Would be nice to have a truly fast laptop and it seems that the most common bottle neck is the storage and specifically the storage latency and random access speed.
Common SSD benckmarks give sequential speeds and are not really a reliable indicator of actual performance in a mixed workload
If I had a desktop, I would buy an enterprise u.3 drive with good random io speeds like the Solidigm D7-P5810 Wendel reviewed a while back.
The hardware
Asus ROG Zephyrus M16 (2022)
12th Gen IntelĀ® Coreā¢ i9-12900H
48GB RAM
4TB Kingston Fury m.2 Gen4 SSD - Should be fast, was dissapointed
It is possible to fit two m.2 drives in the machine with one running at PCIe 4.0 and one 3.0
Running Windows 11 with Bitlocker enabled
Ideas
Is there a way to make use of Optane in such a system? The 380GB m.2 optane is sadly 11cm long so it wont fit in the primary ssd slot but could maybe be made to fit if one would gently drill out the mounting screw for the secondary slot.
Do SSD Latency benchmarks exist? Are there enterprise drives one could use in a laptop?
There was some talk of a storage tiering software that could possibly make use of a smaller Optane module that would fit in the system and speed things up.
Are there any modern m.2 ssd drives that are actually fast with random i/o?
CrystalDiskMark in its settings has a āNVMe modeā that will measure RND4K latency in microseconds. Ditto for KDiskMark on linux. What results do you get from the āReal World Performance +Mixā profile in these?
There are no Gen4 m.2 Optane drives, so donāt worry about the primary slot if thatās your goal.
If it were me, I would use the P1600X (118GB) 2280 as a lean boot drive and then stick any cheap 1-2TB Gen4 m.2 into the primary slot (doesnāt even need DRAM) for bulk storage.
This would probably lead to the best latency on an older 6-10th gen system, but Iām not sure how much it will help on the current architecture. Maxing memory bandwidth/speed might yield better results.
Beware the unintended consequences! Youāre probably not really thinking about it now, but performance has a direct correlation to power consumption - at least in regards to storage.
Check out the power consumption of the fastest nvme drives (by IOPs, or seq bandwidth). Youāll find that these are higher than likely the drive you have in your laptop.
Getting the fastest storage for a desktop system that is permanently tied to the power grid is a reasonable proposal.
In a laptop this will lead directly to shortened battery life. Is this really what you want?
The P1600X was my first thought as well until I saw the power consumption. I found an 800p on eBay instead. The capacity is the same. The performance is worse, but the latency and redundancy is still better than NAND. And itās power consumption characteristics are suitable for mobile computing.
I feel like people are so out of touch about how fast a storage device should be, so many things can effect how well it works. if your using winbows and have a million things loaded into memory (application in the taskbar/background) no NVME/SSDās are going to preform āultra fastā as there available bandwidth is already being utilized by other things. go and get a laptop from 2012-2017 with a HDD (spinning rust) and then you can feel the difference with how āslowā your current performance solid state device isā¦ just my thoughts
Since the laptop I daily drive happens to be a gaming laptop battery life or lack there of is a trade off I am willing to live with.
The issue is that modern consumer drives offer excellent sequential performance but are not great with random workloads.
The smaller optane drives are an ok alternative, but the small space is not something I am able to live with even for a boot drive. Managing the space and hardlinking system folders to another drive is such a hasstle that it would probably be too much.
Maybe a solution like a secondary optane drive of 118GB and software like primo cache could be a good fit?
My quick 2 cents on Primocacheā¦ Primocache will limit your IOPS to substantially less than a raw NVME SSD can achieve. But, for general large file access and traversal in database work Primocache can speed things up (my setup uses 256GB of RAM). Horses for coursesā¦ My fastest NVME SSD āvolumeā for IOPS and low latency is a striped pair of fast 2TB sticks.
bitlocker?
If you donāt encrypt (and decrypt) the files as you write/read them, maybe you would get better results.
