Aorus Gen4 7000s Premium SSD features bulky stacked cooler

[HEXUS]

Latest 2TB SSD uses Phison E18 controller and is capable of 7,000 MB/s seq read speeds.

Read more.

[Irien]

Makes you wonder just how necessary such things are. I look forward to fitting this into my laptop, for example! I get that they're chasing performance at all costs, but if you can't sensibly fit them into computers, a re-think may be in order.

Are we looking at a two tier NVME drive market going forwards - one for use in laptops and confined chassis with one power profile, and another designed around a higher TDP for desktop use?

[ksdp37]

No way this is going to fit on my mobo. The graphics card takes up the space directly above the first m.2 slot and I've got a PCI-E card on top of the second slot. They both use the manufacturer provided heat sinks (with the thermal pads) for the cooling.

How are people supposed to fit a huge cooler like this on an already tight mobo? Do we all need to buy e-atx boards now?

[philehidiot]

Dear Anandtech,

The Destroyer benchmark has another victim for your amusement.

Please order a couple as you'll probably melt at least one.

Regards.

[Tabbykatze]

If a PCIe 4.0 SSD needs this kind of cooler to hit max speed, I don't mind AMD shying away from PCIe 5.0 for their first AM5 iteration.

This is ridiculous and impractical, I would like to a see a performance/temperature comparison with the cooler on, off and with a motherboard based cooler to see how necessary this really is.

[kompukare]

My first questions would be , how does NAND react to temperature?
And does it affect longevity?
Not much chance being able to use these in laptops.
With NAND write and erase uses a lot more power than reading (although a quick search only showed quite old figures), so aside from show, maybe only very write heavy users actually need the big cooler?

[philehidiot]

My first questions would be , how does NAND react to temperature?
And does it affect longevity?
Not much chance being able to use these in laptops.
With NAND write and erase uses a lot more power than reading (although a quick search only showed quite old figures), so aside from show, maybe only very write heavy users actually need the big cooler?

I thought sinks were more for the controller than the NAND itself?

[kompukare]

I thought sinks were more for the controller than the NAND itself?

Hadn't though of that.
Had just assumed since NAND uses a lot of power to erase and write, most of the power usage was from the flash. But looking it up, that doesn't seem to be the case.
Despite NAND being power hungry, for writes it prefers higher(ish) temps, which is surprising.
So some advise was to keep the controller cool, and leave the NAND warm.

EDIT: and just to rub in my mistake, along comes an announcement for Marvel PCIe 5.0 controller:
https://www.anandtech.com/show/16703...sd-controllers

Nearly 10W just for the controller. That may never be useable in a M.2 2280 part.

[Tabbykatze]

EDIT: and just to rub in my mistake, along comes an announcement for Marvel PCIe 5.0 controller:
https://www.anandtech.com/show/16703...sd-controllers

Nearly 10W just for the controller. That may never be useable in a M.2 2280 part.

That's nearly a tripling of the power needs from Phisons initial 4.0 controller offerings and double that of the total power needs of a 980 PRO just for the 5.0 controller!

[kompukare]

That's nearly a tripling of the power needs from Phisons initial 4.0 controller offerings and double that of the total power needs of a 980 PRO just for the 5.0 controller!

I guess Marvel would argue that the metric we should focus on is power per IO and throughput. Marvel actually go on about that one using less than their previous enterprise controller. Guess in a rack server a few 15,000 RPM 40mm fans should keep it cool?

[Tabbykatze]

I guess Marvel would argue that the metric we should focus on is power per IO and throughput. Marvel actually go on about that one using less than their previous enterprise controller. Guess in a rack server a few 15,000 RPM 40mm fans should keep it cool?

Well the 980 Pro can do 1 million IOPS sequential read and the consumption for read is rated at 6.2w giving ~161k IOPS per watt from the total board whereas the marvell silicon produces 1.8 million IOPS (going to assume it's sequential) in a 9.8w envelope just for the controller giving ~183k IOPS per watt. That's only an increase in watt efficiency of 13% and that's discounting the energy requirements of the other surrounding silicon.

Not sure that's a great advertising metric either.

Edit:

The Phison E18 SSD can do 1 million IOPS in a 3W controller power giving ~330k IOPs per watt and that was announced earlier this year and really screws up Marvells power/IOPS rating. The Inland 2TB SSD (E18) review that Anandtech looked at had a maximum PS 0 power of 8.8w for 700k IOPS giving ~80k IOPS per watt, that doesn't look that great compared to the controller capability, I wonder how hard they were pushing the flash. But this could just be the "maximum advertised speeds" versus "what actually happens" so if we're looking at a 30% deprecation in IO with an added 5W just for the NAND flash and supporting components, the Marvell is still operating at nearly double the power requirements of the previous gen and giving little over 90k IOPS per watt if the math is linear.

[Percy1983]

Where is the line of diminishing returns, I am running a PCIE 3 NVME in a PCIE 4 board just due to cost, my drive is one of the fastest 3.0 drives (3000ish MB/s) and at the time of buying the 4.0 drives were pushing 4000-5000.

At what point is it fast enough with no real gains?