Of course, the reason I'm having trouble is they never say what the green plot shows, but it's implied with their thermal image and the associated text:
"We recorded a thermal image of the running SSD as it was completing the write test. The surface temperature of the heatsink 101°C, which very closely matches what the onboard thermal reporting shows."
For it to match the onboard thermal reporting, it should mean the green line is the controller's temperature and the red line is the hottest NAND chip's temperature.
I love that last line on the page. It clearly shows an FLIR image with no heatsink and then says "The surface temperature of the heatsink 101°C." Never mind the incomplete sentence (it's missing the verb "was"), but that is not an SSD with a heatsink and is precisely the problem. Unless by "heatsink" he means "a sticker that on the 990 Pro isn't even copper coated."
Most of the fastest PCIe 4.0 drives will get hot if you run them bare, with no fan or airflow in the neighborhood. Even some PCIe 3.0 drives can get hot in such circumstances. It's why mobos started putting in heatsinks even prior to PCIe 4.0 drives. PCIe 5.0 drives all need a good heatsink for sure.
If you actually had a typical motherboard heatsink on the drive, it would probably drop the temperature way down initially, and then it would take much longer to heat up. These drives basically max out at around ~7W of power draw (they might peak at 10W, but that's just a very short-lived spike), so that's what your heatsink has to cool. In a case with little to no airflow over the primary M.2 slot (plus a lot of heat from the CPU and GPU, potentially), you could still heat up the SSD over time. It would just take a long time.
I definitely think stuff like this represents a very niche problem. To hit these high temperatures, you have to run a sustained write for 10+ minutes, usually a synthetic write test because nothing else puts as much strain on the drive. Downloading a game? No problem, because you're restricted by your connection speed to 1Gbps or lower.
The only way to realistically hit these sorts of sustained loads in a consumer PC would be to copy your entire game library from one fast SSD to a second SSD. Then you might have 2TB of data to write all at once. And you'll probably still have much lower overall transfer rates (and thus, temps) than a synthetic Iometer test.
You're missing my point, which was that it has two separate temperatures, with the lower one being the NAND. I think the temperatures normally reported via SMART are from the NAND, rather than the controller as you previously stated.
And the reason I mention that is to point out that the temperatures on that PCIe 5.0 drive you showed are probably also from the temperature of its hottest NAND chip. As the NAND temperature is what matters most, it's what's most likely to be reported via SMART.
If you had access to a thermal imager, perhaps you could verify it as they did.
The review hasn't posted, but it's the same as all the other E21T + QLC 2230 drives in performance characteristics. If this isn't QLC, I'll eat my hat. Also, I have a photo of the drive's chips. The NAND is labeled "ICCVG96AZA," and while there aren't many hits on that, the few I can find agree it's QLC NAND.
You're missing my point, which was that it has two separate temperatures, with the lower one being the NAND. I think the temperatures normally reported via SMART are from the NAND, rather than the controller as you previously stated.
And you're missing the fact that running a drive bare will obviously cause much higher temperatures, on both the controller and NAND. If TPU measured 101C on the controller and 83C on the NAND with no heatsink or fan, having a heatsink should drop both by at least 20~25C in my experience, more if it's a larger heatsink (plus modest airflow in the case).
But I do have the temperature file for the Samsung 990 Pro 4TB as well. It has three thermal sensors, presumably the first two for NAND and the third for the controller. I say that because the third sensor gets ~20C hotter than the first two. (Again, this isn't consistent on the various SSDs. Some have one temp sensor, some two, and some three. Sometimes, the first is controller and second is NAND, other times it's swapped.)
Max temps in my testing, with a large heatsink, were 46C on the NAND and 65C on the controller. So nearly 40C lower than what was measured with a bare drive and no fan. Granted, it's a 4TB instead of 2TB, but that shouldn't impact the results much.
The review hasn't posted, but it's the same as all the other E21T + QLC 2230 drives in performance characteristics. If this isn't QLC, I'll eat my hat. Also, I have a photo of the drive's chips. The NAND is labeled "ICCVG96AZA," and while there aren't many hits on that, the few I can find agree it's QLC NAND.
Yeah, they are, though TBH the 2TB QLC drives are all fine as well for the Deck. I mean, you're stuck downloading over WiFi, which means ~35 MB/s. Even the QLC can write at around 100 MB/s when it's in degraded (ran out of pSLC cache) mode.
If you fill the drive with 1.8TB of data, leaving just 200MB free, and then have some large game patch, it might end up slowing down a bit. But also, the APU in the Steam Deck isn't super fast for patching games either, and may not push much more than 100~200 MB/s for such a workload.
