Wendell just posted an episode about optane. I never used one (too expensive) and am confused about all those versions-there is one that looks like an SSD (does it plug into a sata port?), there his one that looks like a PCIE card, and there is one that looks like the gum stick NVME drives. And from what I can see, some of them also have a bit of ram? Why are there so many versions?
Wendell showed faster boot speed , lower latency and much greater IOPS even tho the drive had low bandwidth and was operating on PCIE3. There were some oblique references to the fact that optane requires special intel drivers which I did not understand. He also mentioned primocache. I looked up that software and it looks like it is just a RAM disk. So I am a bit confused-was he using primo cache to get that low latency?Are all the optane drives equally snappy even without intel drivers?
optane is just a different flash technology that has super low latency and high write endurace that can be used as cache drives, or standalone.
They are regular block devices and no not require drivers
The connector in question is U.2 and incompatible with sata, they are a cable used in servers to take nvme from somewhere else to the mainboard or a backplane, gotta remember this is enterprise equipment
The rest are just analogous to an SSD just super low latency
This was in reference to the H10 and H20 products, which have both optane and regular NAND flash on the same m.2 device. Intel made special drivers for these devices that would allow Windows to use the Optane as a cache device for certain disk operations and the higher latency NAND for everything else. I think that the software PrimoCache Wendell spoke of can achieve a similar effect using these devices, but I am unfamiliar with it because I donât use Windows.
As @anon55922884 said, all the other devices in the Optane lineup are just block devices that can be used however you want.
EDIT: I forgot to mention the pmem DIMMs that were also part of the Optane lineup. These suffered from being completely vendor locked to Intel Xeon processors at a time when many customers were jumping ship to AMD to get on TSMCâs better process node.
Because Intel was making a really incredible product but at an extremely high cost, they tried all sorts of different iterations of the product in an attempt to appeal to different market segments. In an unfortunate coincidence, they didnât really take off in the enterprise or consumer markets mostly because of the high cost and confusion that was caused by their marketing of the devices. In addition to the price, enterprise didnât show much interest because they didnât want to become entirely reliant upon an experimental product that only Intel was making. Consumers mostly didnât understand it because storage devices are only ever marketed by throughput instead of latency, and âbigger number = more gooderâ is all your average consumer thinks about.
The whole situation is kind an unfortunate clusterfuck because it really is quite an incredible product, but it just didnât pan out money-wise which is all that really matters in the modern Western economy.
Itâs also not really got a lot you can point to besides numbers to really communicate the product well.
How many consumer use cases even benefit much from that lower latency? Just booting into Windows, basically. And the write endurance is also a tough sell, since it didnât take that long for flash to be good enough again after the planar TLC disaster.
Who is optane really for? Basically just people who want the best thing for the most trivial reasons, and a few niche or enterprise software cases. It could have been great for a lot more, but that requires software adoption, which requires hardware adoption, which requires software benefits.
Thatâs an accurate description of the entire enthusiast PC hardware market, you can buy used cars and run down houses in bad neighborhoods that cost less than the current flagship Nvidia product.
Overall though weâre talking about the same circumstance in different ways. Usually enterprise picks up on new trends first because they have use cases that demand new hardware capabilities, so the hardware makers make more of it, then the manufacturers have the production capacity to have a surplus of it, then smaller scale versions of it are designed and given the software support in consumer applications. Intel was just running such a massive deficit bringing this product to life that they needed to make it profitable quickly to appease the investors (fiduciary duty etc.), which led to a bungled marketing strategy that didnât play out like they hoped as they threw fist fulls of ideas at the wall but none were sticking quickly enough, and so the plug was pulled prematurely by the bean counters.
Regardless, Optane was still a great product with lots of potential that could have improved the computing experience for all if it had led to the concept of flash storage that has both high throughput and extremely low latency to really take off. Intel was already stretched thin from the ground they lost to AMD/TSMC over the past few years and their foray into the dGPU market, so between those and the cost to bring Optane to market the timing was incredibly unfortunate.
EDIT: To any readers of this thread who havenât seen it, I highly recommend this TechTechPotato video that serves as a postmortem on Optane and the difficulties Intel had with it.
Basically every operation that moves data from block storage into the CPU, which is a lot of them. But the question is to what degree, and what are consumers willing to pay? Unsurprisingly, NVMe (heck, SATA) SSDs have hit the âgood enoughâ point for most people so the significant cost increase of Optane has made it nonviable for consumer platforms.
Businesses and prosumer. The performance delta between NAND and Optane is not significant enough for consumers to care given the price delta. But itâs a weird spot, as most businesses can just throw $$$ at more RAM capacity in HA server configs and for most use cases that performs better than Optane at about the same price.
For the homelab user with a single storage server though, the Optane firesale is fantastic.
