Crucial T710 2 TB

[Baki]

I don't get it, bunch of SSDs that we don't care for and not WD_black SN8100... And how latest page says it's the fastes when it's clearly not on many tests? I was waiting for this SSD since announcement, and was very happy to see that review is finally here only to end up thinking between Samsung 91000 and Kingston G5..

 

[sysakPL]

Great review, it would be amazing if you could also test the new WD SN8100. Hesitating between it, the Kingston G5 and this Crucial. (Though they're all similar so probably i just should get the cheapest.)

 

[W1zzard]

Great review, it would be amazing if you could also test the new WD SN8100. Hesitating between it, the Kingston G5 and this Crucial. (Though they're all similar so probably i just should get the cheapest.)

Yeah, still waiting for the sample from WD

 

[EsaT]

All gen5 drives run too hot, and don't raise the bar for IOPS or access latency, which seems to be the bottleneck for the vast majority of situations these days.

However, if I was in the market for a Gen5 I would probably be looking at the Kioxia Exceria Plus G4, or something else with Phison's E31 in it. Sure, it's peak sequential writes are lower than the competition at "only" 9GB/s but it's doing that for a little over 4W...

For fixed amount of work, like loading application/game, these drives shouldn't really get that much hotter because of rather short burst nature of load and actually higger effciency under load.
But what sucks is "idle" power consumption starting to be at level where many drives were under load.

With that Kioxia G4 being "good" (or bad) example with over 2W idle draw, which is basically trippled compared to where idle draw used to be:

Kioxia Exceria Plus G4 2 TB Review - Energy-Efficient PCIe Gen 5

The Kioxia Exceria Plus G4 features Phison’s new E31 controller, designed for PCI-Express 5.0 with a focus on power efficiency. This allows for fan-free operation without compromising performance. Pricing is also competitive, it's the most affordable Gen 5 SSD on the market right now.

www.techpowerup.com

Though with power save enabled idle draw becomes very good.
Just have to wonder what kind latencies power save causes.

Anandtech's reviews had measurements for how long it takes for drive to wake up from low power states and there were variations in that...
Like with SM2262 controller being "slow to get up from the bed":

Comparing Two 1TB NVMe Drives with Same NAND, Same Controller: XPG SX8200 Pro vs HP EX950

www.anandtech.com

 

[W1zzard]

For fixed amount of work, like loading application/game, these drives shouldn't really get that much hotter

Exactly, that's why I've added the real-life thermal testing, which confirms that you can run these drives without heatsink just fine during normal usage and they will not throttle

 

[Fluke]

Even then, using HDD's is debatable, too expensive for what they bring to the table.

Right now, I can buy a 24TB HDD for 330€, whereas a 16TB SSD is gonna cost ~1500€.
I see one thing that's too expensive alright - and it's not the HDD.

 

[W1zzard]

Right now, I can buy a 24TB HDD for 330€, whereas a 16TB SSD is gonna cost ~1500€.
I see one thing that's too expensive alright - and it's not the HDD.

But the SSD is almost 100 (!) times faster

 

[Chrispy_]

Right now, I can buy a 24TB HDD for 330€, whereas a 16TB SSD is gonna cost ~1500€.
I see one thing that's too expensive alright - and it's not the HDD.

I can buy a 7.5 tonne van for less than the price of a Porsche GT3 RS, too. It's almost like one is for capacity and one is for speed...

Exactly, that's why I've added the real-life thermal testing, which confirms that you can run these drives without heatsink just fine during normal usage and they will not throttle

My take on Gen5 drives is that if you actually need Gen5, you are after large amounts of very very fast sequential writes that make these drives throttle. A good Gen3 drive can give a very very similar experience in normal bursty usage - unless your workload needs you to write several hundred GB at 12GB/s - and that's precisely when these Gen5 drives drop to Gen3/Gen4 speeds without a massive cooler strapped to them.

Edit:
I just went to look at the throttling data and the heatsinks in question. Thermalright TR-M2 is too big to fit under a GPU so potentially unusable depending on motherboard and layout, but it staved off the throttling for two and half minutes at 12GB/s which is multiple TB of data before it cooks itself and likely to never impact 99.9% of users. The bare drive is cooked in 25 seconds, barely enough time to copy a couple of AAA games before it's running at Gen4 speeds.

Presumably a GPU-compatible M.2 heatsink that most people are likely to be using would provide performance somewhere in between. Maybe 60-90s at 12GB/s which seems like enough for any consumer. If they're regularly reading/writing more than 700GB of data at a time, I don't think you can label them consumers any more

 

[W1zzard]

Maybe 60-90s at 12GB/s which seems like enough for any consumer

That. Also, where is your data coming from at that rate? Isn't your SLC cache full yet?

 

[Chrispy_]

realistically, I

That. Also, where is your data coming from at that rate? Isn't your SLC cache full yet?

Realistically it would be to/from a SAN sitting on a 25Gb switch, or a second SSD that's equally as fast in a slot that also has dedicated PCIe bandwidth; Definitely not consumer territory!

Where's your data coming from when you do throttle-testing? 2.5 minutes is enough to write 2TB to a 2TB drive, so presumably your testing generates random fill or loops a test dataset over and over from RAM?

 

[Tyler-98-W68]

realistically, I

Realistically it would be to/from a SAN sitting on a 25Gb switch, or a second SSD that's equally as fast in a slot that also has dedicated PCIe bandwidth; Definitely not consumer territory!

