, Core i7 uses triple channel RAM, and unless I'm mistaken, Anandtech themselves hinted that there wasn't as much a difference between dual and triple channel DDR3 1066 MHz RAM on this board. In my opinion, that would mean that the difference between dual and triple channel DDR3 1333 MHz RAM would be less. However, that's on the 2.66 GHz model. The 3.2 GHz model's QPI is at 6.4 GT/s, over the 4.8 GT/s on the two lower ones. Calculating it all out, triple channel DDR3 1333 MHz would be the base, with DDR3 1600 MHz being the sweet spot, with probably not much difference about it. With QPI looking to not having many options for overclocking, there wouldn't be a need for anything faster right now. Well, except for the extremist who have bottomless wallets.
Calin - Thursday, October 9, 2008 - link
I think it will be Barcelona all over again - twice the theoretical bandwidth with one and a half the measured bandwidth, for very little gain in performance.Do triple channel RAM makes sense now? Probably not. Did DDR3 made any sense at its introduction? Did DDR2 made sense at its introduction? Not much, and yet we're at triple channel DDR3 when sometime we had single channel PC-100 SD-RAM.
This is proofing for the future, preparing a platform (hopefully) for the long run.
Darkness Flame - Wednesday, October 8, 2008 - link
corrections in order ...Mz -> MHz
(before "With QPI") about -> above
I should proofread before posting ...
npp - Wednesday, October 8, 2008 - link
At least in benchmarks, the gains from the IMC are noticeable, I saw something like 19GB/s bandwith measured on xtemesystems, which is very high indeed, much higher than what you'd get with 450Mhz FSB... The open question is, however - which is the app hte really needs that much...SuckRaven - Wednesday, October 8, 2008 - link
I thought it was JEDEC, not JDEC...'Joint Electron Device Engineering Council'
anyways....
Probably going to wait for the die shrink of i7 (Nehalem), or at the very least, a couple of revisions before I jump on the bandwagon. I am loooooooong due for upgrading my hardware.
Ooooh goody! A new build is always fun.
Butterbean - Wednesday, October 8, 2008 - link
Lately - the main impression I am getting about Nehalem is to just leave it alone a good while.BansheeX - Wednesday, October 8, 2008 - link
This whole 3dmark crowd of SLI, overlock kiddies with 10000w psus who want to pump more volts into their ram to solve their pi benchmarks a millisecond faster has to end. The amount of time and money these people spend on overclocking is way higher than they would have spent purchasing a faster computer to begin with. They've completely taken over forums with their self-inflicted problems, and it's a tech support nightmare. People wonder why interest in PC gaming has lost immeasurable ground to consoles, this is one. Stop dicking with your ram and play your games.