I have a pretty fast computer ( Threadripper PRO 3955WX, a buttload of RAM, RTX 3080 Ti, multiple Samsung 980 Pros), but Iâm endlessly waiting for it â sometimes for things that take long calculations or disk scans, but often just laggy little experiences going from thing to thing in the interface. And, that sucks â its destroys flow state when trying to get some work done. Anything that makes computer interactions go closer to the speed of thought is worth it. A thousand dollars worth of gear, compared to wasted salary hours, is whatâs trivial.
On modern Intel motherboards there is an option to use Optane in fancy ways. The H10 and H20 are Optane and Flash on the same NVMe device.
Probably the easiest Optane to use are the straight NVMe Optane sticks and the PCIe cards, they are basically the same thing but the cards are much larger capacity. They are fast SSDs and show up as an SSD you can use them in the same was as any SSD.
Primo Cache is for caching hard drives, it uses RAM and SSDs to achives SSD like speed. Optane is good for this because it speeds up boot time more than Flash SSD.
Optane cannot compete with NAND flash on price and that is the reason why it âfailedâ from a business perspective.
When it was first introduced the market was much different, but now you can pick up a cheap QLC SSD for 4 cents/gb with many premium drives starting to get close to 6-7 cents when on sale. Meanwhile Optane is still over $1/gb unless you can get it at a huge discount. It just doesnât make sense from an economic standpoint unless you have a very specific use case that canât be covered by NAND and/or spinning rust.
Itâs hard to say if itâs just perception or if it is slow. To be on the safe side Iâd say it was slow.
I would suggest you diagnose the machine by a process of elimination. First do a live boot, like Linux Mint USB and see if that is slow. You may still be able to access most of your files.
You may also try removing all the RAM sticks bar one or two.
Just eliminate parts of your computer until you have something that you would say is fast.
Computers will keep retrying rather than fail, this makes them resilient but also makes them slow if there is a fault.
It will be a while but SK hynix does own the Optane nand now. If they come up with a good controller to leverage it on a modern NVME drive, they could do brilliantly.
You can use a PCIe adapter like that, or an m.2 to u.2 adapter like this. If using an m.2 slot, make sure that the drive you want maximum performance on is in an m.2 slot whose PCIe lanes go directly to the CPU and not to the chipset.
You can usually just remove that heatsink. Itâs not necessary with an adapter cable.
That is not Optane, just regular NAND flash memory. Though it is the high-end enterprise variety. The flash memory used in Optane is referred to as 3D XPoint. I believe that all Optane products say Optane on them.
@AbsolutelyFree Thank you. I appreciate the time you took to reply. I am not sure about what you mean regarding connecting to the pcie slot connecting to the cpu versus the chipset. Since this drive is a pcie3 device, that its slower than pcie4 or pcie5. So why does it matter if I use the chipset or direct to the cpu?
On every computer platform, add-on components connect to the CPU via PCIe lanes. Each processor has a limited amount of PCIe lanes coming from it that can be used for peripherals. Many of the empty peripheral slots on a motherboard have PCIe lanes that go directly from the slot (specifically PCIe slots & m.2 slots on consumer grade hardware) to the CPU.
All consumer grade processors also have chipsets that are used for various purposes that are located on the motherboard. The chipset has a dedicated amount of lanes going from the chipset to the CPU. One of the purposes of the chipset is to serve as a hub that other devices connect to, such as the onboard sound chip, network chip, and SATA devices, but it is also common for some of the PCIe & m.2 slots to be connected to the chipset rather than directly to the CPU. This allows motherboard manufacturers to give you more slots for peripherals than the CPU has dedicated PCIe lanes.
The problem with this though is that all of the devices that are connected to the chipset must share the PCIe lanes that the chipset has dedicated to the connection between the CPU and the chipset. If the chipset only has 8 lanes PCIe lanes to the CPU but the sum of all the devices that are connected to the chipset is greater than 8, the chipset bottlenecks the connection speed. Also since all of the signals for devices connected to the chipset are using the chipset as a middle-man (instead of a PCIe connection that goes straight from CPU to device), they will inherently have higher latency.
I am saying that for the devices that you want the maximum throughput and lowest latency on, you should make sure that the slot they are plugged into goes directly to the CPU and not to the chipset. Your motherboard manual should tell you which slots get direct connections to the CPU and which are connected to the chipset. For instance on the motherboard I use on my desktop, the manual says in regards to m.2 slots:
The slot labeled âM.2_1â is connected to the processor while the âM.2_2â slot is connected to the chipset. This means that my most important M.2 device that I want the best performance on should go in slot âM.2_1â.
are usb sockets on the back of the motherboard routed to the chipset? if I dont have >8 devices routing through the chipset, then it should not matter?
I was talking about a hypothetical scenario, every motherboard is different. Regardless, the most attractive feature of Optane is extremely low latency. If the chipset is between the CPU and the Optane, you are going to have higher latency no matter what your specific configuration is.
