Review Samsung 990 Pro SSD Review: The Return of the King

[Neilbob]

Typical consumers do not need PCI-E Gen 5 anything, and won't for quite some time. Anyone who says otherwise is deluding themselves.

This being PCI-E5 would only result in it being more pricey for little to no actual performance benefit outside of what 3DMark and PCMark say. I suppose that's why there are never any actual real-world tests anymore, because they would show a difference that is far too small to be discernible, and would perfectly demonstrate that a cheapo SSD costing less than 100 would be perfectly sufficient for the vast majority of people.

Only have to say these things cos these reviews and stories bring out my grumpy old codger 👨‍🦳

 

[Makaveli]

It's overwhelming to see all of the Gen 5 SSDs that were suppose to be available by now. /s

Meh won't be much of an improvement to be honest other than double the sequential read and write speeds everything else will stay the same as a PCIe 4.0 drive.

 

[bit_user]

This being PCI-E5 would only result in it being more pricey for little to no actual performance benefit

Also, hotter.

PCIe 5.0 runs at double the clock speed of PCIe 4.0. And after that, they hit a wall - PCIe 6.0 had to use PAM4 signalling, while keeping the same clock speed as 5.0.

 

[bit_user]

Meh won't be much of an improvement to be honest other than double the sequential read and write speeds everything else will stay the same as a PCIe 4.0 drive.

Even sequential doesn't double, because they're already pushing the limits of the NAND chips, there.

It was a similar story with 4.0. It took more than a year after the launch of Zen 2 for the first PCIe 4.0 drives to hit the market that exceeded PCIe 3.0 speeds in any real sense.

 

[tracker1]

guess strike this up to being yet another "pro" branded product that is in no way intended for professionals.
It's probably time for the dictionary curators of the world to add "overpriced consumer electronics" as an official definition for "pro".

I'm not sure that I agree. While I wouldn't get the RGB version, people should be able to if they want. My interest is the best class iops and getting fatter compile times or database throughout in local dev... Others will have their own reasons. And yeah for some it's more vanity than need. I don't mind the vanity buyers as it lets products cross the line from niche and makes pricing more competitive assuming production scales well.

 

[colindog]

I was expecting this to be a PCI-E 5.0 drive, but turns out to be yet another PCI-E 4.0 drive. Which leads me to wonder why Samsung even bothered to make small incremental sequential speed improvements, which most people won't even realize in their day to day use. I mean even if you are transferring huge files all the time, the difference in the sequential write speed is not going to make a huge difference. It is more likely to run into thermal throttling if there is insufficient cooling to allow it to sustain its write speed. So while the title says the return of the "king", its barely faster than the fastest PCI-E 4.0 SSD out there. In my opinion, there is nothing special about it.

Was totally waiting for this to be Gen5 for my new Raptor Lake/4000 build. super disappointing.

 

[DataMeister]

Why do all the Gen4 drives seem to have such low endurance? The Western Digital Red SN700 has a 2000 TBW rating for the 1TB drive, but is only a Gen3 product.

 

[bit_user]

Why do all the Gen4 drives seem to have such low endurance? The Western Digital Red SN700 has a 2000 TBW rating for the 1TB drive, but is only a Gen3 product.

Are the ones you're looking at QLC, perhaps? It typically has lower endurance than TLC.

 

[t-dome]

Why does the 980 Pro suddenly have lower write speeds than in it's own review? I've also seen different values in other benchmarks in your reviews, nothing can be compared to each other anymore.
Are you just sloppy and switch around values in the Excel tables by mistake?

 

[bit_user]

Why does the 980 Pro suddenly have lower write speeds than in it's own review? I've also seen different values in other benchmarks in your reviews, nothing can be compared to each other anymore.

It's a good question, but keep in mind that the test system isn't identical. Even if they use the same host machine (did you check?), the OS receives updates that could enable additional CPU mitigations, which tend to affect things like I/O. Even the benchmark software could've received updates that affect results.

So, the best testing methodology is to re-test all drives on the same host system, for each review. It's a separate question why the results differ between reviews, but doesn't invalidate their testing.

Hopefully, additional light can be shed on this. However, sloppiness or "noisy data" aren't the only possible explanations.

 

[t-dome]

It's a good question, but keep in mind that the test system isn't identical. Even if they use the same host machine (did you check?), the OS receives updates that could enable additional CPU mitigations, which tend to affect things like I/O. Even the benchmark software could've received updates that affect results.

So, the best testing methodology is to re-test all drives on the same host system, for each review. It's a separate question why the results differ between reviews, but doesn't invalidate their testing.

