Gadgetoid

gadg-et-oid [gaj-it-oid]

-adjective

1. having the characteristics or form of a gadget;
resembling a mechanical contrivance or device.

Alienware M15 R3 SSD Upgrade – Crucial Mystery Edition

Upon finding out the Alienware m15 r3 included two 2280 NVMe SSD bays plus a smaller 2230 bay I figured it would be an excellent opportunity to get up to speed with the latest and greatest in absurdly-fast-storage-devices.

What I didn’t expect was to happen upon a mystery. In this article I’ll be walking you through my experience of the Alienware m15 r3 and Crucials “P5” CT1000P5SSD8 SSD.

To spoil it up front, it’s not exactly the most thrilling of adventures. I found during my tests that the P5 showed a weird performance profile that – while on a par with some of the numbers Crucial reported – appeared to exceed their own tests by quite some margin while insisting upon reporting sequential read speeds that fell far, far short of my expectations.

This is particularly curious because sequential reads are something that SSDs are particularly good at.

Fitting the SSD

First thing’s first, you can’t test an SSD without sticking it into your computer. Fortunately the process is incredibly simple with the m15 r3. All of the screws required to access the bottom of the computer are cross-head and all but two of them are captive within the base. The two rear screws must be retrieved fully and set aside, but the remaining screws stay attached to the bottom panel. This is handy, since during the course of my testing I have removed and replaced the bottom panel of the Alienware m15 r3 more times than I’d care to count.

While unscrewing the base, the captive screws also cause the central plastic clips to release automatically – effectively prying the panel off – which is a really nice touch. Despite the SSDs being the only thing that’s supposed to be user-serviceable (lots of m15 owners will re-apply thermal paste), Alienware have gone to great lengths to make opening this machine straightforward.

Further clips release along the front edge of the baseplate, and it lifts off cleanly. Inside you’ll find whatever SSDs are fit in your stock build and you might also notice the distinct lack of spare heatshields for the unpopulated SSDs. This is something of a bugbear I have with Alienware systems. The stock Micron drive shipped in the m15 (made by Crucial’s parent company) isn’t exactly a stellar performer and I’d expect many customers to buy the minimum storage configuration and install two high-performance 1Tb or 2Tb SSDs. On the plus side they do- at least- have the tiny screws required to secure the SSDs, the omission of these screws is a complaint I’ve seen with other systems.

Out comes the stock Micron drive and in goes the uh… Micron drive ?
Upgrading the @Alienware m15 r3 with a @CrucialMemory P5 SSD to see how it chooches. pic.twitter.com/Oqz4VPkDbd

— Phil Howard (@Gadgetoid) August 5, 2020

The SSDs in the m15 r3 slot into a daughter board between the main board and the battery, this is an improvement over the nestled-in-the-battery style that the m15 r2 used.

Stock 512GB Micron SSD vs 1Tb Micron Crucial P5.
I still have no explanation for the P5s weirdly slow read speed.
Working on it! ? pic.twitter.com/ME4IDfZRLH

— Phil Howard (@Gadgetoid) August 20, 2020

The Crucial P5 SSD handily destroys the stock Micron disk in the m15 r3. It seems a tragedy that a gaming laptop specced-to-the-hilt should have such a woefully underperforming disk but it seems almost a given that these drives will be replaced with aftermarket upgrades. I guess you’ve got to leave something for the end user to upgrade.

You will, however, notice that the sequential read speed of the Crucial P5 (pictured on the left) falls short of the write speeds quite considerably, and additionally doesn’t punch too high above the Micron disk. This was typical of every single benchmark I ran using the P5 SSD in the Alienware m15 r3. But *only* in the m15 r3.

Tests

Honestly after conducting these tests I’m sick of unscrewing the bottom of the Alienware laptops. They’re surely among the most easy-to-get-inside computers on the market, but tackling those 8 screws really wears on you quickly.

