Samsung 990 pro real world performance

[Dazzy]

My new 2TB 990 Pro NVME isn't hitting the advertised speeds in Samsung Magician benchmarks. It's supposed to reach up to:

• 7,450 MB/s sequential read / 1.4m IOPS random read
• 6,900 MB/s sequential write / 1.55m IOPS random write

However, I'm only seeing 6,500 sequential reads and 6,000 writes. Random reads/writes are around 1.1m IOPS reads and 0.9m IOPS writes.

I've got it installed in a 12th gen Intel system in a 4x4 M2 Slot (both CPU and Chipset lanes tried), it's not thermal throttling, and it's a brand new, non-OS drive. There's no obvious BIOS setting impacting it and everything is on the latest firmware.

I'm wondering if Samsung might have silently revised the drives, as some recent Amazon reviews mention max reads capping at 6.5k on a PS5.

Could anything else be affecting the speed, or might it be a bad drive?

 

[Mcnumpty2323]

Try a different benchmark
Crystal disk mark,as ssd,or atto for example
A benchmark on only 1 software
Isn't really telling as could be the software for
Some reason

 

[Dazzy]

Try a different benchmark
Crystal disk mark,as ssd,or atto for example
A benchmark on only 1 software
Isn't really telling as could be the software for
Some reason

Thanks for this suggestion. I've now tested on Cristal Disk Mark and it's closer to the advertised speed (7100 read / 6900 write).

I still wonder if they have revised them though as it still seems a little lower than it should be (comparing to older reviews of the same drive). Magician was also benchmarking the advertised speeds for others. For example RND4K Q32T16 random read IOPS of 1.1m vs 1.55m advertised (and achieved in real world reviews on Cristal Disk Mark).

Got it for £80 after cashback so still cheap for the actual speed acheived though.

 

[Mcnumpty2323]

In crystal disk mark
Go to settings
Can't check right now exactly what it says
But there's a setting for nvme drives
Enable that and probably will show
A higher speed

 

[Dazzy]

In crystal disk mark
Go to settings
Can't check right now exactly what it says
But there's a setting for nvme drives
Enable that and probably will show
A higher speed

The figures in my post were already from the NVME mode so it must be something else causing it.

 

[Mcnumpty2323]

The figures in my post were already from the NVME mode so it must be something else causing it

7100/6900 is pretty close
Could try atto test it usually reads higher than others
But those figures are close enough to advertised
That could just be something in the background
Is slightly using the drive

 

[Simon42]

I've got dozens of NVMe drives in various systems and I only ever check they're within the ball park for speeds. If it works, doesn't have any SMART errors and is close enough to show there is not a major driver/system issues then that's good enough for me. Quoted figures are on ideal system with a clean Windows install.

I've got a couple of 990 Pro 2TB and just tested one and it has 7132MB/s Read and 6887MB/s Write and that's with AV running and a PC used for software development including virtualisation. I wouldn't give those results a second thought for there being an issue.

 

[chaparral]

Thanks for this suggestion. I've now tested on Cristal Disk Mark and it's closer to the advertised speed (7100 read / 6900 write).

I still wonder if they have revised them though as it still seems a little lower than it should be (comparing to older reviews of the same drive). Magician was also benchmarking the advertised speeds for others. For example RND4K Q32T16 random read IOPS of 1.1m vs 1.55m advertised (and achieved in real world reviews on Cristal Disk Mark).

Got it for £80 after cashback so still cheap for the actual speed acheived though.

I wouldn't worry about it

My 2TB 990 pro

My 4TB 990 Pro

 

[Rroff]

Just bought another one of these though will take me a day or two to get around to fitting it as I need to take half the system apart to get to the M.2 slots so will do a comparison with my older ones.

 

[Rroff]

Within normal benchmark margin of error for me.

My oldest 990 Pro:

One I bought a few days ago:

Having 2 of them sitting under the same motherboard heat spreader (along with an SK Hynix NVME) isn't doing wonders for the temperatures though - before the one there was running about 5C cooler idle than the one with the factory fitted heatsink on its own slot - now they are all about even temperature wise under idle or up to moderate use - though the ones under the motherboard heat spreader are still cooler under heavy load than the one with factory heatsink.

 

[Dazzy]

Just bought another one of these though will take me a day or two to get around to fitting it as I need to take half the system apart to get to the M.2 slots so will do a comparison with my older ones.

Thanks for doing this comparison. Interesting to see that the performance is just is fast on a new one. Still not clear why mine is getting much lower speeds than everyone else in Magician (and random reads in Crystal Disk Mark) but it's fast enough not to really matter.

