We couldn’t help but notice the excitement in the last day or so as Linus posted a rather detailed Crucial Gen 5 T700 NVMe SSD exclusive. Kudos for snagging that as well as the detailed SSD breakdown for those not familiar with SSDs. Something struck us a bit odd though and we are rather certain that we are walking the line on this, but let’s just say we can only expect a bit more on final release of this SSD. After all, the component layout of the T700 is very similar to our early sample review of the Phison PS5026-E26 reference design. The key components are EXACT in fact.
The Linus report depicted a single Crystal DiskMark result that was obtained through their ‘AMD based test bench’. AMD. As much as most people get very excited about those high sequential speed numbers, the meat and gravy of an SSD lies in random reads and writes. Looking at this Crystal Diskmark result they posted and keeping in mind that they tested Random 4K at Q32 T1, things are just a bit underwhelming:
Actually these results just look really bad and no fault whatsoever of Linus, but rather, it is the inevitable result of the test system in use. AMD. Historically, the consumer has been groomed to believe that high sequential read and write speeds are the be all and end all of SSD performance. In reality though, most might only ever reach this performance plateau in less than 1% of their use. A big thanks to Linus for our use of that result and you will be directed to his video review if you click on the graphic above.
Now if you take a look at our earlier result of this same SSD in its original reference design posted above, but for different firmware we are certain, you might see a significant difference in random performance, even though we had originally tested in Q32T16. Compare that low random 4K Q1T1 read and write result alone if you like. This is the result of testing with an Intel Test Bench, compared to an AMD Bench. To be very clear, these two lower results ARE NOT the result of the T700. We would lay our reputation on the fact that you are going to see higher random 4K IOPS that will easily break the million mark (below) as with our testing, as well as some of the best low random 4K throughput results with reads breaking the 100MB/s mark and writes close to if not over 400MB/s (above).
In fact, we are certain that if Linus tested again with an Intel test bench in the same manner as we demonstrated here to achieve what would be the manufacturers expected low 4K throughput and IOPS, he would see results very close to this, only with that 12GB/s plus sequential speed he has shown. After all, Linus himself opened the video with a complaint of his under par 69MB/s low 4K random read performance which was a low 4K Q1T1 random read speeds True fact is though that an Intel based system is bringing you better 4K random read performance and better 4K write speeds on this same Crucial T700 SSD, at least according to our earlier tests of its reference design.
So…. wrapping things up, we predict that the Crucial T700 is going to be a powerhouse of an SSD, when testing with an Intel based system, with the same sequentials you have already been shown, but with much better random 4K results and IOPS as we have displayed in the bottom two results. After all, the hardware in these SSDs is virtually identical, thanks in part to Phison. Let’s hope for the best in what looks to be a dynamite Crucial T700 release in the near future!
> We would lay our reputation on the fact that you are going to see higher random 4K IOPS
You may put your qualification on the line explaining:
1. Why Linus’ CDM has Random 4K in Q32T1 mode and your own in Q32T16 mode?
1. How does the performance difference come about?
Because a) it’s possible to install the same NVMe driver on Windows for both systems (iirc) b) lock both processors to the same frequency c) probably possible to pin the driver to the same core or adjacent cores
d) replicate the benchmark numbers under Linux and compare to Windows figures e) disable CPU vulnerability mitigations and compare those against the default baseline
Yes, both systems are different and both tests vary as well. Our testing of random performance is from the initial tests of our Phison gen 5 reference design and matches the industry standard where the client wants to validate that the SSD they just bought matches that of the manufacturers specifications. Thank you for commenting as we have amended our report slightly to more accurately reflect the intent in our statements. If it assists, we will do our best today to retest same reference design gen 5 drive in same configuration that Linus used exactly. We are confident that one will see a noticeable difference in results as is already evident in the bottom Q1T1 result.
As I recall, ALL of your AMD/Intel comparisons over the years have shown the same differences with AMD lagging behind.
I am with you: I don’t pay much attention to sequential speeds, as I will never be dragging 100GB files around all day, but the Q1T1 makes all the difference to me, as I use my PC the same as the majority of users do: Internet, videos, pictures, etc. I want my PC to be as ‘snappy’ as possible, which is what high Q1T1 speeds give us. That is why I became interested in the Samsung 990 Pro you reviewed: excellent Q1T1 results (along with the rest of the results). I guess they got the ‘quick death’ issue fixed with a new firmware release. At least I hope so.
Do you think AMD is aware of their SSD performance issue?
However, in the end, it looks like their new Ryzen 7000 Zen 4 CPUs are showing excellent benchmarks even with the SSD issue.