Hopefully, additional light can be shed on this. However, sloppiness or "noisy data" aren't the only possible explanations.

I have searched only a bit but I couldn't find 2 identical test systems for different reviews. Most of the reviews don't even list the test bench.

Another example: SK Hynix Gold P31 review

Here, the 2 TB version was tested (presumably) later, but the test bench system is listed only on its page, not on the original review with the 1 TB.
In the verdict, they say that the 2 TB is only a little slower that the 1 TB. In the Final Fantasy test, it is 3 seconds slower (12 vs 9 s), all the other drives have similar huge differences in the same bechmark.
Really??? Did they suddenly use a Ryzen 19999X with 12 GHz from the year 2030 in the 1 TB test? There is 100% no way that the difference for the P31 (and the other drives) can be that big, this is more than suspicious.

 

[gamo62]

Not really worth the upgrade if you already have a 980 Pro.

 

[bit_user]

I have searched only a bit but I couldn't find 2 identical test systems for different reviews. Most of the reviews don't even list the test bench.

Another example: SK Hynix Gold P31 review

Here, the 2 TB version was tested (presumably) later, but the test bench system is listed only on its page, not on the original review with the 1 TB.
In the verdict, they say that the 2 TB is only a little slower that the 1 TB. In the Final Fantasy test, it is 3 seconds slower (12 vs 9 s), all the other drives have similar huge differences in the same bechmark.
Really??? Did they suddenly use a Ryzen 19999X with 12 GHz from the year 2030 in the 1 TB test? There is 100% no way that the difference for the P31 (and the other drives) can be that big, this is more than suspicious.

Changing antivirus or switching CPU mitigations on/off could easily account for that amount of difference.

 

[seanwebster]

Why does the 980 Pro suddenly have lower write speeds than in it's own review? I've also seen different values in other benchmarks in your reviews, nothing can be compared to each other anymore.
Are you just sloppy and switch around values in the Excel tables by mistake?

A few reasons. Differences you see between scores in different reviews are due to either a new test system being used or different data/benchmark/testing methodology being compared (change from iometer to CDM for synthetic metrics for example). You need to see which test platform is being used if you are comparing older reviews and test platforms.

We just retested 100 drives these past months on a Z690 & i9 12900K setup, so that is what we show on current and future reviews. We will also have a few more comparison articles with benchmarks from all our testing. Older reviews were on a Z590 & 11900K setup.

Additionally, we only show 1MB block size results for our sustained sequential write speeds in more recent times while we used to show 128KB metrics back in my early reviews. In the Samsung 980 Pro's review we threw in both the 128KB (since that is how Samsung rated it) and 1MB block size data, but in our reviews, we mainly compare 1MB block results nowadays.

I have searched only a bit but I couldn't find 2 identical test systems for different reviews. Most of the reviews don't even list the test bench.

Another example: SK Hynix Gold P31 review

Here, the 2 TB version was tested (presumably) later, but the test bench system is listed only on its page, not on the original review with the 1 TB.
In the verdict, they say that the 2 TB is only a little slower that the 1 TB. In the Final Fantasy test, it is 3 seconds slower (12 vs 9 s), all the other drives have similar huge differences in the same bechmark.
Really??? Did they suddenly use a Ryzen 19999X with 12 GHz from the year 2030 in the 1 TB test? There is 100% no way that the difference for the P31 (and the other drives) can be that big, this is more than suspicious.

In this case, the 500GB and 1TB models were tested on an X570 3600X setup (back when we didn't list the test platform on my old reviews) and the 2TB model was released as we were transitioning to the Z590 rig (when we started to list the test platform).

Our test system is usually configured with locked clocks, although for Alder Lake we decided to leave the default Asus UEFI settings for boosting, and once the OS is installed, services are locked down including updates to prevent interfering with benchmarking results. Normally, the system does not see software updates after it is configured and used that way until the next platform releases. If updates or software changes are performed for specific reasons, benchmark data is cross compared to ensure the updates do not influence benchmark results. If they do, the original configured system imaged is restored.

 

[b1k3rdude]

The Return of the King : Read more

Isn't the title completely misleading, was is a sponsored by Samsung article..?..? in your own benchmarks the drive no better than a bunch of the other drives in this list. In fact when we dig down we can see the Rocket4/KC3000 are both better for sustained reads and writes.

What is the point of posting GB/s per watt figures? imho is utterly pointless when we are in the 0-5W range, I can guarentee absolutley no-one cares (am Uk based & very power aware, electricity is £0.35/Kwh atm). If and when we start seeing double-digit power useage, then at that point it will start to matter.