Don't get me wrong: I despise QLC drives. I have a couple of laptops running Windows 11 with QLC drives, and they suck on Windows updates and even driver updates. (Also, one is 512GB and the other is 1TB, and I think they're earlier drives that lack the pSLC caches on newer models.) But for SteamOS, I haven't generally noticed the impact of QLC much, thanks to the hefty pSLC caches.
1TB TLC versus 2TB QLC for the Deck? For the price, the 1TB is a better value, but if you just want capacity, you basically have to do QLC or the WD SN770M. So it's $220 for the WD, or ~$150 for the least expensive QLC (Silicon Power UD90 2230). I definitely would prefer the WD drive... but would I prefer it enough to pay nearly 50% more? (Personally, no, but then I don't use the Steam Deck hardly at all.)
Yeah, they are, though TBH the 2TB QLC drives are all fine as well for the Deck. I mean, you're stuck downloading over WiFi, which means ~35 MB/s. Even the QLC can write at around 100 MB/s when it's in degraded (ran out of pSLC cache) mode.
If you fill the drive with 1.8TB of data, leaving just 200MB free, and then have some large game patch, it might end up slowing down a bit. But also, the APU in the Steam Deck isn't super fast for patching games either, and may not push much more than 100~200 MB/s for such a workload.
Don't get me wrong: I despise QLC drives. I have a couple of laptops running Windows 11 with QLC drives, and they suck on Windows updates and even driver updates. (Also, one is 512GB and the other is 1TB, and I think they're earlier drives that lack the pSLC caches on newer models.) But for SteamOS, I haven't generally noticed the impact of QLC much, thanks to the hefty pSLC caches.
1TB TLC versus 2TB QLC for the Deck? For the price, the 1TB is a better value, but if you just want capacity, you basically have to do QLC or the WD SN770M. So it's $220 for the WD, or ~$150 for the least expensive QLC (Silicon Power UD90 2230). I definitely would prefer the WD drive... but would I prefer it enough to pay nearly 50% more? (Personally, no, but then I don't use the Steam Deck hardly at all.)
I don't want QLC in my house (at least not in SSDs), mostly because of their endurance. Although the jury is still out on my 1 TB Kioxia BG5 in my current Deck...
I know that it's just a theoretical limit and even 450 TBW would last me 50+ years. I have now 5 TBW (with write amplification maybe 10) on my Kioxia after 13 months of moderate usage, but the games are split between it and my microSD card. But I just don't want to make any compromise and worry all the time about how many TBW I still have left.
I have seen several times that patches ran at more than 700 MB/s for at least 1 minute on this Kioxia, and games are downloading with more than 50 MB/s here over wifi on my launch Deck.
Btw, that SN770M isn't that great either when the SLC cache runs out, with its 250 MB/s. Normal TLC speed should be 1.5-2.2 GB/s all the way (see the NM790 and most DRAM drives).
I never said otherwise. However, this is also a lower-powered drive than some of the PCIe 5.0 drives that want monster heatsinks. While it might do alright with a passive heatsink, that doesn't mean you'd never have throttling from one of the higher-powered drives with a passive heatsink. At least, I don't have enough information to rule it out.
If TPU measured 101C on the controller and 83C on the NAND with no heatsink or fan, having a heatsink should drop both by at least 20~25C in my experience, more if it's a larger heatsink (plus modest airflow in the case).
But I do have the temperature file for the Samsung 990 Pro 4TB as well. It has three thermal sensors, presumably the first two for NAND and the third for the controller. I say that because the third sensor gets ~20C hotter than the first two. (Again, this isn't consistent on the various SSDs. Some have one temp sensor, some two, and some three. Sometimes, the first is controller and second is NAND, other times it's swapped.)
Don't just think of write endurance. If you're putting it in some device you might not use very regularly, you should also think of data-retention. QLC is at a fundamental disadvantage there, because we're talking about the charge in each cell not being able to stray by more than 1/16th, before the cell starts reporting the wrong value for that group of 4 bits it's holding. While the drive uses error-correction, that only saves you if just a few outlier cells have drifted. If the entire chip has gone long enough without power, then your data is toast.
This exact thing happened to me with an eReader. I don't know what technology its built-in NAND used, but I went 6 months without using it during the pandemic (I had left it at the office) and when I turned it on again, it lost off the books and documents I had copied onto it. The irony is I paid extra for the 32 GB model - maybe if I'd just gone with the base 8 GB model it'd have used an older, lower-density NAND with better retention.
Congratulations, you found an error in a massive table that had to be manually input. Those things do happen when you have to punch in all of this data, but don't worry, I'll fix it. Also, I corrected all the typos that said "Micon" and they now say "Micron." Instead of acting like it's an intentional error, or "fishy," maybe next time just point out the mistake and don't act like there's some ulterior motive here.
