Two SSDs, different results

[Kazlehoff]

So i have a 250GB Samsung 980, and a 500GB Samsung 980. The 250GB is in the M.2 closest to the CPU. (boards a 390 chipset, 8600k cpu). OS on the 250GB. both SSD's also have a 20% Overprovision.

 

The attachment below is what Samsung magician is giving me for preformance comparison. both benchmarks were done on a freshly booted copy of Win11 with nothing running in the background.

 

Now i know that smaller SSDs have less chache and should be slower. what im confused by is the dramatic difference in sequential reads and Radom Writes (higher on the smaller drive) and the Sequential Writes on the bigger drive.

 

Can someone explain whats up here?

(this is in general instead of troubleshooting because i dont think something is the matter, and am not seeking troubleshooting to fix this, im just looking for general knowledge that might explain why it is happening)

[curiousmind34]

How full are both of those drives as a %? I'm asking because an   almost full drive will generally run slower than their empty counterparts.

[Kazlehoff]

5 minutes ago, curiousmind34 said:

How full are both of those drives as a %? I'm asking because an   almost full drive will generally run slower than their empty counterparts.

Excluding the 20% over provision on each drive (bringing total space to 208GB on the 250 and 419GB on the 500), and the other partitions windows makes on the boot drive, the 250GB drive has 69.8GB free, and the 500GF drive has 160GB free.

[mariushm]

Think of it like single channel DDR4 vs dual channel DDR4  ... the ssd controller uses multiple channels, spreading reads and writes across flash memory chips, and flash memory chip layers. The 250 GB version may only use half the channels because there's half the memory chips or half the memory layers within the chip.

 

Also note these use HMB  (host memory buffer), where the drives reserve a portion of the RAM to cache things in, so that could affect tests.

 

[Kazlehoff]

3 minutes ago, mariushm said:

Think of it like single channel DDR4 vs dual channel DDR4  ... the ssd controller uses multiple channels, spreading reads and writes across flash memory chips, and flash memory chip layers. The 250 GB version may only use half the channels because there's half the memory chips or half the memory layers within the chip.

 

Also note these use HMB  (host memory buffer), where the drives reserve a portion of the RAM to cache things in, so that could affect tests.

 

Ah. Good point to mention, thank you.

 

Host system has 16GB DDR4 3000 (Corsair LPX modules), dual channel, and with each test the OS was freshly booted with nothing running in the background. only Magician.

with the top slot being on the CPU directly, and the bottom being on the PCI bus (i think thats how boards are doing it... its an asus Z390-E gaming) would the top slot be preforming better due to the PCIe bus not being involved?

[mariushm]

No dude, it's like this - and I'm super simplifying so it may be a bit incorrect due to over simplification.

 

Both drives have a single flash memory chip.

But picture the 250 GB flash memory chip as having the actual memory cells arranged as 4 levels of 64 GB  and the 500 GB chip as having 8 levels of 64 GB each.

The SSD controller has a fixed number of channels, let's say 4, and they can be connected to levels on the flash memory chip. 

With the 250 GB chip, each channel can send a write request for 512 KB (or some amount of data) and then it may take a few ms or us (microseconds) for the level to become available again for writing.. so for a bit of time, all 4 channels are stuck waiting for the 4 levels of memory to be available again. 

With the 500 GB chip, the ssd controller can send write requests to levels 1,2,3, and 4 but instead of waiting for a level to become available for writing, the controller can simply connect to levels 5,6,7, and 8 and put some data on those levels and then quickly go back to previous four channels and wait the remainder of the time until the levels become available again.

So with the 500 GB memory, the controller spends less time waiting for the flash memory chip to accept new data, as it can bounce between the extra layers available in the memory chip.

That's why I made the analogy with single channel vs dual channel but i guess it's not quite correct, maybe better would have been single rank vs dual rank single channel or dual channel.. dual rank is like having two levels, the cpu can write some stuff in first rank on the slot, but then that rank becomes busy for some nanoseconds until data is settled, but the cpu can switch to writing data in the second rank of memory on the same ram stick.

Dual channel makes it possible for the cpu to use two channels in parallel to send or receive data from ram sticks, so you get double the read speed or double the write speed.

 

 

The m.2 near the cpu in theory is connected with pci-e 3.0 x4 to cpu, so you get maximum 4 x 985 MB/s (max. theoretical bandwidth)

 

The second m.2 is provided by chipset which also receives data from other sata ports, from usb controllers and so on, and mixes together data and sends it to CPU using the DMI link which should be also equivalent to pci-e 3.0 x4.

In theory you get less maximum bandwidth, but the drive itself can't saturate the bandwidth either way so it doesn't matter.

 

It should be irrelevant where the SSDs are connected.  The difference in construction of the drives causes difference in performance between the drives.

Also the boot drive could be affected if you have applications running in background writing or reading from ssd - and you most certainly do have some services running in background if it's a fresh windows

It's Windows 11 also ... beta.  The Samsung drivers are probably not tuned for it.