Why are m.2 SSDs so disappointing in real world performance?

[petteyg359]

So then the big question is why do SSDs have access times at all? HDDs have them because the head needs to physically move around so the computer spends all day waiting for the head to constantly move into position. But SSDs have no moving parts so why cant an SSD transfer 5000 files of 5MB at the same rate as one at 25,000 MB? Does RAM scale the same way?

Even DRAM has "seek time" in the form of latency between addressing. Current high-end NAND Flash from searching Google is 15 microseconds (15,000 nanoseconds). DRAM is 20 nanoseconds. Every address change is going to take that long. If a file is (or four files are) fragmented across four locations on Flash, that means 60,000 nanoseconds of time for the controller to access the (pieces of) the file(s). I'm not sure of the electrical reasons that STR "ramps up" (hence random reads being slower), though; I know 60 microseconds ought to be nothing at all observable to the user.

 

[mackerel]

It still takes a finite amount of time to move between even solid state storage locations. It is many magnitudes faster than a hard disk, but is still not unlimited. Current SSDs are limited in that they're still block devices, in that there is a minimum read/write size. I'm kinda hoping that Optane will drop enough in price as it is a new tier of solid state non-volatile storage performance. Performance wise is it somewhere between flash and ram, and you can address it in finer granularity. For now, it is held back by the high price it commands. I'm thinking if they introduce a 64GB model, I want to get one to test. Current 32GB is just a bit too tight for even a single game install and I don't want to raid a pair. It doesn't push performance for sequential or high QD random, but for single thread random, it is another level faster than what we know as SSDs.

 

[Woomack]

Yep, and then there's hotrodding overclocking fools that MUST have the fastest of everything. Which wouldn't be so bad except they have to brag about it too.

I see there is one benchmark that both SATA and NVMe are about the same.

Test rig looks like this...



While in my gaming PC I have MX300 1050GB and it's enough. Above results were made on stock i7-7900X/ASUS TUF2.

You can see that higher CPU/memory performance = higher random bandwidth. There is no separated storage processor. Because of that it's sometimes hard to compare results on different rigs.

 

[EarthDog]

So then the big question is why do SSDs have access times at all? HDDs have them because the head needs to physically move around so the computer spends all day waiting for the head to constantly move into position. But SSDs have no moving parts so why cant an SSD transfer 5000 files of 5MB at the same rate as one at 25,000 MB? Does RAM scale the same way?

https://www.extremetech.com/extreme/210492-extremetech-explains-how-do-ssds-work

 

[Tír na nÓg]

I don't really care about Windows load time, as I boot once a week.

But I have installed Adobe CC, STeam and Oculus games directories on the NVMe, and apps load quicker.

Nothing huge, but noticeable.

 

[batboy]

Woomack, you say the results were made on a stock i7-7900X/ASUS TUF2, but what drives are the benchmark for?

 

[Woomack]

Woomack, you say the results were made on a stock i7-7900X/ASUS TUF2, but what drives are the benchmark for?

2x Patriot Hellfire 240GB in R0
Results from Alphacool Eisblock HDX-2 thread. Many members wanted to see results and later barely anyone checked this thread:
http://www.overclockers.com/forums/...l-Eisblock-HDX-2-M-2-PCIe-Cooler-RAID-results

 

[batboy]

Thanks for the info, I wasn't aware of that other thread. I looked at your review. Impressive benchmarks with those striped m.2 drives. Nicely done article. I really want to try running a pair of Samsung 960 Pro drives striped one of these days. Your article says the drives ran hot, but not too hot. What were high temps during benchmarking? Like over 70 C?

 

[Woomack]

These drives run up to 70°C without any cooling. With Alphacool heatsinks temps were closer to 40°C. When I was performing tests then I couldn't check temps but later it was working in other soft.
Considering that max temp for commercial usage is about 80-85°C then there is no need of additional cooling, especially when is good airflow in the case. However PCIE cards/controllers have higher bandwidth so it's possible to pass ~4GB/s DMI limit. Worse is that you can't set RAID on 2 separated cards without RAID controller and without hardware key for motherboard. It's possible to set only stripped volume in Windows but can't boot from that. On the other hand it's maybe better to use SATA SSD as boot/OS drive and all applications/games keep on PCIE drives.