X220 memory speed

[yha]

Hello,

As is the case with my first post, I did search the forums before posting this, but I couldn't find the answer I'm looking for, so here we go.

I have a ThinkPad X220 with an i7-2620M CPU and 4GB of RAM. I would like to upgrade to 16GB which I know is possible from the many people who successfully did it. What I want to know is which speed of DDR3 memory I should be getting because apparently there are multiple factors that affect this.

My firmware is at version 1.24, so it's before Lenovo limited RAM speed from 1886 MHz to 1333 MHz, BUT, according to my CPU specs, memory type is "DDR3 1066/1333". Does that mean that even if I don't upgrade the BIOS and use 1886 MHz memory, the memory will still operate at 1333 MHz because of the CPU?

And assuming I upgrade my official Lenovo firmware to the latest version, does the modified BIOS available at McDonnell Tech really "re-enable" 1866 MHz memory speeds despite the fact that Intel says the CPU works (only?) with DDR3 1066/1333? Like how does that work? My understanding is that the motherboard (chipset/memory controller), CPU, and RAM must have common support for a particular speed to operate or synchronize at (in my case, the "highest common denominator" or highest frequency supported by ALL components would be 1333 MHz).

I know other people say they checked with CPU-Z and it reports 1866 MHz, but maybe that's only the reported speed, and the actual speed is limited by the component with the slowest supported frequency? (similar to how fake SD cards report much higher capacity than there actually is).

I hope all of this makes sense. In the end, I want to squeeze as much performance as I can from the RAM I will buy (in fact, I want to max out everything to the tiniest of details). I would also like to upgrade to the latest Lenovo BIOS (assuming it doesn't affect my RAM speed as explained above) because I do believe these upgrades provide better stability to the system overall (there are 19 upgrades between my version of 1.24 and the latest of 1.46, and the change log suggests many of them are important updates or new features).

So to summarize, should I be getting 1333 or 1886 sticks? And should I keep my current BIOS, upgrade it, or upgrade it + flash modded BIOS?

Thanks!

[RealBlackStuff]

Install McDonnell's BIOS 1.46 plus any DDR3 (non-L) 1333 or 1866 RAM you can find.
The difference between RAM-speeds is so minimal, it'd hardly worth even describing it.
Look here: https://techbuyersguru.com/does-ram-spe ... 2400-games
And here: https://www.anandtech.com/show/7364/mem ... -haswell/8

Do NOT update to Lenovo's newest BIOS, McD's BIOS is much more advanced.

[31336]

Should I be getting 1333 or 1886 sticks? And should I keep my current BIOS, upgrade it, or upgrade it + flash modded BIOS?

From my experiments, the best it can do is 1866 MHz. I'm using Kingston HyperX Impact HX321LS11IB2K2/16, supposed to do 10-11-12-? @ 1866 MHz and 11-12-13-? @ 2133 MHz. It runs at 10-11-12-30, command rate 1T.

I use ValdikSS's BIOS 1.43 as it had a newer Intel VBIOS (which was reverted back to the original one in the following versions due to issues with Linux). Any later BIOS version was just Spectre mitigations, outdated by now anyway, and they are likely to make the laptop slower.

There're two other parts that could potentially be even better:

• Crucial Ballistix Sport BLS8G3N18AES4 [2×: BLS2K8G3N18AES4] supports 10-10-10-30.
• Kingston HyperX HX321LS11IBK2/16 supports 10-10-10-? @ 1866 MHz and 11-11-11-? @ 2133 MHz.

I think both are practically unobtainable now: in particular, the Kingston part is an earlier revision of the one I have. But you could definitely check out one of these if you run into them somehow. I tested the first one for a while but don't have it anymore, and I don't remember what the timings were exactly, but as far as I recall 10-10-10 was not possible.

I also tried:

• G.Skill Ripjaws F3-1866C10S-8GRSL [2×: F3-1866C10D-16GRSL] only worked at 1600 MHz for me: looked as if it was failing memory training at 1866 MHz, which manifested itself with a noticeably longer POST at initial boot. Possibly it could still be made to run at full speed with a tool like Thaiphoon Burner but I didn't try.

Other comparable parts I know exist:

• Corsair Vengeance [2×: CMSX16GX3M2B1866C10] 10-10-10-32
• Patriot Viper PV38G186LC0S [2×: PV316G186LC0SK] 10-10-10-32
• Kingston HyperX Impact HX318LS11IB/8 [2×: HX318LS11IBK2/16] 11-11-11-?

The first two are even more rare. The third one should be available but is practically the same price as the faster Kingston counterpart I think.