Crucial P5 1TB

With repeated tests Crucial’s 1TB P5 SSD benched with a sensible read speed in the Alienware m15 r2 (my workstation) but at almost half the expected value in the m15 r3. Switching betweeen AHCI and RAID, system disk or auxiliary disk and any other permutation of tweaks failed to bring the m15 r3 tests in line with the r2. Truly baffling.

Above: Crucial P5 1TB, tested in Alienware m15 r2

Above: Crucial P5 1TB, tested in Alienware m15 r3

Even after coaxing Intel’s Rapid Storage Technology drivers onto my system by manually installing the device, and then switching the disks into RAID mode and rebooting didn’t magically recover all of the mysterious missing performance. However it did seem to make somewhat of an impact- raising scores from around 1.6GB/s to 2.2GB/s.

I prefer running in AHCI, however, since the RAID controller sits between the SSDs and the system and obfuscates useful benchmarking info such as disk temperature. Indeed in RAID mode neither HWINFO64 nor Crucial Storage Excutive could tell me the temperature of the SSDs, so I couldn’t tell if a drop in performance was due to wind direction, stray cosmic rays or overheating.

Crucial P5 2TB

Since I wanted to eliminate the issue being with the one specific 1TB P5 SSD I was testing, and additionally wanted to use the same disk as Crucial’s in-house test, I requested a 2TB P5 to replicate my experiments.

I tested this disk fresh out of box in both the Alienware m15 r2 and r3 and received very similar results to the 1TB part.

Both CrystalDiskMark and AJA System Test showed the same stonking write speeds alongside very similar and baffling slow read speeds. In the case of AJA System Test the write test is performed first and as far as I can tell the lower reported read speed (versus CrystalDiskMark) is a result of the drive getting pretty toasty and throttling to keep its temperature under control.

Above: Crucial P5 2TB, fresh out of the box in the Alienware m15 r3

Above: Crucial P5 2TB, fresh out of the box tested with AJA System Test

Control Tests

Trying a different drive…

To verify I wasn’t seeing some general issue with the m15 r3 I grabbed one of my Samsung 970 Evo drives out of the Alienware m15 r2 and tested it in the r3. The result was much for muchness- these drives were bought at the same time with one used as a system disk and one used as a gaming disk. Their usage will have varied, but the results are within a reasonable margin for error from each other. It’s clearly not something wrong specifically with the Alienware m15 r3, but rather with its interactions with other disks.

Above: Samsung 970 Evo tested in Alienware m15 r2

Above: Samsung 970 Evo tested in Alienware m15 r3

Linux…

It also occurred to me to try Linux (Ubuntu 20.04 to be specific) on the m15 r3 so that I had a system free of Windows driver weirdness that could confirm if the SSD weirdness was due either to firmware/controller issues (hardware) or Windows driver issues (software). Not entirely to my surprise I found Linux giving me read speeds of around 3GB/s sustained with a few drops just under. Even testing 1000 reads of 512GB. This clearly implicates some driver level issue that somehow only affects a combination of Windows 10, the Alienware m15 r3 and the Crucial P5 series SSDs.

Of course these tests are not especially comparable to CrystalDiskMark results, so it’s quite possible I’m seeing a false positive correction of the disk performance and it’s some system or firmware level issue with the m15 r3.

Either way there’s clearly some obtuse issue with compatibility here, though I can’t seem to find where to point the finger.

The stock SSD…

The stock Micron SSD that shipped with the Alienware m15 r3 is pretty tragic compared to its high-performance contemporaries, but isn’t all bad. It’s clear that whatever issue plagues the P5 in this particular configuration, it doesn’t affect Micron’s less fancy drives.

Above: Stock Micron drive tested in Alienware m15 r3

Hot, hot, hot

If you plan to use the Crucial P5 in a laptop you will want to make sure you get a full complement of heat spreaders for your drives. Running the P5 without any kind of heat spreader quite considerably impacted sustained performance. I noticed a significant drop in speed when copying 200GB of Xbox Game Pass games from the 1TB disk over to the 2TB one. The 1TB – with heat spreader – remained around 65 degrees, while the 2TB without tended to hit 75 very quickly. CrystalDiskMark tests immediately after this bulk copy and before allowing the disk to fully cool revealed throughput of around 1GB/sec read and 880MB/sec write. Ooof.

