UPDATE on SSDs and Super RAID 5 - easy PCIe 5.0x4 OCuLink port
After proving the Titan 18 HX AI A2XWJG is perfectly stable running 128GB of RAM, and having been using that amount of RAM so far with no problems, it is time to test SSDs and RAID systems.
After contacting MSI regarding SSDs, OEM thermal pads and SSD covers, I got the following information:
"We have mainly tested the below SSD for you refer, but the specific support list is based on the inventory of service centers or the store.
PCIe GEN4x4 NVMe,M.2-2280 M-KEY,2TB,,SAMSUNG/MZVL22T0HDLB-00BT7
PCIe GEN5x4 w/DRAM NVMe,M.2 2280 M-KEY,2TB,_,SAMSUNG/MZVLC2T0HBLD-00BT7
PCIe GEN4x4 w/o DRAM NVMe,M.2-2280 M-KEY,1TB,,SAMSUNG/MZVL81T0HELB-00BTW
PCIe GEN4x4 w/o DRAM NVMe,M.2-2280 M-KEY,2TB,,MICRON/MTFDKBA2T0QGN-1BN1AABYY
PCIe GEN4x4 w/o DRAM NVMe,M.2-2280 M-KEY,1TB,,MICRON/MTFDKBA1T0QGN-1BN1AABYY
PCIe GEN4x4 NVMe,M.2-2280 M-KEY,1TB,,SAMSUNG/MZVL21T0HDLU-00BT7
PCIe GEN5x4 w/o DRAM NVMe,M.2 2280 M-KEY,2TB,_,PHISON/ESR02TBTLCA3-E1J-2MS
To avoid compatibility problems, it is recommended to install the modules (brand and specification) which was tested by msi.
We are sorry that we haven't tested the 4TB Samsung 990 Pro SSDs. From my personal point of view (Personal opinion only, not official), you can also try to use the SSD which the store announced compatibility with your laptop. But we still need to explain that this material has not been MSI tested with your laptop, and we do not guarantee its stability, compatibility and performance."
Regarding the purpose of the black SSD covers:
"The purpose of the black SSD covers is to shield other components from interfering with the SSD signal."
Regarding gen5 slot cooling pipe and OEM thermal pads thickness:
(sent by MSI)
Btw, this OEM part can be ordered with OEM pads included from the MSI factory in China if you want (e.g. UK):
"E31-0007050-B22 - 45 EUR net
Shipment to UK - 35 EUR net
Total (incl. VAT): 96 EUR"
So, when it comes to SSDs, we all know these systems are compatible with many SSDs on the market, even though MSI only recommends what they tested in lab (as expected). But I have a 12-year-old GT70 2OC (GTX 780M) still running the same Win8.1 on a RAID0 system of 2x Plextor SSDs, and my latest Titan GT77 HX 13V (RTX 4090 - now sold) always had a RAID0 system of 2x 2TB Samsung 990 Pro that worked just fine.
Looking online, you quickly find several people saying that with today gen5 SSD speeds, any RAID0 system offers no advantage - TLDR this was the conclusion of my tests (more on those later). For these, I bought a 4TB Samsung 9100 Pro gen5 to replace my laptop 2TB OEM Samsung MZVLC2T0HBLD-00BT7, and used an extra 4TB gen4 Samsung 990 Pro to create different RAID0 configurations and test/compare against the use of a single gen5 SSD. In the process, I also replaced OEM thermal pads on the gen5 cooling pipe/plate with
0.5mm GELID GP-Ultimate pads, and also added
1.5mm copper heatsinks with 1mm GELID GP-Ultimate pads to my gen4-slot SSDs - all with the intention to measure SSD temperatures and check if new pads reduced heat.
This was the process to exchange original OEM pads (left) of the gen 5 slot cooling pipe/plate to new GELID pads (right):
Overall, this is how my system currently looks with laptop (back cover) closed:
- The copper heatsinks on the gen4-slot SSDs are now working in the same way as the black SSD covers, which is why I do not use the covers anymore.
