Intel Optane 905P 1.5TB SSD goes on sale for $299

[Admin]

Newegg has issued a promo code to reduce the cost of the Intel Optane 905P 1.5 U.2 SSD by $50. It costs $299, making it a viable choice for those with write-intensive systems and services due to its low access time, high endurance rate, and warranty support of up to 5 years.

Intel Optane 905P 1.5TB SSD goes on sale for $299 : Read more

 

[JRStern]

It has a ton of write endurance, but can it do continuous writes or does it overheat and slow down?
Barring that use case my understanding is Optane makes a great SSD except that since the technology doesn't stack 128 or 256 layers deep the density isn't as good as newer flash, and the production price per gigabyte is many times higher than current flash - but Intel still has a couple of warehouses of the stuff and is eager to clear them out for pennies on the dollar.

 

[Jame5]

These drives do require some kind of active cooling. They are not low power by any means. In my case I have a 905 mounted directly behind the front 120mm intake fan to ensure proper cooling.

That being said, I would love to get my hands on one of the 5800x's for the boosted performance metrics from the newer tech and PCIe4.0.

However, they are still stupid expensive.

905P SSD 1.5TB = $300 on sale
5800x SSD 1.6TB = $2800.

 

[Notton]

but can it do continuous writes or does it overheat and slow down?

continuous
but, it's 2,200MB/s, so very slow by today's standards.
https://www.tomshardware.com/reviews/intel-optane-ssd-905p,5600-2.html

Oh, and this one is a U.2 form factor, so unless your mobo has the connector for it, you'll need an adapter.

 

[rluker5]

I picked up a 1TB a year or two ago and it came with the m.2 adapter. I also have a 900p 480GB that was $600, but it has run great since 2017 and is still roughly as fast as the fastest NAND on game loading and is faster than all NAND for my OS drive.
That being said, if the rest of your PC is fast switching from a 60hz to a 120 hz display will make a bigger difference in your felt responsiveness.

 

[thestryker]

Just as a warning for anyone kicking this around the 1.5TB model is just the bare drive without M.2 adapter. These drives are still significantly better than anything NAND based for low queue depth work, but much worse at sequential so it's all about workload.

 

[Amdlova]

Once? Still the fastest drive... unbelievable low latency. You will not it on benchmarks but on everyday task feels way polished and faster to do some of the work.

 

[bit_user]

LOL, in 2022 I bought a 400 GB P5800X and it's still sitting unused. Can't decide whether I should use it as an OS drive, or try to wait and see if prices ever go up above what I paid for it. I know I probably should just go ahead and use it, because...

The next gen of storage devices is already coming onto the market, and they will smoke any Optane drive ever made. However, they require a platform with CXL support:

https://www.tomshardware.com/pc-com...rage-tiering-comes-to-remote-memory-over-pcie
​

I'm thinking the main reason to put DRAM and NAND in the same device is if you can use NAND as a transparent, backing store for the DRAM. So, if it has a NVMe mode, then even my Optane P5800X is already decidedly obsolete.

 

[bit_user]

Anyone who wants to risk it can find pretty good deals on the P5800X on ebay. Here's the same drive I got (except with Dell branding) for less than half what I paid:

https://www.ebay.com/itm/285811104023
​

They claim it's new.

 

[Li Ken-un]

This drive was once called "the fastest SSD ever."

It still is, depending on what metric you are measuring. 😉

If you are comparing to a P5800X? Even the 905P gets smoked.

For consumers like us, the metric that matters (low latency at low queue depths) is the primary reason to buy it. The endurance is really for the professional and enterprise users, although the 905P is probably artificially gimped like the earlier M10 and 800P series to go into read-only mode no matter what after a certain amount of writes.

 

[bit_user]

although the 905P is probably artificially gimped like the earlier M10 and 800P series to go into read-only mode no matter what after a certain amount of writes.

If it goes into read-only mode after the spare blocks have been exhausted, then that's in most users' best interest. To keep writing it with no spare blocks remaining, you're risking almost certain corruption and data loss.