Alienware are still shipping laptops with heat spreaders installed only on pre-installed drives, I noticed this both in my r2 and r3 which both shipped with one 512GB stock SSD that I immediately replaced and stowed away for emergencies. This stock drive came with a heat spreader installed, but additional spreaders for the extra bays (1 additional bay in the r2 and two additional bays in the r3) are nowhere to be found.

To be fair most workloads don’t necessitate copying 200GB of data from one disk to another, so this may not be a dealbreaker but you really should grab those heat spreaders to get the best out of your drives. If you’re buying for a desktop system then a heatsink should suffice- really you’re just looking for something to give the disk enough thermal mass to survive short bursts of sustained use that might occur when loading a game, rendering a video, copying files or running “npm install”.

Some of my more aggressive benchmarks (not on these SSDs mind) actually required me to take the bottom off the laptop and point a fan directly at the SSD. With games getting bigger and the “next-gen” titles expecting near continuous, high performance streaming of assets from disk, active SSD cooling could be an important addition to the gaming laptop space.

Crucial’s Magic Momentum Memory Caching

I’d like to say good things about Crucial’s Momentum driver for their SSDs. On paper it’s a good idea to use remaining system memory as a memory buffer for SSDs, and in a battery-backed system like a laptop there is, in theory, little chance of data loss. In practise, however, enabling Momentum Cache threw me into a nigh-on inescapable Windows 10 repair loop. It was only upon following these bizarre instructions specified by Ron on Microsoft Answers that I was able to boot the system and finally see Momentum in action. Not a good start.

Things took a turn for the better when actually benchmarking the memory-boosted storage, however. In my case sequential read/write speeds shot up to 8GB/s with dramatic performance increass all across the board.

With 32GB RAM in the Alienware m15 r3 it was very tempting to leave this caching enabled, but having read that it potentially causes issues with anticheat software (ugh) and being just a smidge afraid of spontaneous data loss I lost my bottle and opted to leave it disabled.

Since RAM caching only affects “hot” data that’s being read/written repeatedly there are few practical implications for this performance increase. Since CrystalDiskMark creates and reads files in a short period of time, it makes sense that they would stay in RAM and give an unnaturally positive outlook on caching. It may be more useful for increasing the endurance of your SSDs- something that has thus far not been a problem for me- than increasing performance. Tasks such as compiling software projects may see more immediate benefit from this caching and if I can summon the courage to enable Momentum Caching again I’ve certainly got some significant software projects that would be worth A/B testing the compile times of.

Conclusion

While this mystery remains unsolved you would be forgiven for avoiding Crucial P5 series SSDs for ugprading your Alienware m15 r3, but in spite of these findings the day to day performance of the SSDs is… fine. I don’t notice any practical difference between performance on my r2 with Samsung 970 Evo SSDs and the r3 with Crucial P5 SSDs.

The P5 is clearly Crucial’s answer to the 970 Evo and Evo Plus but, with both the price and performance keeping pace, it will be difficult to convince the average aftermarket upgrader to forego the “nobody got fired for buying IBM” choice of Samsung.

Until I’ve run these SSDs into the ground I can’t stick my neck out and be the one to say “It’s fine to pick Crucial” but Micron SSDs seem to find their way into plenty of OEM builds and while SSD failures are inevitable there doesn’t seem to be the crises of crashes that gamer Samsung proponents would have you believe plague other manufacturers. So- I’ll happily say “pick these with confidence” and with any luck you’ll have forgotten who said that if they ever fail.

On the subject of the weird benchmark figures… I’ve lodged an ongoing tech support query with the right person at Dell and it’s being looked into.

Saturday, September 12th, 2020, Personal Computing.