Optane is technically the lowest latency m.2. Even a first-batch 32GB will eat a SSD up for breakfast, especially low4k. Though their density is incredibly low.
And the reason they canāt push low4k random speeds is precisely that ā capacity. Dense NAND and low latency donāt mix.
Kind of like how high capacity RAM usually lowers maximum clock (as do buffering/registering them).
It is true that I lose performance with bitlocker. It is how ever a reasonable trade off for information security on a laptop with information stored on it that is worth protecting.
The majority of consumer Optanesāunfortunatelyādo not support OPAL. The P4800X and P4801X series are the only ones which Intel claims to do so. (I own SSDs from that series and I can confirm.) Now I said āmajorityā which means there might be one consumer Optane series which supports OPAL: the H20. The drives have a PSID on the sticker, which would only be present if some OPAL-like interface were exposed.
Straight from the horseās mouth (Microsoft): āTypically, thereās a small performance overhead, often in single-digit percentages, which is relative to the throughput of the storage operations on which it needs to operate.ā
https://www.reddit.com/r/hardware/comments/16nbq7o/nvme_bitlocker_performance_degradation_post_your/ āThe 4kQ1T1 was benchmarking really well on this SSD (83MB/s read, 355MB/s write), but this reduced to around 65MB/s read and 156MB/s write post-encryption, at 90% full. I have found that when not encrypted, this SSD benchmarks identically whether empty or at 95% full. The post-encryption result for this benchmark is very similar to my 970 Evo Plus. It seems that there is some kind of bottleneck for these 4kQ1T1 readings.ā
As usual, youād have to test on your own hardware under your usage scenarios to get the most accurate data.
@mtchetch, youāre either going to have to find your compromise point between Bitlocker tanking performance and NAND having five to ten times higher latency, or self-experiment with H20.
Thanks for the feedback. I have looked into hardware encryption using OPAL. The Kingston drive I have does not support it so I have a Samsung 990 pro on the way with support for HW encryption. I will see what happens with a 118GB Optane and primocache if it makes any positive difference or not. Based on the comments I am not expecting much and have other uses for the optane if it does not perform.
The main cause for going down this particular rabbit hole was finally diagnozed and solved. Everything felt slow because of a āfeatureā of the Kingston Fury NVMe drive series. Stale data became extremely slow to read. When reading old data it caused drive latency to go to 200ms+ and read speeds were at around 12-19MB/s.
When running benchmarks I found that they wrote new data that performed at the expected level. This made me believe that the drive is fine and since the change was gradual it was not noticed and was only sanity checked with the occasional crystal disk mark run.
Decrypting the drive (un-bitlockering) took 26 hours to complete.
Having re-wrote all the data on the drive the performance returned to a somewhat normal level. I got the Samsung 990 Pro drive and that performed even better, so that issue got solved. Also spent some quality time with Kingston support engineers and was promised a firmware fix for the performance regression on stale data.
On a side note optane + primo cache does not do anything usefull with modern NVMe drives so that was not adopted into service.
Lessons learned:
Consumer SSD drives are bad, mkay
There is such a thing as stale data on SSD-drives and that can perform extremely poorly
if you feel like going off the deep end on the stale data issue:
It still surprises me how little this issue comes up in reviews and general conversation considering seemingly all consumer SSDs suffer from it (with a few exceptions).
Also Iām skeptical Kingston will be able to develop a firmware update that will fix the charge decay on the SSD, doing so would burn P/E cycles at an accelerated rate which I assume is part of the reason it wasnāt implemented to begin with.
I agree with you. It is unlikely that the manufacturers will fix the issue for the reasons you outlined. They had to say something to get me to not rma the drive back to them and so here we areā¦
The tools for rewriting data to prevent the issue should be talked about and be easily findable. I had to spend some time on the issue to actually find out what is going on and find a fix.
In my case most of the problematic data was 1-2 years old so doing a full drive refresh every year would not really cause premature wear on the device in a meaningfull way.