 

[DavidC1]

continuous
but, it's 2,200MB/s, so very slow by today's standards.
https://www.tomshardware.com/reviews/intel-optane-ssd-905p,5600-2.html

Slow? What and how often of your workload would reach quarter of that speed consistently?

You could have a PCIe 7.0 SSD with 32GB/s bandwidth but it'll never reach the 10us latency that this is able to give you. Your fancy schmancy 10GB/s NVMe SSD is essentially 10x the latency.

Latency = benefits every workload, and without needing buffers, or caching of any sort. Fast the first time.
Bandwidth = You need parallelism to reach max bandwidth meaning enough chips in parallel + certain workload size + certain CPU threads running. You could have 1TB/s bandwidth, but it won't help you in responsiveness over a drive that delivers only 100MB/s bandwidth but at 1/2 the latency.

$300 for 1.5TB full on Optane is dirt cheap. If I was looking for a drive I would get one.

If it goes into read-only mode after the spare blocks have been exhausted, then that's in most users' best interest. To keep writing it with no spare blocks remaining, you're risking almost certain corruption and data loss.

The endurance is really for the professional and enterprise users, although the 905P is probably artificially gimped like the earlier M10 and 800P series to go into read-only mode no matter what after a certain amount of writes.

It's rated at 27 PetaBytes. It has 575K 4k IOPS, meaning ~2.2GB/s bandwidth in random read/writes.

27PB / 2.2GB = 12.3 million hours or 1400 years. The drive's electronics will likely fail anywhere between 1/10th to 1/100th of the write lifespan. Or, maybe it'll fail in 10-20 years from thermals and thermal cycling.

Even if the drive was capable of 128GB/s bandwidth comparable to a latest Zen 5 gaming PC's dual channel DDR5-8000 memory, it still would take 24 years of 24/365 to reach that point.

You'll NEVER worry about the media running out of write cycles even if you max it out and run it 24/365. People worrying about lifecycles on Optane are in the mental hospital paranoid territory.

 

[Li Ken-un]

If it goes into read-only mode after the spare blocks have been exhausted, then that's in most users' best interest. To keep writing it with no spare blocks remaining, you're risking almost certain corruption and data loss.

It might be, but they really ought to let the user decide. They do let P4800X owners decide: there is no read-only mode, and the controller simply caps the write speed to protect the media.

The two other notable differences are the much higher warranted endurance and support for OPAL SED.

Some have speculated that the limits are firstly for market segmentation, and secondly purposely conservative because they didn’t even know the true limits of 3D X-Point media themselves.

It's rated at 27 PetaBytes. It has 575K 4k IOPS, meaning ~2.2GB/s bandwidth in random read/writes.

27PB / 2.2GB = 12.3 million hours or 1400 years. The drive's electronics will likely fail anywhere between 1/10th to 1/100th of the write lifespan. Or, maybe it'll fail in 10-20 years from thermals and thermal cycling.

Even if the drive was capable of 128GB/s bandwidth comparable to a latest Zen 5 gaming PC's dual channel DDR5-8000 memory, it still would take 24 years of 24/365 to reach that point.

You'll NEVER worry about the media running out of write cycles even if you max it out and run it 24/365. People worrying about lifecycles on Optane are in the mental hospital paranoid territory.

You got the math wrong there.

905P 1.5 TB’s warranted endurance: 27.37 PB	P4800X 1.5 TB’s warranted endurance: 164 PB
= 27,370,000 GB	= 146,000,000 GB
… / 2.2 GB/s = 12,440,909 seconds	… / 2.2 GB/s = 66,363,636 seconds
= 207,348 minutes	= 1,106,061 minutes
= 3,455 hours	= 18,434 hours
= 143 days	= 768 days
= less than 5 months	= over 2 years

Now we have a lower bound on the length of time it takes to reach the warranted write endurance.

Anecdotally, Optane controllers have been reported to fail. But the example I have seen was not before the warranted endurance had run out many times over. It’s speculated that the media was fine and would have continued humming along had the controller not died.

