Review: Kioxia Exceria PLUS 2TB M.2 NVMe SSD (PCIe Gen3 Going for 3400 MB/sec)

Published by

Click here to post a comment for Review: Kioxia Exceria PLUS 2TB M.2 NVMe SSD (PCIe Gen3 Going for 3400 MB/sec) on our message forum
https://forums.guru3d.com/data/avatars/m/196/196426.jpg
2TB version: https://www.amazon.de/KIOXIA-EXCERIA-NVME-2280-2000GB/dp/B087CPPP4L/ € 430 Way more expensive than a Samsung Evo Plus https://www.amazon.de/Samsung-SSD-970-EVO-Plus/dp/B07MLJD32L/ € 370 1TB version: https://www.amazon.de/KIOXIA-EXCERIA-Plus-Gen3x4-LRD10Z001TG8/dp/B087CPFYP5/ € 175 More expensive than a Samsung 970 Evo (and only slightly cheaper than the Evo Plus) https://www.amazon.de/Samsung-MZ-V7E1T0BW-970-Interne-NVMe/dp/B07CGJNLBB/ € 167 Kioxia is not starting on the right foot with these prices.
https://forums.guru3d.com/data/avatars/m/274/274425.jpg
As soon as I saw the DRAM chips in the photos, I figured the performance would be at least decent. (I don't even bother reading m.2 storage reviews where they aren't plainly visible.) I wonder what the delta is, in terms of the BOM, to include a couple of DRAM chips and a controller that can properly utilize them, versus not? $4.00?
data/avatar/default/avatar34.webp
Impressive performance. @Hilbert Hagedoorn: Sadly the scores are not comparable with your previous tests because of the CPU choice. These new, fast SSDs that push against the PCIe limits are very dependent on the CPU speed, but also its security mitigations. On the 970 Evo bench a 6600K was used with months of security mitigations in place. A far better CPU and platform was used for newer tests. Just as an example, on my 8700K/970 Evo+ combo, running at default with latest BIOS/microcode and mitigations would score a rather low 15800 in AnvilPro. Running the CPU at 5GHz with earlier an BIOS would score 18200 in the same bench. Can be reliably reproduced. TLDR is that if you are benching the CPU instead of the SSD basically and that the Intel mitigations mess pretty hard with SSD benchmarks.
https://forums.guru3d.com/data/avatars/m/16/16662.jpg
Administrator
No that's an incorrect assumption. I've performed A-B testing once the Meltdown/Spectre drama kicked in and made a decision to make sure that at the time the vulnerability migations would not be installed (to not mess with results). So the test system was not updated and thus not hindered in performance (especially 4K results showed an offset). However, you can only skip updates so long, ergo that is one of the reasons we have moved to a Ryzen platform.
data/avatar/default/avatar31.webp
It's actually pretty east to test by running something like a 8700K at default settings (i.e. 4.3GHz boost in Anvilpro) and then OCing at 5GHz, it should give you 1000 points at least (for good SSDs like Evo+). If you watch the CPU during SSD benching, a thread will get 100% usage so it's really no surprise that with such a bottleneck there would be an impact on the score. Should be quite reliable to test, for example here's my 970 Evo+ with stock 8700K and 5GHz 8700k on an older BIOS. The newest BIOS, with the D6 microcode collapses the score below 16K. I'd say such a reliably reproducible difference in score is significant, as it puts a 500GB 970 Evo+ well above the 16k scores you obtained with the 6600K, for example. Oh, and both these are with all mid 2019 mitigations enabled up to and including MDS, basically before Plundervolt. Your 2GB Evo+ should score better than this 500GB drive for sure. [spoiler]https://i.imgur.com/Id4jAsy.png https://i.imgur.com/3dvcyvT.png [/spoiler]
https://forums.guru3d.com/data/avatars/m/262/262190.jpg
Hilbert Hagedoorn:

(...)we have moved to a Ryzen platform.
Thank you for the review! Have you tested the impact of the Infinity Fabric/RAM speed in NVMe transfer rates?