 

[Simon42]

Thanks for doing this comparison. Interesting to see that the performance is just is fast on a new one. Still not clear why mine is getting much lower speeds than everyone else in Magician (and random reads in Crystal Disk Mark) but it's fast enough not to really matter.

I wouldn't spend time investigating, but a key question is whether your motherboard, Windows install, drivers or other software is making this small difference.

I've got lots of systems so testing such issues is simple, but if I only had one system I wouldn't worry about it given its close enough to spec. Chasing down the last few percent on parts of an overall system's performance isn't worth the time for me.

 

[Mujja]

I have 2x 2TB 990 Pros, ran benchmarks in Samsung Magician for comparison.

M.2 slot to CPU

• 7,345 MB/s sequential read / 1.40m IOPS random read
• 6,913 MB/s sequential write / 0.96m IOPS random write

M.2 to Chipset

• 7,452 MB/s sequential read / 1.40m IOPS random read
• 6,928 MB/s sequential write / 1.57m IOPS random write

I have another 4x 1TB 990 Pros in a ASUS HYper M.2 card, ran the benchmark on one of these

• 7,451 MB/s sequential read / 1.20m IOPS random read
• 6,914 MB/s sequential write / 1.59m IOPS random write

All reach Samsung's numbers from your OP.

 

[Shaolin Chicken]

I have 2x 2TB 990 Pros, ran benchmarks in Samsung Magician for comparison.

M.2 slot to CPU

• 7,345 MB/s sequential read / 1.40m IOPS random read
• 6,913 MB/s sequential write / 0.96m IOPS random write

M.2 to Chipset

• 7,452 MB/s sequential read / 1.40m IOPS random read
• 6,928 MB/s sequential write / 1.57m IOPS random write

I have another 4x 1TB 990 Pros in a ASUS HYper M.2 card, ran the benchmark on one of these

• 7,451 MB/s sequential read / 1.20m IOPS random read
• 6,914 MB/s sequential write / 1.59m IOPS random write

All reach Samsung's numbers from your OP.

Surprised the chipset has higher performance numbers than direct to the CPU, it's usually the other way round.
What motherboard do you have?

 

[Mujja]

Yes, never ran any benchmarks until now and i was expecting lower performance. I only have USB peripherals connected, no SATA devices. The network wasn't doing much, so I guess the chipset had enough bandwidth available.

The Hyper M.2 card is in PCIE_1, I'm using IGP. If that was in PCIE_2 (chipset) it may have affected the numbers.

Motherboard is an ASUS B650 MATX WiFi, nothing special.

 

[Rroff]

Surprised the chipset has higher performance numbers than direct to the CPU, it's usually the other way round.
What motherboard do you have?

The 990s seem to have a fair bit of run to run variance - if I ran them multiple times I can get anywhere from 7000 to 7450MB/s and 1.2 to 1.6M IOPS. With full power mode (I'm using standard in screenshot above) the CPU connected slots give very slightly more consistent results on the higher side.

 

[freddie64]

I'm only seeing 6,500 sequential reads and 6,000 writes. Random reads/writes are around 1.1m IOPS reads and 0.9m IOPS writes.

My 4TB 990 pro is also running similar speeds. 6486 read and 6204 write, IOPS 1277099 read and 1038574 write.

It is running off the chipset, but apparently it shouldn't make much difference.

It is partitioned into 2 x 2TB partitions but again, have read this shouldn't matter.

Things i have tried to fix it

Change pcie from auto - gen 4
Samsung Magician set to full performance mode
Windows power mode set to best performance
Latest chipset drivers
4K aligned

The only other thing i can think of is to try it in the slot from the CPU.

 

[Dazzy]

My 4TB 990 pro is also running similar speeds. 6486 read and 6204 write, IOPS 1277099 read and 1038574 write.

It is running off the chipset, but apparently it shouldn't make much difference.

It is partitioned into 2 x 2TB partitions but again, have read this shouldn't matter.

Things i have tried to fix it

Change pcie from auto - gen 4
Samsung Magician set to full performance mode
Windows power mode set to best performance
Latest chipset drivers
4K aligned

The only other thing i can think of is to try it in the slot from the CPU.

I wonder if there is some sort of unit variance. CPU vs chipset made no difference for mine.

 

[Mujja]

My 2TB drives are partitioned in to two 1TB partitions. These benchmarked the same as the 1TB drives, with a single partition.

The two 2TB drives had similar results, with one connected to the CPU and the other to the chipset.

I have changed any Windows power or NVME mode settings. Using the latest chipset drivers.

 

[freddie64]

Not tried the top slot on the motherboard yet, but altered a setting in power managent "Link State Power Management" set to off.

It's improved a little bit, but not much.

Spoiler: Benchmark

Benchmarks done one after the other and not sure why Samsung Magician is lower than CrystalDiskMark.