ADATA Swordfish 1 TB M.2 NVMe SSD

[W1zzard]

At just $115 for the 1 TB version, it'll be hard to find a more affordable M.2 NVMe drive than the ADATA Swordfish. Despite the low price, performance is still very decent; there's now very little reason to buy a SATA drive instead of an NVMe SSD.

Show full review

 

[trparky]

To reduce manufacturing cost, ADATA isn't including a DRAM cache chip on the Swordfish. DRAM on an SSD is used as fast temporary storage for the drive's internal mapping tables, which translate between physical disk addresses (as seen by the OS) and the actual location of where the data is stored in the flash chips: "which chip, at which location." Using DRAM has a speed advantage as it operates much faster than flash, but it's a cost/performance trade off. A 1 TB SSD typically uses 1 GB of DRAM, which costs a few dollars. If you can save that, you'll be able to position your drive more aggressively, leading to more sales, or you'll enjoy higher margins. The disadvantage is that random IOPS performance is lower because the controller has to access the flash more often to process the mapping tables. For nearly all consumer workloads, random write performance over a large area doesn't really matter as our benchmarks show. Consumer workloads are very read-heavy anyway, and if writes happen, they are localized over a relatively small area, which is exactly why DRAM-less SSDs were invented.

Correct me if I'm wrong but isn't there a standard where the SSD could use a portion of system RAM to serve the same purpose of onboard DRAM?

 

[TheLostSwede]

Correct me if I'm wrong but isn't there a standard where the SSD could use a portion of system RAM to serve the same purpose of onboard DRAM?

Yes, Host Memory Buffer. Doesn't seem to have been a very popular feature though.

The Toshiba RC100 SSD Review: Tiny Drive In A Big Market

www.anandtech.com

 

[trparky]

Yes, Host Memory Buffer.

Oh.

Doesn't seem to have been a very popular feature though.

Why is that?

 

[TheLostSwede]

Oh.

Why is that?

I guess it's not popular because of a couple of reasons.
As per above, a certain amount of system RAM has to be always allocated as the HMB, how much seems to depend on the controller maker, but if we assume a like for like comparison, a 1TB SSD normally has 1GB of RAM for storing the mapping tables.
If you have a cheap notebook, they normally still ship with only 4GB of RAM, of which some is already shared with the integrated GPU.
So if we assume a cheap notebook ships with a 256GB SSD, you lose another 256MB of RAM.
So the user would only be able to access something like 3.5GB at best, which might reduce the overall system performance.
On top of that, if there's a power cut (not likely with a notebook unless you're low on power), you have a bigger chance of losing data vs. a normal SSD, even though most consumer SSD's don't tend to have any kind of power failure protection, there's at least a shorter path to the DRAM and for the data in the DRAM to be written to the drive.
I'm sure there are more reasons, but as per the link above, it seems to be mainly drives targeting system integrators (big or small) that use HMB. I have not seen any off-the-shelf drives that supports it.
Also, judging by the testing done by Anandtech, it seems to have fairly limited performance improvement, considering the trade-offs.
I have a feeling that the pseudo SLC cache is offering better performance overall, as seen in this review.

 

[trparky]

If you have a cheap notebook, they normally still ship with only 4GB of RAM, of which some is already shared with the integrated GPU.
So if we assume a cheap notebook ships with a 256GB SSD, you lose another 256MB of RAM.
So the user would only be able to access something like 3.5GB at best, which might reduce the overall system performance.

Even more reason not to buy a cheap notebook but you knew that was coming.

 

[TinyPilot]

Ah... Come on, Crucial P1 for almost same price is easy better option! What the heck is going on?

 

[TheLostSwede]

Ah... Come on, Crucial P1 for almost same price is easy better option! What the heck is going on?

You registered just to say that?
Anyhow, there are a few reasons why, which you'd known if you'd read the P1 review here. Mainly bad pSLC cache implementation (in comparison) and thermal throttling issues. Also, not sure the value graphs get their pricing update, you'd have to ask the boss @W1zzard about that.

Crucial P1 NVMe M.2 SSD 1 TB Review

Crucial has finally released their first M.2 NVMe solid-state drive. The Crucial P1 SSD is built using QLC NAND flash, which is the next generation in flash memory, promising lower cost per GB than previous technologies. In our testing, we see good results that are competitive with popular TLC...

www.techpowerup.com

 

[ironwolf]

So among SSDs with a heat spreader pre-attached, this is basically the best bang for the buck?

 

[Hossein Almet]

If u intent to use the drive for games only, SATA SSD is still a better option, because with SATA SSD u don't need to worry about throthing.

 

[W1zzard]

So among SSDs with a heat spreader pre-attached, this is basically the best bang for the buck?

Fixed it for you

If u intent to use the drive for games only, SATA SSD is still a better option, because with SATA SSD u don't need to worry about throthing.

