• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Intel announces new memory, XPoint

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.
Well that's the thing with Optane, its a NVMe tech that can be leveraged across multiple interfaces: Memory or Storage. Intel's Optane M.2 "memory" is a bit of a downgrade from their original design intent. Now from a consumer side that depends on utilization. Optane is not a storage replacement for client. Neither is it intended as such for the market. It is a NVMe for quick storage and high speed read write for data access. The best part is that it can survive a power cycle event (brown/black out). Allowing the user to not lose any data during those events. You can find other means to use it but honestly your M.2 or PCI-E expander NVMe's provided by Samsung or others are equivalent in terms of speed.

Intel has been the Google of client/server side tech for years. They test and tease something, and then pull it when tracking is forecast to be 0 in the near future.
 
You can find other means to use it but honestly your M.2 or PCI-E expander NVMe's provided by Samsung or others are equivalent in terms of speed.

Samsung are not even close to Optane when it comes to low QD random reads, which I think is where most of the benefit will come. This could be the biggest step since SSD replace HD. Depending on who's benchmarks you look at, Optane is many times faster than top end flash SSDs like 960 Pro. We go from tens of MB/s to hundreds. Home/consumer workloads are unlikely to hit the high QD required to extract good random performance from flash. Even the 16/32GB modules marketed as caches, when used as a SSD, showed that level of performance in benchmarks.
 
Eh the random reads is explained easily by the memory access architecture.

The NAND cell array is arranged in blocks just like HDDs. This means that when you read/write to a specific area, you access the entire block. Random isn't best because you may write only once in a block than switch to another. Takes a lot of time to do that.

Xpoint memory is a crosspoint architecture, meaning you can single out individual cells with each read/write. It may take several more write/reads (traditionally over NAND) but it is much faster access time, which makes up for the penalty.

Now what's cool is when 3D stacking and 5nm NAND meet, you'll see smaller blocks. This means that the penalty for access may be hidden and the total time could be comperable to crosspoint. Crosspoint may always have a natural advantage but it lacks in current technology development compared to NAND.


NAND 101
https://www.ece.umd.edu/~blj/CS-590.26/micron-tn2919.pdf

Crosspoint Explained
https://arstechnica.com/gadgets/201...veil-3d-xpoint-a-brand-new-memory-technology/
 
Optane Memory is an Intel only standard, and is not compatible with AMD/ARM
Huhwhat? These are PCIe drives running a bog-standard NVMe protocol. Stick them in whatever AMD or Cyrix or Samsung or Qualcomm system you want. If you mean "DRAM replacement" Optane, that's not happening in any sane system. It may be faster than NAND, but DRAM still wipes its *** with Optane.
 
The increase in transfer rate I understand to be due to the access latency. Xpoint is just a lot faster than nand. Hence this is a low QD problem for flash, you can sequentially access random parts faster with Xpoint. With high QD, parallel accesses mask the latency differences and the gap closes.

This technology is here, right now. You can buy it if you need it, or skip it if not. If you do need it, who would wait for a possible future evolution to flash that might or might not catch up? Maybe the future nand alternative ends up cheaper per capacity, and that'll be its selling point, but it wont affect anything I could do today.
 
Not sure how to interpret that reply. The different technologies are not exclusive. If any survive or die depends on someone finding value in it. Xpoint has a niche above nand, below ram. If future nand can move up to squeeze it out is something my crystal ball doesn't tell, but it is a good looking solution for today.
 
And I was saying that NAND will win for a long time to come. The industry as a whole needs to be able to adopt to other materials outside of Silicon. Samsung doesn't have production grade anything outside of Silicon, and neither does GF or TSMC. Not to say they are not experimenting with new materials, but they do not have plants like Intel/Micron for this type of tech. That's what I meant by industry adoption.
 
Done some more poking around.

Interesting testing here, but note it appears to be paid for by Intel: https://www.shroutresearch.com/blog...mance-and-implications-on-testing-methodology
Of particular note, it seems that keeping the CPU in an active state can have a significant impact.

Putting aside various claims, I also tried to look at price per capacity position, expressed in £/GB. My pricing is from one major UK e-tailer who is the only one I found taking pre-orders so far, apart from the enterprise part I had to look elsewhere to get values.

We have 3 classes of Optane product so far: enterprise SSD, consumer SSD, consumer HD cache. The enterprise SSD I'm finding hard to get pricing on, but it seems to work out just over 4.0. Consumer cache works out 2.3 to 2.5. The 900p SSDs work out 1.3 - 1.4. In general, bigger capacity per category is slightly better value.

Into conventional high end SSDs, 960 Evo works out 0.43 to 0.47, 960 Pro works out 0.55 - 0.58. That puts Pro around 25% more than Evo, and 900p is up to 2.5x the Pro. Yes, that's a hefty price premium... so you're going to have to be pretty sure it is worth it to buy the 900p.

What about as ram substitute? If I just find the cheapest DDR4 16GB kit, that works out £9/GB, so over double the enterprise Optane, and 6x - 7x the 900p. I don't know where the ram-substitute Xpoint pricing will fall, or how that compares to the more likely ECC ram used in enterprise workloads. Obviously whoever might eventually buy this stuff will need to look at the best combo for their use case.

On competition to Xpoint, I have now seen vague references to Z-NAND and Z-SSD. I guess this is the "Gen Z" Dolk mentioned yesterday, but I didn't find it at the time due to the name difference. I've not turned up much information on it, other than claims of lower latency than conventional nand, allowing it to be a possible Xpoint competitor. I look forward to finding out more but for now there is nothing like a 900p.

The UK place I'm looking is claiming stock on 10th so just over a week away. I'm not going to pre-order to keep my options open as others should get it around the same time and I can see if there is any wiggle room for pricing.
 
So I had to do some research into Z-NAND it it looks like it's just Samsung's next gen V-NAND which is their 3D stacking. 3D stacking can actually become xpoint like memory when you create smaller blocks like I proposed earlier. As note, this is not Gen Z. Gen Z is all about bus and protocal, not the actual technology that interacts with it. Technically a DDR2 memory module can work in Gen Z, but I digress.

Optane based RAM is an Intel want but the technology is proving hard to mature. There is the M.2 alternative that acts like RAM cache, but it is only supported on specific Intel motherboards and CPUs (link was provided in a previous post). I'm not really sure what is public knowledge on the Optane memory as I haven't seen too many articles on it. It's not really your standard memory but it is very much like it in all sense of purposes. It's just that it can hold state after power is removed.
 
Back