And then there is the COMPLETE lack of random 4k QD1 benchmarks, for all the drives. For desktop users random read/writes are an important metric when decideing what SSD to upgrade too. @Admin , can you post screenshots from Crystaldiskmark for the 990, Rocket4 & KC3000 please!? As Im betting its not much better than my Samsung 960 pro.

 

[seanwebster]

Isn't the title completely misleading, was is a sponsored by Samsung article..?..? in your own benchmarks the drive no better than a bunch of the other drives in this list. In fact when we dig down we can see the Rocket4/KC3000 are both better for sustained reads and writes.

What is the point of posting GB/s per watt figures? imho is utterly pointless when we are in the 0-5W range, I can guarentee absolutley no-one cares (am Uk based & very power aware, electricity is £0.35/Kwh atm). If and when we start seeing double-digit power useage, then at that point it will start to matter.

And then there is the COMPLETE lack of random 4k QD1 benchmarks, for all the drives. For desktop users random read/writes are an important metric when decideing what SSD to upgrade too. @Admin , can you post screenshots from Crystaldiskmark for the 990, Rocket4 & KC3000 please!? As Im betting its not much better than my Samsung 960 pro.

Our reviews are NOT sponsored by anyone. The best is not simply decided by what drive delivers the highest sequential. In general user experience, 4K random and fast cache recovery matter the most, but there are still many other aspects that weigh into evalutation.

Efficiency metrics matter for laptop users upgrading their storage. Some M.2 drives can consume up to 11W sustained under load... a lot harder to cool than 5W. Differences in battery life between SSDs can be hours. Samsung's 980 Pro 2TB enables 3 hours longer battery life than the Seagate FireCuda 530 2TB in the exact same laptop, apples to apples.

4K random metrics are shown in the ATTO/CDM section. You just have to click through the gallery to view them.

Ex. We include both IOPS and latency metrics as shown below.

 

[b1k3rdude]

• In general user experience, 4K random and fast cache recovery matter the most,
• Efficiency metrics matter for laptop users upgrading their storage. Some M.2 drives can consume up to 11W sustained under load... a lot harder to cool than 5W.
• Ex. We include both IOPS and latency metrics as shown below.

• Yep, and inline with what I said. Cache recovery is new thoough, as my main drive dosent have this issue (960pro)
• Thats a fair comment, but all drive in this article are all below 5W.
• IOPS can be missleading, specially if the block size is not mentioned (I prefer direct CDM screenshots where possible) -
  • So assuming you guys used CDM defaults (4KiB) 25133 IOPS gives 100MB/ reads which is much better than my 960 pro at 64MB/s(when 960 was new). But I beleive my 960 dosent fall of a cliff like the 990 does after 100GB.
  • No 4K QD1 Write speed numbers
  • Latency has no context or meaning for me. And on the above graph bar the crucual, all the drives are with in a few micro seconds of each other.

Ive attached a screenshot of my 960 when new(see below), this is what I compare other SSD to. Ive had the drive since 2017 and the 4k QD1 r/w are now 57MB/s and 205/MB/s.

And checking out KitGuru's review its actually only slightly faster than my 960 on reads @82MB/s but slower on writes at 230MB/s - https://www.kitguru.net/components/ssd-drives/simon-crisp/samsung-990-pro-2tb-review/4/

 

[seanwebster]

• Yep, and inline with what I said. Cache recovery is new thoough, as my main drive dosent have this issue (960pro)
• Thats a fair comment, but all drive in this article are all below 5W.
• IOPS can be missleading, specially if the block size is not mentioned (I prefer direct CDM screenshots where possible) -
  • So assuming you guys used CDM defaults (4KiB) 25133 IOPS gives 100MB/ reads which is much better than my 960 pro at 64MB/s(when 960 was new). But I beleive my 960 dosent fall of a cliff like the 990 does after 100GB.
  • No 4K QD1 Write speed numbers
  • Latency has no context or meaning for me. And on the above graph bar the crucual, all the drives are with in a few micro seconds of each other.

Ive attached a screenshot of my 960 when new(see below), this is what I compare other SSD to. Ive had the drive since 2017 and the 4k QD1 r/w are now 57MB/s and 205/MB/s.