- Yes you could add more SSDs - but remember that each SSD you are adding to the laptop is a heat source of 30°C (idle) which if not properly cooled can add up to 70°C. And I have 3 heat generators already in there in such a small confined space - just remember that every time you want to add more SSDs. The final system has OS only on the gen5 slot (4TB Samsung 9100 Pro) with the other gen4 slots having a gen4 4TB Samsung 990 Pro and the gen5 2TB OEM Samsung SSD, which will only be used for extra storage (e.g. big AI datasets) and therefore will only be used now and then - minimizing heat generation inside the laptop.
- Also notice that I prefer to use the two gen4 slots high above on the left for extra SSDs, instead of the lower one on the right, because the laptop back cover has more holes higher up as well. In addition, the lower-right gen4 slot is closer to the gen5 slot which technically reaches higher temperatures just because it reaches higher speeds - so there's a lot of temperature there added to the heat generated by the RAM kits.
- For the 4th gen4 slot, I might buy an
M.2 NVMe to OCulink SFF-8612 adaptor, which in essence gives the laptop a PCIe 4.0x4 OCulink port. Just as you can buy from MSI an extra gen5 cooling pipe/plate, you can also buy a new extra laptop back cover, which I intend to play around with to make a precision hole for the OCulink SFF-8612 port.
- Even better, these laptops give you something extremely uncommon in the market - the possibility to use the gen5 M.2 NVMe slot to connect an OCCulink adaptor which would give you access to a PCIe 5.0x4 lane for eGPUs - this would be the fastest eGPU connection ever - same PCIe 5.0x4 lane that internal dGPUs use to connect to CPU as well. You could then create a RAID0 system of 2 or even 3 gen 4 SSDs for the OS and not lose too much speed compared to a single gen5 SSD - synthetic sequential speeds would be even higher. It's a shame the latest eGPUs are only being made with Thunderbolt 5 connection (which is limited to PCIe 4.0x4 = 64Gbps) - with this laptop gen5 slot we could have full external GPU performance with absolutely no loss..
Given the hardware above, I then proceeded to test SSD speeds and temperatures. For the tests, I used HWiNFO (temperatures), CrystalDiskMark and PCMark 10 Full System Drive Benchmark (FSBD). If you google you'll see that CrystalDiskMark gives only synthetic read/write speeds under specific workloads which do not represent real-world, day-to-day SSD speeds - and from these we should pay more attention to the random speeds (instead of sequential speeds). This is why PCMark 10 FSBD was made, as it emulates those real-world workloads, and it better shows real-world, day-to-day SSD performance. For all tests I used same software versions, and always fresh-installed new OS with latest drivers (same drivers and programs among different OSs + MSHybrid Graphics Mode + Extreme Performance) when the case - idea being to try and keep everything standard - I even fixed my IETS GT600 v2 laptop cooler to a test-standard of 800 RPM (which is the speed I always use everyday at home - not loud). In addition, I run tests with nothing else running in parallel (clean system tray with only MSI Centre icon) - no Real-time protection (Windows security), no internet on, nothing opened (a bunch of Windows Services I never use were also always disabled with every fresh OS install). When the test is about SSD performance, I also just reference the InstallAware Pre-Compiled Runtimes Tool time (if I have it) for the people who are interested in it - it's a benchmark that does not focus on SSD performance but instead the combination of CPU+RAM(+SSD).
Below you can find examples of the battery of tests I ran:
1. Temperature/speed test: 4TB Samsung 9100 Pro (gen5) in gen5 slot with OEM pads (no OS)
VS 2TB OEM Samsung (gen5) in gen5 slot with OEM pads (with OS)
I always started and ended HWiNFO with the beginning and end of the CystalDiskMark (All) test, so the average/max temperatures you see above are a good indication of average temperature throughout the entire bench test.
- Average temperatures are pretty much the same for both SSDs even though the 2TB OEM Samsung can reach slightly higher max temperatures than the 4TB 9100 Pro. Also remember the OEM Samsung had the OS, but I always minimized everything during test runs, and run only CrystalDiskMark and HWiNFO at same time.
- Write and test speeds are on par as well, with the 9100 Pro just slightly edging the OEM Samsung on the more important SSEQ1M Q1T1 and RND4K Q1T1 tests.
- All in all, I must say the 2TB OEM Samsung MZVLC2T0HBLD-00BT7 is actually an excellent SSD. Throughout the tests was always on par with the latest, more expensive Samsung SSDs. If you don't care about extra storage, just stick to this gen5 OEM SSD and it will be pretty much one of the best speed/temperature performance SSDs - as we'll see even better than any RAID0 system as well.
