Samsung to release M.2 980 Pro SSD within two months

Damn it, just got 2 M.2 970 EVO Plus. Just one day earlier if I knew it I would wait.

 

I think our I/O bottleneck is the file system? NTFS is very old.

 

New Technology File System. Still new, still good.

I'd like to see devs take advantage of these extreme speeds like PS5 is touting, but sadly, it will not be implemented to much benefit. Every time I see people say, "It's in consoles now, so PC will benefit immensely since console is always the focus!" That trickle up (not down;that's insulting) will not be realized. Mark my words.

 

Phison E16 and only up to 1TB. This was not worth waiting for.

It is sad to see that Samsung stopped developing their in-house controllers.

 

Yeah, I thought Samsung is all about its own controllers.

 

Even if they do to really see improvement in games the i/o needs to be changed a lot or to completely get redesigned this is why the difference between sata ssds and nvme drives on game loading times the difference is tiny especially in comparison to the speed increase.

 

Is this I/O optimization done at the game's engine or is it related to the OS (Windows)?

 

Not only , the files are compressed on your ssd most likely with kraken . Say you want to pull an asset 100mb from the ssd .... The way it works on computer s now is you have to decompress then send it to your ram then send it to the gpu ... From what i got the controller sony uses atm allows em to straight up fetch the asset from the ssd skip all the in between steps and load it directly . We most likely need different /extra controller s on the ssd it self and ofc the os to be aware of it.... I was snob about the ps5 ssd claims till i saw what they did with it !

The way it works right now they will split the install in data chunks say you have the same building in 30 chunks needs to be there on all 30 of em so everytime it gets loaded to the rum..... If all ps5 claims are true you would not need chunks anymore since you can directly access the ssd with out having to threw the hoops to use an asset , look at it is rather interesting when i was looking about it i was constantly saying........owwwwwwh that's why they claim that and this.

 

IS this the one that rumors had as NVMe in PS5? (QVO)

PS5: NVMe (kraken data) -> Fetch to ASIC for decompress PCIe 4.0 x4 -> Send to shared memory pool at I/O speed between APU and Decompression ASIC = Up to 22GB/s (Typical 9GB/s) according to Sony.
PC: NVMe (kraken data) - > Fetch to CPU for decompress PCIe 4.0 x4 -> Send to System memory at 40~50GB/s or GPU at 64GB/s

PS5's custom NVMe controller with decompression capabilities may or may not deliver Kraken faster than CPU. It depends on how efficient it is and how high clock it has.
(Please note that Cerny stated that their ASIC decompresses data as fast as 9 Zen 2 cores.)

RAD states that Kraken provides 10 times faster decompression than LZMA and 2~3 times faster than zlib. According to benchmarks, it is 10 to 15 times faster than lzma.
One of many situational examples is test with i7-2600K @4,4GHz decompressing on single thread 1,2~1,5GB/s. With SMT one modern core is likely able to decompress over 2GB/s.

While looking for data, I actually found what was Epic talking about when they "rewrote" their code for PS5's storage.
There was rather extensive test of different compressions. Time it took to decode and load game. And beyond certain read and decompression speed UE4 did not actually load any faster.
Basically it did not matter if they got decompressed data at 400MB/s or 800MB/s. Time to load level was practically same.
So Epic did not listen to people over years telling them that there is something wrong. It took storage system capable to deliver magnitude more data per second than was their limit to actually move finger.
This change will help everyone with better storage.

 

PS5's advantage is not in sustained transfer speeds. It's how the controller, filesystem and OS is able to handle fetch requests, which makes it way faster and more responsive than a PC, in such terms as what a videogame needs to run smoothly. Linus from LTT initially also thought that the PS5 SSD was about raw throughput, and made bold claims that you could already get storage solutions for PC, which was faster than the PS5. But he eventually had to make this apology video, to correct his own false initial assumptions, where he also explains the actual advantages of the PS5 storage architecture:

 

Will be interesting to see how that API Microsoft was working on and other OS improvements will change the PC situation in regards to not just transfer speeds but other advantages as well but changing some of these might be a long-term gradual improvement without risking breaking stuff or trying to implement some newer model for file system or updating NFTS. (I don't think ReFS is intended for regular consumer systems and might never be intended to replace NTFS?)