Sauces:

• AnandTech: Intel Optane SSD DC P4800X 750GB Hands-On Review
  “When the P4800X exhausts the rated write endurance, it will switch to a ‘write protect’ mode where writes are throttled to 30MB/s.”
• Level1Techs: BVD’s reply to Would Optane be good for running virtual machines from?
  “‘Hey, you remember those drives you were asking me about last year? Well one of them FINALLY DIED!’”
  “Intel P4800x - 375GB
  Power on hours: ~55400 (6 years and 4 months)
  Lifetime writes: ~156.3PB”

 

[Jame5]

Honestly, the main argument for these drives in any use case is extremely low latency, low queue depth work. OS work is something that matches with this use case. The big reason for this is that the Optane tech can hit it's full performance at T1/Q1 loads. This is very unlike NAND based drives that can only hit their rated performance numbers with a large amount of simultaneous accesses. (Think T8/Q32 type benchmarks.)

 

[bit_user]

It might be, but they really ought to let the user decide. They do let P4800X owners decide: there is no read-only mode, and the controller simply caps the write speed to protect the media.

The two other notable differences are the much higher warranted endurance and support for OPAL SED.

Some have speculated that the limits are firstly for market segmentation, and secondly purposely conservative because they didn’t even know the true limits of 3D X-Point media themselves.

Thanks for including the ark.intel.com links. Did you try using the product comparison feature of that site? It has an option to highlight the diffs.

One thing that immediately jumps out at me is the difference in peak temperature. The consumer version allows up to 85 C, while the datacenter version is capped at 70 C. I don't know about 3D XPoint, but NAND endurance is highly sensitive to temperature.

As for why they allow the consumer version a greater temperature excursion, I'd speculate they're aware that people are going to be caught off guard at how hot these things get and will tend to under-cool them. I'll bet you've seen these PCIe add-in-cards you attach the drive to, directly. Sitting in a back, bottom corner of someone's PC probably isn't going to provide enough ventilation.

Other things I'm noticing are that the consumer version goes a little heavier on read performance and that MTBF is only 80% of the datacenter version. Don't know what to make of that, but perhaps the hit on read hit comes from a stronger ECC algorithm.

We could also speculate that the 3D XPoint chips, themselves, are binned in some way. Maybe they saved the chips with the fewest bad cells for the datacenter drive, which both gives the DC drive more capacity of reserve blocks and tends to correlate with generally better natural endurance.

Anecdotally, Optane controllers have been reported to fail. But the example I have seen was not before the warranted endurance had run out many times over. It’s speculated that the media was fine and would have continued humming along had the controller not died.

How about temperature? Was the drive well-ventilated?

 

[bit_user]

Honestly, the main argument for these drives in any use case is extremely low latency, low queue depth work. OS work is something that matches with this use case. The big reason for this is that the Optane tech can hit it's full performance at T1/Q1 loads. This is very unlike NAND based drives that can only hit their rated performance numbers with a large amount of simultaneous accesses. (Think T8/Q32 type benchmarks.)

Not only QD=1 IOPS, but how about 14M high-QD IOPS from a single drive?

https://www.phoronix.com/news/Linux-14M-IOPS-Per-Core
​

Granted, that was using the P5800X, but I wouldn't be surprised if Axboe could squeeze a couple million out of even the 905P or DC4800X. The drive he used for these tests was rated at only 1.5M read IOPS, which suggests Intel's testing was heavily bottlenecked by the host they used.

 

[0112358]

Very good to see. This is one of the most uplifting bit of news I've seen in quite a while as regards SSD latency/availability. Done deal for me -- plenty tired of waiting for Gen5 latency improvements to show up, and as best I can tell even when the best-of-the-best Gen5 finally gets here, Optane 905P will still beat it substantially. Can now finally get off of the SSD-latency carousel. Love me some Samsung 990 PRO...but Optane remains way out in front of that for low queue depth latency. "The End."