News MSI Latest SSD Pushes PCIe 4 to the Limit: Up to 7.4 GB/s and 1M IOPS

[Admin]

MSI's Spatium M480 Pro may be the world's fastest PCIe Gen4 SSD available, now or ever.

MSI Latest SSD Pushes PCIe 4 to the Limit: Up to 7.4 GB/s and 1M IOPS : Read more

 

[Dr3ams]

Probably need a heatsink the size of a truck to cool it.

 

[thisisaname]

Probably need a heatsink the size of a truck to cool it.

Was just wondering what cooling it would need to perform this well.

 

[atomicWAR]

Probably need a heatsink the size of a truck to cool it.

I have a PCIe gen 4 Nextorage 7400MBs drive and a WD 850X 7400MBs drive. If you want them to perform well, a good heat sink is needed for sure. I went with Acidalie M.2 SSD heatsinks that use a couple of heat pipes and a decent truck sized heat sink instead of the motherboard ones which would throttle so fast due to heat.

https://www.amazon.com/gp/product/B0B1M2J61F/ref=ppx_yo_dt_b_asin_title_o03_s00?ie=UTF8&psc=1

 

[HideOut]

Why is this even reportable. PCIe Gen 5 drives are shipping and trash this. Its like clickbait.

 

[edzieba]

Probably need a heatsink the size of a truck to cool it.

Assuming their specsheet is correct, the >12W peak power would mean it can be adequately cooled by a small passive heatsink over the controller (please do not attempt to cool the NAND dies, it will only reduce drive longevity). If anyone remembers passive northbridge pin-grid heatsinks, that's the sort of size for that power dissipation.

Why is this even reportable. PCIe Gen 5 drives are shipping and trash this. Its like clickbait.

If you're ignoring actual performance and buying SSDs based on "bigger bus number means more betterer" then, well, at least the marketers will be happy for the easy money.

 

[flagrantvagrant]

I use these for my PCI-E 4 nvme's. https://www.amazon.com/gp/product/B07KDDKDNN/ Work perfectly fine, but unlike most people I don't care about fan noise or pretty electric light show art cases. I run my beast with no panels and ripped off what few embellishments it had. Replaced the trainer wheel fans on the radiators with noctua 3kRPM industrial PPCs and run them with aggressive fan curves. I also bolted a 140mm fan facing inwards towards the low airflow area below the gpu, and put a large GPU backplate fan/HS on that, so all the heat sinked components in those areas are always getting a solid draft.

(please do not attempt to cool the NAND dies, it will only reduce drive longevity)

Okay sure, I'll make sure to totally not do that, even though doing real world stress transfer tests (like copying a 150+GB archive from one drive to another with 64k clusters, then touching the exposed NAND BGAs til I figure which one is taking the entire sequential write after DRAM and over provisioning is expended) is CLEARLY getting very warm because the 1TB NAND BGA (tested this on a 2TB Soldigm P44 and a 2TB Hynix P41) is getting written to sequentially at 2-2.5 GB/s, and is pulling at least as much power as the controller itself. Yeah, I think I'll be fully encasing my nvme's NAND chips and controller and keep air moving over them, TYVM.

Setup basics

MSI X570 Unify (oh no the NB fan is so loud how will I survive even though it prevents it from flying up to 90C!)
64GB gSkill (2x32GB) 4000 (downclocked and overvolted to get tighter timings running 1:1:1)
5950X tweaked to get all core all thread 4.5-4.6GHz AVX (CB23 AVG 29.5k to 30.5k pts)
Corsair H115i 280 AIO modified. Hung from DIY'd machined trusses since it didn't fit in the case; 3kRPM noctua PPC fans. No iCUE software, all controlled with BIOS settings. P95 AVX torture test 30 minutes: CPU maxed at 75C.
MSI liquid x suprim. Replaced radiator fans with 120mm 3kRPM noctua PPCs. Manual fan profile. OC maxes 2.8GHz with Bright Memory Benchmark maximum fidelity RT torture loop. 30 minutes: 65C with 90% fans.
Furmark torture loop: maxes 3+GHz. 30 minutes: 70C with 90% fans. Pulls almost 600Watts in both cases despite the whining and gnashing of teeth of "the community" at the "paltry" 525W PL in the VBIOS.
1x8TB and 1x16TB HGST SATA3 rust spinners for mass data retention. Both read/write 225-250MB/s steady sequential after cache expenditure.
1TB hynix P41 for OS only in slot 1. 4TB WD SN850X in slot two solely for games. Solidigm 2TB P44 in slot 3 for back up purposes. All using 64k cluster size (reduces MFT latency when assigning or reading block checksums, making read/write faster) All running full PCI-E 4x4. Max speed from one drive to another AFTER DRAM and over provisioning expenditure with very large sequential read/writes: 2-3GB/s depending on drive adjacency.
PSU: Corsair 1200 watt something. It's big and heavy. Who cares what model or brand it is as long as it gets the job done and has appropriate power conditioning, good input filtering caps (Y filter pattern FTW) high quality electrolytics and a well designed switching circuit.
Case: Its a metal box that parts are mounted in. Who cares.

 

[edzieba]

Okay sure, I'll make sure to totally not do that, even though doing real world stress transfer tests [...] is CLEARLY getting very warm

Getting warm is the goal. Higher temperatures at time of write means less power needed for every bit write, and less power needed for block erase. Operating at elevated temperature both makes NAND more efficient, and reduces wear by minimising power dumped into the NAND cells during write ops.
But more importantly, not cooling the NAND dies means the dies are not constantly expending energy attempting to reach their setpoint temperature (which they never can, because you're force cooling them), and because the NAND dies heat themselves by running block erases (the most energy intensive thing they can do) keeping the dies too cool is actively burning NAND cell lifetime.

 

[Li Ken-un]

lol. no.

The fastest PCIe 4.0 SSD hands down will be the Optane P5800X for a while, in terms of sheer sequential throughput, random throughput, and latency.