Suppose there's a lot to this from how the OS itself currently handles stuff both HDD and SSD wise and then I/O and on the file system side of things, fetching or storing to RAM and then transfer bottlenecks like PCI Express or others although maybe caching critical data to RAM could help at the expense of a bigger increase in RAM requirements possibly even recommended speeds without storing the entire data like some RAM drive.

Might matter more for PS5 exclusives too than multi-platform projects though HDD PC's are still common for storage purposes and even some that only have HDD's so lowest common denominator and how that will work out but getting full effect and taking advantage of these new consoles might also take a bit of time although I can see why developers are really happy getting that bottleneck done with.

 

I heard that before. Listen to him again. And carefully think why he is making apology.

 

I don't think we will get the same real scenario speed as next gen consoles, they are much tightly integrated. Going forward in the future with MCM apu design + on chip memory (HBM) + on chip NAND storage, general purpose computer will be lagging more and more, we need some serious architecture shift and rethink the whole motherboard concept if we want to keep up in the future.

 

okey ps5 ssd is quicker, but we the master race can just buy more ram? preload the whole thing into ram, now try to beat that with nv storage, finally a reason to get +64gb for gaming

 

Because that he has misrepresented what Tim Sweney said, and not bothered to try and understand what it is about the PS5 storage solution which makes it special, compared to PC solutions.

 

No, because he come to belief that Mr. Sweeney said something that he did not say.
PS5 fanboys said those things, Linus took it as Sweeney said it. (Since they claimed it was him.) And went ballistics on those stupid claims.

As part of apology he highlighted some strengths of PS5. (Free marketing.) But it still does not make those Fanboy's claims any more real than before.

 

Besides that it is slower than GDDR6 memory and needs an additional step before hitting the GPU?
And then comes additional time used loading a game into memory every time, I didn't run over any "smart" caching solution that does a real difference so far, even with smaller games it really doesn't get there.
The PC and OS needs a proper controller with a co-processor for NVME and onboard GDDR6 besides OS support for it all to get these 3 seconds game loading times.
Shouldn't forget that the Xbox Series X also is blazing fast, so MS would be stupid if it is not in their interest to push the technology to the PC.
I'm quite sure that PC's will get there in time, question is more if they can do it via addon to existing hardware or if it needs a complete new motherboard.

 

3 seconds loading time not the question of NVMe being faster/slower on PC. It is question of compression used on data. In other words it is choice of game developer.
Often, developer makes choice that's most fitting old HDD. That's to get maximum compression ratio regardless of decompression cost on CPU as HDD reads usually 80~120MB/s.
But then this heavy compression hits CPU wall with SSD and it can't pull from HDD more than 200~250MB/s. Change compression and you suddenly can read 1000~1500MB/s from NVMe per CPU core (with one thread). Run it with SMT and get maybe to 2GB/s.

Is it an problem to sacrifice entire core when game is loading? (Thread that loads data from SSD with inefficient compression uses entire core anyway as it becomes limiting factor.) Or maybe two cores to get higher throughput? It depends if you really can(need) use all 8C/16T, your gaming PC has available, for rendering while game actually load data from storage. Answer to this is no. Games are at edge where 4C/8T is not OK (can't maximize fps under all usual gaming scenarios). Sometimes 6C/6T shows reduced performance too. But 6C/12T is OK.

And that summarizes current issue. Game engines often look backwards to support HDD. If they used tooling adequate for SSDs, then NVMe-s would already provide considerable boost too. If games were not dumb coded, but were bit more CPU aware, then many more workloads would have been spread across more cores.
Stupid java had parallelism built in for loops long time ago. Engine Developer should just ask simple question: "Is this CPU heavy loop self iterative or can it be parallelized?" And this applies to any processing.

 

Well i´m thinking about buying a M2 SSD so this is interesting for me. The problem is the size and price because with games becoming bigger and bigger, i don´t know if a 1Tb SSD is going to be future proof...