By that logic you should buy a VW Beetle instead of a sports car, because you can't get a ticket with the Beetle.
Most NVMe SSDs can write continuously at 1 GB/s without throttling, 2x what SATA can achieve

Also, not sure the value graphs get their pricing update

Prices in old reviews do not get updated, this review has the latest prices

 

[TinyPilot]

You registered just to say that?
Anyhow, there are a few reasons why, which you'd known if you'd read the P1 review here. Mainly bad pSLC cache implementation (in comparison) and thermal throttling issues. Also, not sure the value graphs get their pricing update, you'd have to ask the boss @W1zzard about that.

Crucial P1 NVMe M.2 SSD 1 TB Review

Crucial has finally released their first M.2 NVMe solid-state drive. The Crucial P1 SSD is built using QLC NAND flash, which is the next generation in flash memory, promising lower cost per GB than previous technologies. In our testing, we see good results that are competitive with popular TLC...

www.techpowerup.com

I have P1 1tb.
No thermal throttling, but I'm using heatsink, that comes with my motherboard.
Raw read/write speeds are quite as written on the package, but it comes with software, which has ability to switch some area of memory to more productive one. And that is a beast. Linear read/write speeds are close to 9GBps, 4k blocks are coming to around 500MBps.
If you buy electric bicycle and turn off electric part, that means you are dumb, not manufacturer. Same with these memories. Some come with software which add great performance through caching mechanism, but not everybody is turning it on.
I've heard arguments from people, that buy Samsung Evo series memory. They tell their memory is faster, yet, nobody is crazy enough to turn off caching mechanism.

So, let's be honest, if something has caching options, it has to be turned on.
At the same time, I have co-worker, who is certain he's saving hardware by turning all imaginable caching off. I wish he would be able to turn off CPU L1 &L2 cache.

 

[Hossein Almet]

Fixed it for you


By that logic you should buy a VW Beetle instead of a sports car, because you can't get a ticket with the Beetle.
Most NVMe SSDs can write continuously at 1 GB/s without throttling, 2x what SATA can achieve


Prices in old reviews do not get updated, this review has the latest prices

Actually, I quite like a VW Beetle It's a pity I couldn't afford it.

 

[BSim500]

By that logic you should buy a VW Beetle instead of a sports car, because you can't get a ticket with the Beetle. Most NVMe SSDs can write continuously at 1 GB/s without throttling, 2x what SATA can achieve

Except in this case (BF V: 24.8s Swordfish vs 24.3s MX500 / Far Cry 5: 17.3s Swordfish vs 16.2s 850 EVO / SotTR: 20.7s Swordfish vs 20.5s 850 EVO / Win10 Boot: 18.1s Swordfish vs 17.8s MX500, etc) ends up more like building a shiny new 2020 sports car only to have it still lose to two 2 & 6 year old VW Beetle's when testing on real-life roads that involve more than just straight line drag-strip peaks...

Personally I picked up a 2TB MX500 the other day for £160. Absolute zero regrets vs "losing out" on £250-£460 NVMe's of same capacity (or £170 for half the capacity). And every single SSD review including TPU's, I now completely skip all pages of synthetics and jump straight to Page 8 where the real-world stuff starts. I'll give you and TPU as a whole a big +1 for doing real world stuff though. A lot of reviews infomercials on other sites do little more than copy / paste CrystalDiskMark, IOMeter, etc, which is like testing 8 vs 4 core CPU's by declaring "a 3700x will get exactly double the fps of a 3300X" on the back of Cinebench scores - and then not actually testing any games that end up looking like this... So I do actually love these TPU SSD reviews for avoiding that nonsense.

 

[W1zzard]

So I do actually love these TPU SSD reviews for avoiding that nonsense.

Thanks! Working on PCIe4 test rig right now, adding more real-life tests, too

 

[Hossein Almet]

Except in this case (BF V: 24.8s Swordfish vs 24.3s MX500 / Far Cry 5: 17.3s Swordfish vs 16.2s 850 EVO / SotTR: 20.7s Swordfish vs 20.5s 850 EVO / Win10 Boot: 18.1s Swordfish vs 17.8s MX500, etc) ends up more like building a shiny new 2020 sports car only to have it still lose to two 2 & 6 year old VW Beetle's when testing on real-life roads that involve more than just straight line drag-strip peaks...

Personally I picked up a 2TB MX500 the other day for £160. Absolute zero regrets vs "losing out" on £250-£460 NVMe's of same capacity (or £170 for half the capacity). And every single SSD review including TPU's, I now completely skip all pages of synthetics and jump straight to Page 8 where the real-world stuff starts. I'll give you and TPU as a whole a big +1 for doing real world stuff though. A lot of reviews infomercials on other sites do little more than copy / paste CrystalDiskMark, IOMeter, etc, which is like testing 8 vs 4 core CPU's by declaring "a 3700x will get exactly double the fps of a 3300X" on the back of Cinebench scores - and then not actually testing any games that end up looking like this... So I do actually love these TPU SSD reviews for avoiding that nonsense.

