Owners of a WD Black SN850 SSD have long reported that they write much more slowly in an M.2 slot connected to the X570 chipset. ComputerBase can confirm this and has approached Western Digital about it. The manufacturer now wants to get to the bottom of the cause.
Readers put the editors on track
The reports about a low write rate on the WD Black SN850 have been going back a long time and also found their way into the ComputerBase forum. The editors were made aware of this through tips from readers and have taken on the matter.
The thesis: Regardless of the manufacturer or motherboard model, the WD Black SN850 (test) proves to be considerably slower in sequential writing if it is not in the M.2 slot directly connected to the Ryzen processor, but in an M.2 Slot that receives its PCIe 4.0 lines from the X570 chipset. Mainboard manufacturers confirm that higher latencies when using the chipset can reduce performance, but have referred to WD regarding the blatant loss of performance in the SN850.
Benchmarks confirm problem in chipset slot
With its own measurements on a Gigabyte Aorus X570 Master from the SSD test system, ComputerBase was able to reproduce the problem on the third M.2 slot, which is attached to the chipset. CrystalDiskMark then reports a sequential write rate of just over 3,200 MB/s for the SN850 with 1 TB. In the first M.2 slot, which in turn is attached to the CPU, it is over 5,200 MB/s, which is close to the manufacturer's specification of 5,300 MB/s. The drop in performance in the chipset slot is therefore almost 40 percent and is even more pronounced in the SEQ1M Q1T1 setting.
«Previous CrystalDiskMark 8.0.1 (write) CrystalDiskMark 8.0.1 (read) Next»
- 24 entries CrystalDiskMark 8.0.1 (write) Unit: megabytes per second (MB/s)
- SEQ1M Q8T1:
- Mushkin Gamma 2 TB – PCIe 4.0 via CPU6.761.9
- MP600 Pro 2 TB – PCIe 4.0 via CPU6.752,1
- Mushkin Gamma 2 TB – PCIe 4.0 via X5706.335.0
- MP600 Pro 2 TB – PCIe 4.0 via X5706.293.2
- SN850 1 TB – PCIe 4.0 via CPU 5.254.8
- SN850 1 TB – PCIe 4.0 via X570 3,247.8
- Mushkin Delta 1 TB – PCIe 4.0 via X5701.940.0
- Mushkin Delta 1 TB – PCIe 4.0 via CPU1.937.6
- SN550 2 TB – PCIe 3.0 via X5701.883.8
- SN550 2 TB – PCIe 3.0 via CPU1.880.4
- SEQ1M Q1T1:
- Mushkin Gamma 2 TB – PCIe 4.0 via CPU5.927.5
- MP600 Pro 2 TB – PCIe 4.0 via CPU5.799.2
- MP600 Per 2 TB – PCIe 4.0 via X5705.336.0
- Mushkin Gamma 2 TB – PCIe 4.0 via X5705.324.7
- SN850 1 TB – PCIe 4.0 via CPU 5,255.0
- SN850 1 TB – PCIe 4.0 via X570 2,972.0
- Mushkin Delta 1 TB – PCIe 4.0 via X5701. 982.2
- Mushkin Delta 1 TB – PCIe 4.0 via CPU1.934.0
- SN550 2 TB – PCIe 3.0 via X5701.882 , 7
- SN550 2 TB – PCIe 3.0 via CPU1.878.5
- RDN4K Q32T1:
- SN850 1 TB – PCIe 4.0 via X570 660.4
- SN850 1 TB – PCIe 4.0 via CPU 652.8
- Mushkin Delta 1 TB – PCIe 4.0 via X570620.0
- Mushkin Gamma 2 TB – PCIe 4.0 via X570604,5
- Mushkin Delta 1 TB – PCIe 4.0 via CPU596,7
- MP600 Pro 2 TB – PCIe 4.0 via X570592,9
- MP600 Pro 2 TB – PCIe 4.0 via CPU580,3
- Mushkin Gamma 2 TB – PCIe 4.0 via CPU567.9
< li class = "chart__row chart__row - hidden toggle-body-item nojs-tr"> Kingston A2000 1 TB494,2
- SN550 2 TB – PCIe 3.0 via X570469,4
- SN550 2 TB – PCIe 3.0 via CPU433,1
- SEQ1M Q8T1:
- RDN4K Q1T1:
- Mushkin Gamma 2 TB – PCIe 4.0 via CPU268,1
- MP600 Pro 2 TB – PCIe 4.0 via CPU260,9
- SN850 1 TB – PCIe 4.0 via CPU 250.0