Here's the full table from my Excel sheet, which other than the SSD names was manually copied or input. Meaning, I had to go through every past review of each of the more than 100 SSDs to check the controller, NAND type, and PCIe version and input that into the table, plus pulling the review URL. It took quite a few hours. You're welcome.
Congratulations, you found an error in a massive table that had to be manually input. Those things do happen when you have to punch in all of this data, but don't worry, I'll fix it. Also, I corrected all the typos that said "Micon" and they now say "Micron." Instead of acting like it's an intentional error, or "fishy," maybe next time just point out the mistake and don't act like there's some ulterior motive here.
Here's the full table from my Excel sheet, which other than the SSD names was manually copied or input. Meaning, I had to go through every past review of each of the more than 100 SSDs to check the controller, NAND type, and PCIe version and input that into the table, plus pulling the review URL. It took quite a few hours. You're welcome.
I didn't say anything about you having an ulterior motive. I pointed out the strange thing about the UD90 2230, and it turned out that you wrote "QLC" there just by guessing, instead checking the serial number on the chip or asking SP, or just putting "?" in there. Although SP state very clearly that the 2 TB variant has 1200 TBW - so who should I trust more now - SP or you?
That combined with the mistake about the MP600 mini too. Very fishy review.
I pointed out the strange thing about the UD90 2230, and it turned out that you wrote "QLC" there just by guessing, instead checking the serial number on the chip or asking SP, or just putting "?" in there. Although SP state very clearly that the 2 TB variant has 1200 TBW - so who should I trust more now - SP or you?
He gave you the serial number off the NAND on their review part, and while Micron doesn't have their QLC catalog online (that I could find) you'll find two SSD reviews here at Tom's, one on TPU and two entries in the TPU SSD database citing that serial number as Micron 176 layer QLC NAND. If you look up the full length UD 90 on TPU's SSD database you'll find 3 different NAND types (you could also check reddit etc for users citing they got QLC) used and chances are it's a similar situation for the 2230. Just because they're willing to provide a warranty for 1200 TBW doesn't mean you're guaranteed TLC NAND.
I didn't say anything about you having an ulterior motive. I pointed out the strange thing about the UD90 2230, and it turned out that you wrote "QLC" there just by guessing, instead checking the serial number on the chip or asking SP, or just putting "?" in there. Although SP state very clearly that the 2 TB variant has 1200 TBW - so who should I trust more now - SP or you?
That combined with the mistake about the MP600 mini too. Very fishy review.
Did you not read? Seriously, it is a QLC chip. I showed you the write saturation chart from the upcoming review to prove this point. This article is not a review; it's a performance hierarchy using all of the collected tests from the past 2~3 years. Let me put it in images, though.
TL;DR:
ICCVG96AZA = QLC
ICCIG94AYA = TLC
The first bit can vary by capacity, but the AYA/AZA appears to be the determining factor for Micron on whether a chip uses TLC or QLC.
Silicon Power UD90 2TB:
Inland QN446 2TB:
Addlink S91 2TB:
WD Black SN770M 2TB:
Teamgroup MP44S 1TB (also QLC, but different capacity):
Corsair MP600 Mini 1TB (note the difference in the NAND, as this is TLC as you correctly pointed out):
Sabrent Rocket Q4 2230 2TB (uses the same Phison E21 controller):
Are you aware that the TBW (or PBW) number can be affected by how much spare capacity the NAND chips have? Maybe the QLC chips have enough extra cells that they can offer the same endurance.
Did you not read? Seriously, it is a QLC chip. I showed you the write saturation chart from the upcoming review to prove this point. This article is not a review; it's a performance hierarchy using all of the collected tests from the past 2~3 years. Let me put it in images, though.
TL;DR:
ICCVG96AZA = QLC
ICCIG94AYA = TLC
The first bit can vary by capacity, but the AYA/AZA appears to be the determining factor for Micron on whether a chip uses TLC or QLC.
Maybe you didn't read or understand my comment. Although the SN770M is maybe 3x faster than the UD90 when the SLC cache is full, it's still dog-slow, even if it is touted with the super-duper TLC memory.
So, inferring chip types for tiny 2230 SSDs from performance charts from an upcoming review doesn't impress me that much, especially since I already saw the exact same graph in your colleague's review from 9 days ago:https://www.tomshardware.com/reviews/inland-qn446-2230-ssd-review/2 (image 5/6) .
But you have showed now pictures of the various similar SSDs with the serial numbers, so it's now clear for me. Thank you.