And checking out KitGuru's review its actually only slightly faster than my 960 on reads @82MB/s but slower on writes at 230MB/s - https://www.kitguru.net/components/ssd-drives/simon-crisp/samsung-990-pro-2tb-review/4/

Power and heat depend on workload, system, and how the system is configured. Power consumption is higher in sustained write workloads. Some drives hit 8-9W in our copy test in this review, which in some laptops will result in the drive going offline because on some, the M.2 slot can only provide up to 3.3v at 2.5A, not 3A. I've had it happen with a few as well as thermal-throttling shutdowns. Efficiency really matters most for mobile platform users. We are working on building a better power database that includes CDM runs to show more in-depth analysis with corner-case loads, too.

Read through the review and look at the gallery of benchmarks before jumping to so many conclusions, please. The data is there if you look. Block sizes and queue depths are listed. In 4K random tests, 4K is the block size. For sequential metrics, 1MB is the block size. 4K write numbers are in the review in the exact same gallery as stated in the previous reply directly after the 4K random read metrics.

Latency is simple - lower = better/faster response to requests. Lowest = most responsive in daily use.

You can compare that benchmark to any you like, but it will not be accurate. Performance varies due to many variables across different systems and software packages. For example, going from an X570 to a Z690 test platform results in a massive random performance uplift, but slower sequential benchmark results.

 

[TR909]

Not the "outstanding SSD". Too much hype in the article's title.
KC3000 remains a very strong option in terms of Value for Money.

 

[t-dome]

Ofcourse is the Kingston is better value, I think the Fury Renegade is even better. They cost less than half than the 990 Pro, and in practically all the cases you will see no performance difference. The SN850x is even faster than the 990 Pro.
But for laptops the Samsung is the much better choice, here probably even the 980 Pro because of the price.

 

[mark204]

This review is incomplete! The magician software is necessary for performance optimization on each system (I’ve been using Samsung SSD’s since 840 Evos, built dozens of systems with them). The magician software has not worked correctly, thus the drives do not run optimized, since 980 pros. I currently have 2 systems: a Dell Precision 7560 with a 980 pro and an MSI B560M PRO-VDH ProSeries system I built yesterday with a 990 pro. Magician’s functions do not work on either; according to crystal disk mark both drives are significantly underperforming in reads and writes. Samsung support responses: hmm first we are hearing about it...I seriously doubt that. As you can see from this review, if you are unable to turn on full power mode (which I am not) then the drives get beat in several tests and therefore not worth the $50+ more than better-performing drives.
You are better off purchasing a drive that does not rely on optimization software to run correctly; from now on I will only go with drives that are significantly less expensive and more reliable.

 

[USAFRet]

This review is incomplete! The magician software is necessary for performance optimization on each system (I’ve been using Samsung SSD’s since 840 Evos, built dozens of systems with them). The magician software has not worked correctly, thus the drives do not run optimized, since 980 pros. I currently have 2 systems: a Dell Precision 7560 with a 980 pro and an MSI B560M PRO-VDH ProSeries system I built yesterday with a 990 pro. Magician’s functions do not work on either; according to crystal disk mark both drives are significantly underperforming in reads and writes. Samsung support responses: hmm first we are hearing about it...I seriously doubt that. As you can see from this review, if you are unable to turn on full power mode (which I am not) then the drives get beat in several tests and therefore not worth the $50+ more than better-performing drives.
You are better off purchasing a drive that does not rely on optimization software to run correctly; from now on I will only go with drives that are significantly less expensive and more reliable.

I don't know why your Magician does not allow Full power mode.
Possibly you have an out of date version applied?

This is my 980 Pro, before and after enabling 'Full power'

 

[ithemask]

There's one point this review doesn't cover, that Samsung is lying about the hardware encryption support (class 0) of this product. The 990 Pro with the latest firmware (1B2QJXD7) does not support hardware encryption at all, Samsung support simply ignoring tickets and community discussion (just search for "990 pro hardware encrypted not supported")

 

[Deleted member 14196]

I don't know why your Magician does not allow Full power mode.
Possibly you have an out of date version applied?

This is my 980 Pro, before and after enabling 'Full power'

Looks like you don’t need full power mode at all.
Eh I’m done with Samsung

too expensive and not worth it. Look at the last debacle’

 

[bit_user]

Looks like you don’t need full power mode at all.

Here's a pure hypothetical: what if Full Power mode mostly impacts performance once the pseudo-SLC buffer is full?

I think that's at least consistent with the fact that the greatest impact is on 4k writes. In the above data, RND 4k Q32 T1 speeds up by 15.1%, while RD 4k Q1 T1 speeds up by 7.8%.

Once you fill up the SLC buffer, then the drive switches over to doing read-modify-write operations for each small write. And the writes are slower, because they're at full TLC density, rather than fast pseudo-SLC writes.