- Mushkin Delta 1 TB – PCIe 4.0 via CPU233.6
< li class = "chart__row"> MP600 Pro 2 TB – PCIe 4.0 via X570233,5
- Mushkin Gamma 2 TB – PCIe 4.0 via X570232,9
- SN550 2 TB – PCIe 3.0 via CPU225,1
- SN850 1 TB – PCIe 4.0 via X570 217.8
- Mushkin Delta 1 TB – PCIe 4.0 via X570208,7
- SN550 2 TB – PCIe 3.0 via X570200,4
- SEQ1M Q8T1:
- MP600 Pro 2 TB – PCIe 4.0 via CPU7.207.6
- Mushkin Gamma 2 TB – PCIe 4.0 via CPU7.191.0
- SN850 1 TB – PCIe 4.0 via CPU 7.067.5
- Samsung 980 Pro 500 GB6.865.2
- Mushkin Gamma 2 TB – PCIe 4.0 via X5706.369.8
- MP600 Pro 2 TB – PCIe 4.0 via X5706.347.6
- SN850 1 TB – PCIe 4.0 via X570 6.304.2
- Mushkin Delta 1 TB – PCIe 4.0 via CPU4.715.1
- Mushkin Delta 1 TB – PCIe 4.0 via X5704.715.1
- Samsung 980 1 TB3.566.7
- SN550 2 TB – PCIe 3.0 via CPU2.609.5
- SN550 2 TB – PCIe 3.0 via X5702.609.4
- SEQ1M Q1T1:
- SN850 1 TB – PCIe 4.0 via CPU 4,375.6
- SN850 1 TB – PCIe 4.0 via X570 4,008.5
- MP600 Pro 2 TB – PCIe 4.0 via CPU3,872.0
- Mushkin Gamma 2 TB – PCIe 4.0 via CPU3.869.0
- MP600 Pro 2 TB – PCIe 4.0 via X5703.655.4
- Mushkin Gamma 2 TB – PCIe 4.0 via X5703.627.2
- Mushkin Delta 1 TB – PCIe 4.0 via CPU2.346.7
- Samsung 980 1 TB2.307.0
- Mushkin Delta 1 TB – PCIe 4.0 via X5702.296.1
< li class = "chart__row"> SN550 2 TB – PCIe 3.0 via CPU2.120,7
- SN550 2 TB – PCIe 3.0 via X5702.103.2
- SN850 1 TB – PCIe 4.0 via X570 712.8
- SN850 1 TB – PCIe 4.0 via CPU 705.8
- Mushkin Delta 1 TB – PCIe 4.0 via X570674.7
- MP600 Pro 2 TB – PCIe 4.0 via CPU668,7
- Mushkin Gamma 2 TB – PCIe 4.0 via X570668,5
< li class = "chart__row"> Mushkin Delta 1 TB – PCIe 4.0 via CPU662.6
- SN850 1 TB – PCIe 4.0 via CPU 83.2
- SN850 1 TB – PCIe 4.0 via X570 81.1
< li class = "chart__row"> Mushkin Gamma 2 TB – PCIe 4.0 via CPU72.5
As a cross-check, the editors tested other PCIe 4.0 SSDs in the form of the Corsair MP600 Pro (test) and Mushkin Gamma (test to follow) in the same way. There was also a slump here, but with a loss of less than 10 percent, it is far less severe. PCIe 3.0 SSDs like the WD Blue SN550 do not seem to lose any of their performance. When reading sequentially with the SEQ1M Q8T1 setting, all fast PCIe 4.0 SSDs lose a good 10 percent in the chipset slot, which is a side note.
Manufacturer is looking for cause
Ultimately, it can be said that the SSD performance in the M.2 slot on the X570 chipset is generally somewhat worse than in the M.2 slot on the CPU, which is not very surprising and can be explained with higher latencies with longer cable paths . However, the WD Black SN850 loses a disproportionate amount of power when writing sequentially, which cannot be explained by this and is apparently a specific problem with this SSD series.
A concrete confirmation of this While not yet available, Western Digital has confirmed to ComputerBase that it is investigating the problem and its cause. As soon as new knowledge is available, it will be reported here.
Spurensuche
The “Maximum Payload Size ”on the chipset slot suspected to be the cause. However, the cross-check showed that this limit also applies to other SSDs and therefore does not appear to be a general problem.
Note on the screenshots: HWiNFO recognizes the WD Black SN850 as a “Sandisk Device”.
SN850 in the 1st M.2 slot (via CPU): Maximum Payload Size 512 Bytes 
Other NVMe SSD in the 3rd M.2 slot (via X570)): Maximum payload size 128 bytes