So, by all means get 1866 MHz (PC3-14900) memory or better. Performance for different modules running at 1866 MHz will be comparable but be aware that some of them might not work at full speed by default, and since timings cannot be adjusted by hand even in the modified BIOS, it'd be best to check compatibility at time of purchase.

[atagunov]

..

@31336: thx a lot for the detailed write-up; I guess the most interesting question now is how much difference can be measured between the creme de crop RAM you enjoyed and cheaper varieties like random 1333 or 1067.. many people say there is virtually none; I'm wondering if there are other people who have noticed any and how much

[yha]

First of all, thank you so much for the detailed answer. I guess I'll go with 1866MHz memory then.

Secondly, does that mean that the CPU datasheet is just marketing crap when it says "DDR3 1066/1333"? Perhaps it's the "preferred" frequency to work with but then you could use any other speed if you wish? (kinda like "native resolutions" on screens: it doesn't mean the screen can't do scaling and work with other resolutions).

Thanks once again.

[ThreeIguanasPerSecond]

it's not 'marketing crap', that's what the CPUs have been tested and certified for. you have to also consider that not all CPUs are the same in terms of performance, since any of them can perform differently, even if they are identical models (e.g: my 2500K overclocks better than my friend's 2500K). now, you should be wondering how this plays with your memory modules.

the memory controller integrated into Sandy Bridge isn't guaranteed to support DDR3-1600 modules, as there's always a 'bad' batch that underperforms and brings the rest of the silicon down. because of that, they had to downgrade to DDR3-1333 or DDR3-1066 modules that just happened to work with every known Sandy Bridge-based CPU, no matter how bad their memory controllers are. this is why they are stated to run at DDR3-1066 or 1333, not DDR3-1600 or 1866 or 2133 as you would expect.

if Intel tested out DDR3-1600 and faster modules for every single CPU they had, it would result in a logistics nightmare that would pump out a mess of CPU sub-models just to indicate that their memory controllers' support DDR3-1600 and/or 1866 and/or 2133. it's not a realistic or feasible approach by any stretch of your imagination, it's as if you were to take some silicon from a wafer and slap on a suffix on it to indicate that it can operate at the same clock frequency of other silicon but by a difference of -0.001v in voltage. do you see Intel doing what i'm describing? no, they aren't. it isn't worth it to make a segment just for these special CPUs that are better than the rest by simply having the ability to operate on negligible voltage differences at the same clock frequency. likewise for the CPUs that operate on a slightly higher voltage at the same clock frequency, they're not going to label it with some funky suffix to separate it from the rest. makes sense, doesn't it?

[31336]

I guess the most interesting question now is how much difference can be measured between the creme de crop RAM you enjoyed and cheaper varieties like random 1333 or 1067.. many people say there is virtually none; I'm wondering if there are other people who have noticed any and how much

I'd expect the performance difference to be nontrivial, mainly because the system memory is also used by the integrated GPU. Would not matter that much if the GPU had its own memory.

I ran some tests before with standard 1067 and 1333 MHz memory I had lying around and performance was noticeably worse but that would mostly be because with only a single 2 GB module and two 1 GB ones I had, I could only have 3 GB installed at any given time. Increasing memory capacity up to at least 8/16 GB would have improved things a lot, regardless of the speed.

So if you already had 16 (2×8) GB of slower memory or could get it for free, it's probably not worth to upgrade. If however you are only looking to buy it now, then I'm not sure whether it would make sense in terms of price to buy anything but the best, or if such slower memory is even still available (as the faster one is backward-compatible). For example, Micron's kit of 2 low-end/standard 1866 MHz memory modules, CT2K102464BF186D, is $86 on Amazon right now, while the slower 1600 MHz counterpart, CT2K102464BF160B is $90, and the 2133 MHz Kingston HyperX I mentioned before is also $90 on NewEgg. (Of course, maybe you could also get some of these more cheaply second-hand, depending on where you are located.)

As for the datasheet specs, everything they put in there constitutes a commitment on Intel's part, so they're careful not to include anything there other than the bare minimum required. I wouldn't really consider it BS, more like "anything above it might work but if it doesn't, you're on your own." There's a disclaimer in the datasheet to that effect. In practice, much higher speeds should usually work, with i7 CPUs at least. It's possible higher-speed memory wasn't even available by the time the datasheet was being written, and once the product is EOL they aren't going to update it just because it makes no sense for them as a business decision, it would also open them to potentially new liabilities. As an aside, if you want a real example of Intel's BS, you don't have to look far: for example, they now sell 65W TDP CPUs that can draw 300W.

Edit: Benchmark results for Kingston HyperX HX321LS11IB2K2/16 (2× KHX2133C11S3L-8G):


If anyone wants to try it too we can compare.