My 2TB MX500 was delivered this morning, for A$349.

 

[TheLostSwede]

Prices in old reviews do not get updated, this review has the latest prices

The 1TB P1 seems to be $105 now, if you buy it directly from Crucial.
So it's technically cheaper, but I'd say it's 50/50 which drive is "better".

I have P1 1tb.
No thermal throttling, but I'm using heatsink, that comes with my motherboard.
Raw read/write speeds are quite as written on the package, but it comes with software, which has ability to switch some area of memory to more productive one. And that is a beast. Linear read/write speeds are close to 9GBps, 4k blocks are coming to around 500MBps.
If you buy electric bicycle and turn off electric part, that means you are dumb, not manufacturer. Same with these memories. Some come with software which add great performance through caching mechanism, but not everybody is turning it on.
I've heard arguments from people, that buy Samsung Evo series memory. They tell their memory is faster, yet, nobody is crazy enough to turn off caching mechanism.

So, let's be honest, if something has caching options, it has to be turned on.
At the same time, I have co-worker, who is certain he's saving hardware by turning all imaginable caching off. I wish he would be able to turn off CPU L1 &L2 cache.

A beast? It's a budget SSD with meh performance imho. You're using some kind of RAM cache to hit those speeds, which is cheating and it's not something you can compare with other drives, as it's not native drive performance.

And no, it doesn't have to be turned on at all, as you risk your data using those kind of features. In theory, someone can do the same with a good old spinning rust drive, do they then have an SSD? I don't think so. A RAM cache is NOT the same as turning of caching, you clearly don't seem to understand the difference, nor what you've enabled. I suggest you do some research before you come on here and try to show off and tell people what to do.

 

[Twistedtyrant66]

The asenno 1tb is 99 u.s also. And in Canada I will pay 170 for the adata swordfish and then 70 for import fees. And the asenno I only pay 130. It's the cheapest I found. And the speeds in the 1 TB are not bad.

 

[ThrashZone]

Hi,
Besides price per gb
Read/ write look as anticlimactic as you can get :/

Sequential R/W performance (max.)​

Up to 1800/1200MB/s​

 

[Maxx]

Likely TLC as IOPS increase up to 1TB (so there's more interleaving as you go) and 900 MB/s at 250GB requires 16-way interleaving which is impossible with current QLC (IMFT's being 4-plane but at most 2 dies at that capacity). Likely candidate is Micron's B27B which is 512Gb/four-plane as I've seen on the Falcon which has the same write perf. at that capacity.

 

[HEisUS20]

Yeah. I am still gonna buy crucial p1 1tb.

 

[TinyPilot]

You registered just to say that?
Anyhow, there are a few reasons why, which you'd known if you'd read the P1 review here. Mainly bad pSLC cache implementation (in comparison) and thermal throttling issues. Also, not sure the value graphs get their pricing update, you'd have to ask the boss @W1zzard about that.

Crucial P1 NVMe M.2 SSD 1 TB Review

Crucial has finally released their first M.2 NVMe solid-state drive. The Crucial P1 SSD is built using QLC NAND flash, which is the next generation in flash memory, promising lower cost per GB than previous technologies. In our testing, we see good results that are competitive with popular TLC...

www.techpowerup.com

I replied to you with some numbers, but reply was never posted. I guess, moderator doesn't like my answer, based on numbers.

 

[TheLostSwede]

I replied to you with some numbers, but reply was never posted. I guess, moderator doesn't like my answer, based on numbers.

You're still confused. Using RAM cache is cheating. It doesn't show the capabilities of your drive, it shows the capabilities of you RAM. Clearly you don't get that.

 

[Space Lynx]

this is my AS SSD score on my new WD Black SN750 1tb drive... does this all look good to you? @TheLostSwede i paid $129.99 and that included tax. i am not good with storage and that 0.072 seems a little high not sure why that is, sometimes it even says 0.125 ish range instead of 0.072

 

[TheLostSwede]

this is my AS SSD score on my new WD Black SN750 1tb drive... does this all look good to you? @TheLostSwede i paid $129.99 and that included tax. i am not good with storage and that 0.072 seems a little high not sure why that is, sometimes it even says 0.125 ish range instead of 0.072

I don't normally use AS SSD, so I couldn't tell you. I have the older WD Black so yours should be faster.
Not sure how the 4K scores compared to CrystalDiskMark, but it seems a tad slow, but a lot of that depends on the host system as well.

@lynx29 CDM results from my drive below. Note the version of CDM though, as they've moved on to 7.x now, which is a bit different in how it tests. This is my OS drives as well.