I believe the when we reviewed them, Teamgroup only had A440 Pro ]called "Pro" on the 1TB/2TB variants, and only called "Pro Special" on the 4TB model. So I guess we need "Special" on our 4TB model, yes. I'll go fix that. Thanks for the heads up.
The review for the 990 pro is off and not true. I get 7495 for seq reads and 6990 for my seq writes. Also the random read iops for the 4tb 990 pro is 1,600,000 not 45k. And the random write iops is 1,550,000.
My 990 pro 4tb that I benchmarked with crystal disk mark gets way better scores then what this review posted. Not sure who did this benchmark review but you need to go back to hardware testing 101 and get your cert.
The review for the 990 pro is off and not true. I get 7495 for seq reads and 6990 for my seq writes. Also the random read iops for the 4tb 990 pro is 1,600,000 not 45k. And the random write iops is 1,550,000.
My 990 pro 4tb that I benchmarked with crystal disk mark gets way better scores then what this review posted. Not sure who did this benchmark review but you need to go back to hardware testing 101 and get your cert.
And you might need to go back to reading comprehension class, as the random IOPS is for QD1, not QD32 or QD256 or whatever. Different test PCs, different test settings, lots of variables that can easily account for a ~5% difference in score.
Is there a test anywhere that shows the performance of the drives in a PCIe 3 M.2 slot?
Wondering how some of the lower performing 4.0 SSDs would do against the faster 3.0 ones in a system with a 3.0 slot.... The reason for this is there are a few of these with low power consumption nearly at the level of the SK Hynix P31 Gold, with faster performance on that chart AND they are available for significantly cheaper than the P31... but I am assuming these are tested in a 4.0 slot and I'm wondering how they'd do when compared apples-to-apples in a system that only supports 3.0, as a lot of laptops still do... would that negate the performance of the 4.0 drive to the limit of the BUS, or neuter it altogether so the P31 was still faster, etc?
Is there a test anywhere that shows the performance of the drives in a PCIe 3 M.2 slot?
Wondering how some of the lower performing 4.0 SSDs would do against the faster 3.0 ones in a system with a 3.0 slot.... The reason for this is there are a few of these with low power consumption nearly at the level of the SK Hynix P31 Gold, with faster performance on that chart AND they are available for significantly cheaper than the P31... but I am assuming these are tested in a 4.0 slot and I'm wondering how they'd do when compared apples-to-apples in a system that only supports 3.0, as a lot of laptops still do... would that negate the performance of the 4.0 drive to the limit of the BUS, or neuter it altogether so the P31 was still faster, etc?
While I've not seen any reviewers go over this specifically in what little first hand experience I have the only difference between PCIe 4.0 and PCIe 3.0 performance has been in sustained R/W. Any drive improvements outside of that are maintained (my P41 is much better performance wise than P31 even outside of sustained workloads) so long as it would not exceed the available bandwidth.
It's a slippery slope. If we do QD1 random reads and writes as two separate lines... then shouldn't we do the same for sequential reads and writes? And what about all the other tests we already run?
We decided to focus just on these aggregate read+write metrics for the hierarchy, and the final column (of specs) links to the individual drive reviews where you can see all of the other tests. Not a perfect world, but our tables really look poor if you start adding too many columns (IMO).
Hi Jarred and team,
I’m about to upgrade the SSD in my laptop. (ASUS ROG Strix)
I read your great reviews, on all drives. I decided to go with a 2TB.
So I looked up all the favorites individually.
Can you help me with one thing.
In the ranking of the best SSDs, recently updated, you say the best 2TB SSD is the Corsair MP600 Elite 2TB, far ahead of the T500.
Far worse than the T500, but on the "
We've tested over 100 different SSDs over the past few years, and here's how they stack up." page it's #1 ?
But here
For me the most important thing is fast access times, when opening lots of applications and a second place would be, due to my laptop, energy efficiency.
With which one should I go ?
Corsair MP600 Elite 2TB or T500 2 TB
The prices are the same for me.
If desktop PC, does the motherboard have a heatsink for that slot? If not, are you comparing prices of the heatsink version or non-heatsink? There are some fine aftermarket SSD heatsinks, and they're generally not very expensive, but you should be sure to consider that.
If you care about performance, make sure any SSD you put in a desktop has a heatsink. For laptops, try to prefer the non-heatsink version that has the lowest average temperatures or power consumption, as that will reduce the amount of thermal throttling.
"We recorded a thermal image of the running SSD as it was completing the write test. The surface temperature of the heatsink 101°C, which very closely matches what the onboard thermal reporting shows."
Facebook
Twitter
Instagram