Where's your data coming from when you do throttle-testing? 2.5 minutes is enough to write 2TB to a 2TB drive, so presumably your testing generates random fill or loops a test dataset over and over from RAM?

You will not transfer 2TB of data is 2.5 minutes because no drive is going to have enough SLC cache to provide those 12gb/s speeds during the whole transfer

 

[Chrispy_]

You will not transfer 2TB of data is 2.5 minutes because no drive is going to have enough SLC cache to provide those 12gb/s speeds during the whole transfer

You're right - not 2TB of unique data, no. Editing an existing dataset in pSLC cache is a different story, I think - and presumably how W1zzards throttling test does it:

12GB/s of writing for 230 straight seconds. That's obviously all going to pSLC, and not exhausting the cache.

 

[W1zzard]

Realistically it would be to/from a SAN sitting on a 25Gb switch, or a second SSD that's equally as fast in a slot that also has dedicated PCIe bandwidth; Definitely not consumer territory!

so only 2.5 GB/s. You can write at that speed all day without temps increasing much. Second SSD: not a lot of boards with 2x Gen 5. Didn't mean to be rude btw, sorry if I came across like that

Where's your data coming from when you do throttle-testing? 2.5 minutes is enough to write 2TB to a 2TB drive, so presumably your testing generates random fill or loops a test dataset over and over from RAM?

Yes, it's generated, fully random

 

[drSeehas]

... Second SSD: not a lot of boards with 2x Gen 5. ...

There are 38 boards:

Motherboards with Socket: AM5, M.2 M-Key PCIe 5.0 x4: from 2x Price Comparison Skinflint UK

compare prices and read user reviews for Motherboards with Socket: AM5, M.2 M-Key PCIe 5.0 x4: from 2x

skinflint.co.uk

 

[Chrispy_]

Didn't mean to be rude btw, sorry if I came across like that

You're good; It did not come across as rude. Facts are facts.

As for dual Gen5 boards, I haven't found many, the last video editing rigs I built used the MSI X870E Tomahawk Wifi not because it's necessarily the best board for the job, but because it was the first one I found that was both in stock and where I could easily find info on PCIe lane layout and sharing. That info is missing from so many options if you're simply browsing for boards!

Even then, their bottleneck is usually getting the raw footage from a portable SSD onto their Gen5 drive dedicated to proxies. That's only ~4.5GB/s thanks to USB4's limit - and despite the board having two Gen5 slots, you lose 40Gbps USB4 if you want a second Gen5 drive. In a world with unlimited budget they'd be on Threadrippers, but instead they have to suffer their OS and application drive hanging off the chipset at Gen4. I have no idea how much that hurts their experience, but I wouldn't imagine it's a big deal otherwise they'd all be moaning at me.

 

[W1zzard]

There are 38 boards:

Motherboards with Socket: AM5, M.2 M-Key PCIe 5.0 x4: from 2x Price Comparison Skinflint UK

compare prices and read user reviews for Motherboards with Socket: AM5, M.2 M-Key PCIe 5.0 x4: from 2x

skinflint.co.uk

I looked at that same page before my post and came to the conclusion that that is not a lot, out of like 300 boards on the market

 

[Chrispy_]

There are 38 boards:

Motherboards with Socket: AM5, M.2 M-Key PCIe 5.0 x4: from 2x Price Comparison Skinflint UK

compare prices and read user reviews for Motherboards with Socket: AM5, M.2 M-Key PCIe 5.0 x4: from 2x

skinflint.co.uk

Far fewer if you consider that many of those are just duplicate models in white or with wifi.

 

[drSeehas]

Far fewer if you consider that many of those are just duplicate models in white or with wifi.

Far fewer?
many?

Out of the 8 white models are only 3 Gigabyte models duplicates.

I haven't compared the wifi models, but I assume in the 300 boards from W1zzard are some/many duplicate models with wifi.

 

[10tothemin9volts]

I asked AIs (used lmarena.ai) (local Qwen3 LLMs can solve this too) how 0.874W at L1 ASPM enabled would change the idle-like power-up time vs a 0.049W SSD one and the result was 11.3% higher power consumption, I think (I think I used 0.1 Watt instead of the following my own calculation of 0.049W).

Then I did my own calculation:
A modern 84Wh Strix Point 14" notebook, without a dedicated GPU, using WiFi and 150 nits screen brightness lasts 13:53 hours at idle-like tasks like light browsing (=6.05 Watt draw, the L1 ASPM is being used, not only in sleep mode, in sleep mode even deeper states like L1.2 may be used).

Let's say the used SSD is a Lexar NM790 2TB, like I have 2 (0.049W at L1 ASPM enabled). Now let's compare it to the Crucial T710 2TB (0.874W at L1 ASPM enabled) (provided the T710 would draw this much):
(84Wh/6.05W)/(84Wh/(6.05W-0.049W+0.874W)) = 13.63% higher power consumption (checks out with AI's result).

Changing the Wh doesn't change to result, so the calculation can be simplified:
13.63% = (6.05-0.049+0.874)/6.05.

13:53 hours - 13.63% = 11:59 hours.

Almost 2 hours fewer, this is a pretty crazy difference.

Would this be actually like this, maybe someone can test it?

 

[W1zzard]

Would this be actually like this, maybe someone can test it?

Looking at your math I'd say you are right or close enough