2. Temperature test: OEM Samsung (gen5) in gen4 slot no copper heatsink (with OS)
VS OEM Samsung (gen5) in gen4 slot + copper heatsink (with OS)
When I reached this phase, I noticed I had to start standardising the time between bench runs as well. I started with 4 tests without the copper heatsink, but the time between these tests was not measured and I'm pretty sure they were higher than 2 minutes to let SSD temperature drop after each run. On the other side, the 5 tests above ran with heatsink have exactly a time of 2 minutes of wait time between them to let SSD temperature drop, but this way I keep wait times between runs consistent from now on. This is important because the starting (minimum) temperatures of the SSD affect the max and average temperatures of the bench test run in the end - so from this point on (including tests below) I always standardised the time between bench test runs with a strict 2 minutes wait time between runs.
- Firstly, for the first run only I always waited to let SSD reach its minimum temperature before I started the test (and opened a new HWiNFO window) - as you can see above, on the first of 5-run SSD tests, all three SSD temperature sensors can reach a lower temperature of 26°C vs 26/28/26°C (without heatsink).
- Secondly, even for the 2-minute-limited 5 runs with heatsink the maximum temperatures are lower than the ones in the 4-run tests without heatsink.
- Despite the 5 runs with heatsink having less wait time between runs, which can be seen by the 1-2°C higher start (minimum) temperatures, they still show a final average temperature smaller than the 4 runs without heatsink.
- All in all, these 1.5mm copper heatsinks I bought definitely work and help to reduce temperatures on SSDs.
3. Temperature test: Samsung 9100 Pro (gen5) in gen5 slot with OEM pads (no OS)
VS Samsung 9100 Pro (gen5) in gen5 slot with GELID pads (no OS)
In this temperature test, I wanted to check if changing the gen5 slot cooling pipe/plate OEM pads to GELID ones would decrease heat generation. For the run with OEM pads I am still using the inicial results from (
1.) where the 4 runs had no fixed wait time of 2 mim between runs. The above 5 runs with GELID pads already include the standardized 2min wait time between runs.
- Firstly, for the first run only (on the left) when waiting to let SSD reach its minimum temperature before I started the test, we can see that all SSD sensors reached a minimum of 27°C when using GELID pads vs 29/30/29°C when on OEM pads.
- When it comes to average temperature during the runs, they are pretty much the same on both pads with the GELID having just one sensor with +1°C. When looking at the maximum temperature we see that all 3 SSD sensors on GELID pads also have a +1°C. This would mean they are on par with the OEM pads - but because GELID pads allow SSD to cool more when CPU/SSD iddle, it's also possible that the pads conduct heat so well that the SSD cooling pipe actually conducts more heat from the big GPU/CPU cooler to the SSD cooling plate if that differential in temperature is bigger.. It's hard to say - this would ask to redo the test with OEM pads again with the standardized 2min wait time, but I just didn't want to touch the SSD once GELID pads were on to avoid any pad damage.
- Either way, just like the OEM Samsung SSD, I think it's safe to say the OEM pads are excellent too. Again, leaving the system OEM as it comes from the store will give you some of the best performance.
4. Temperature/speed test: 2 TB OEM Samsung (gen5) in gen4 slot + heatsink (with OS)
VS 4TB Samsung 990 Pro (gen4) in gen4 slot + heatsink (no OS)
This is a quick test just to check how the gen5 OEM Samsung SSD performs on a gen4 slot in comparison to the gen4 990 Pro on the same gen4 slot.
- Regarding the less important SEQ1M Q8T1 and RND4K Q32T1, the OEM Samsung is almost on par with the 990 Pro. But on the more important SEQ1M Q1T1 and RND4K Q1T1 the gen5 nature of the OEM Samsung gives it an edge specially when it comes to SEQ1M Q1T1. This makes the gen5 OEM Samsung SSD slightly faster than a 990 Pro on a gen4 SSD slot.
- Nevertheless, the OEM Samsung runs slightly hotter than the 990 Pro with a maximum of 55°C vs a maximum of 51°C on the 990 Pro. Note that any of these temperatures are still extremely good for SSDs that can reach a 70°C maximum.
5. Temperature test: OEM Samsung (gen5) in gen5 slot with MSI cooling pipe/heatsink (OEM pads)
VS OEM Samsung (gen5) in gen4 slot with copper heatsink (GELID pads)
- Looking at these figures in (
1. bottom) and (
4. top), we can see that actually the OEM Samsung SSD reaches higher average temperatures when used on the gen4 slot (with gen4 read/write speeds only) even with the copper heatsink - this means the MSI cooling pipe/heatsink solution is actually extremely good considering that the SSD is performing on gen5 speeds, generating a lot more heat, and still HWiNFO measures a smaller average temperature across the 3 SSD sensors.
6. RAID0 Speed tests:
6.1. RAID0 (stripe size = 4KB) with 9100 Pro (gen5) in gen5 slot + 990 Pro (gen4) in gen4 slot with heatsink
Best InstallAware Pre-Compiled Runtimes Tool time: 10:25 (less is better)
6.2. RAID0 (stripe size = 64KB) with 9100 Pro (gen5) in gen5 slot + 990 Pro (gen4) in gen4 slot with heatsink
Best InstallAware Pre-Compiled Runtimes Tool time: 10:23 (less is better)
- I started by creating the same two RAID0 systems above of 9100 Pro (gen5) + 990 Pro (gen4) with two different stripe sizes: a minimum of 4KB (
6.1.) and the default recommended 64KB (
6.2.) stripe size by the Intel Rapid Storage Technology (in MSI BIOS). Even though I was not expecting any significant difference (since these are SSDs), the 64KB stripe size shows better speed results, especially for the all important SEQ1M Q1T1 (and just slightly better RND4K Q1T1). This is then reflected in the real-world PCMark 10 FSDBenchmark where the stripe size of 64 KB shows higher bandwidth speed, slightly lower average access time, ending in better benchmark scores.
6.3. RAID0 (stripe size = 64KB) with 9100 Pro (gen5) in gen5 slot + Samsung OEM (gen5) in gen4 slot with heatsink + 990 Pro (gen4) in gen4 slot with heatsink
Best InstallAware Pre-Compiled Runtimes Tool time: 10:20 (less is better) but with a lot more inconsistent runs (higher delta): e.g. 10:32
- Keeping now the stripe size of 64KB fixed for further tests, I created a new RAID0 system with 3 SSDs (
6.3.), 9100 Pro (gen5) + Samsung OEM (gen5) in gen4 slot + 990 Pro (gen4). By looking at CrystalDiskMark sequential speeds we can see that there is some sort of cyclic behavior where results show the performance of either two SSDs (11000 MB/s SEQ1M read) or the actual three SSDs together as it should have been with +15000 MB/s SEQ1M read speed. Clearly, one of the three SSDs was not being used all the time making the results fluctuate between the performance of two or three SSDs in RAID0 - the Super Raid 5 system here seems to not be working properly which does not allow three SSDs in RAID0 to perform very well. This is evidenced by looking at the PCMARK 10 FSDB bandwidth and scores. They are actually worse than the RAID0 system (
6.2.) with only the 9100 Pro (gen5) + 990 Pro (gen4).
7. Speed test: Samsung 9100 Pro (gen5) in gen5 slot
VS RAID0 systems above
Best InstallAware Pre-Compiled Runtimes Tool time: 10:20 (less is better), reproducible on both Extreme "MSHybrid Graphics Mode" and "Discrete Graphics Mode".
- Finally, above are the results for the Samsung 9100 Pro (gen5) in gen5 slot - and they are easy to interpret vs any of the previous RAID0 systems. The more important SEQ1M Q1T1 and RND4K Q1T1 tests show much higher read and write speeds than any RAID0 system (with two or three SSDs), and this is reflected in the PCMark 10 FSDB with much higher bandwidth by 70 MB/s, much lower average access time by 13µs, and finally higher benchmark scores up by 500 points.
- The results are obvious: a single gen5 SSD in the gen5 slot will always be faster than any RAID combination of gen5 and/or gen4 SSDs, even when including the same gen5 SSD in the gen5 slot.
- Once again, if you don't care about extra storage, stick to the original OEM Samsung gen5 SSD with OEM pads and don't even bother creating a RAID0 system. OEM as is on these Titan 18 HX AI A2XWJG systems will be the best performance you will get.
drive.google.com
