I think it would be great if you guys would color code your graphs depending on if higher or lower is better - I feel like that would make them much easier to read at a glance.
Seconded. Maybe a little logo on the left or right for lower/higher or something, so you don't halve the available colors for big graphs, but that would make them much easier to read, especially just watching through the video rather than pausing to read each one in detail.
I loved the final bit about PCI 5.0, it's the most important message of the video. Desktop is starting to get the same PCI lane benefits of server-class platforms of a few generations ago, in a much more slim package. PCI 6.0 might be the tipping point of too much bandwidth, where we come out with novel ways of using it all.
I would be excited to see 2x lanes for GPU shrouds connecting over USB4 or 5, imagine how small we can get the power of a steam deck to be in another 5 years. The use wouldn't be for 16x slots, and maybe at some point 8x slots become the norm for desktop GPUS
@@endmymisery3623 8x 4.0 GPUs are already becoming the norm for the mid and lower end lineup of cards which is really good for anyone who might need a heck lot of PCIE cards for home servers. hoping that CPUs and mobos also support for 32+ lanes on consumer chips because can benefit a lot from that also.
@@Blackdune00006 8 lane GPUs aren't in any way an advantage over x16 lane cards. It only reduces flexibility. It hurts performance when installing the cards in a previous pcie gen motherboard.
It only applies sometime in the future where all devices attached via CPU lanes are gen 5. An RTX 4090 in a gen 5 slot that's been reduced to 8 lanes will work with gen 4 x 8 bandwidth. Also add in cards for more drives are expensive. E.G a 4x m.2 nvme Gen 4 adapter. That and motherboard manufacturers have stopped building boards with split x8 x8 CPU lanes for the most part with the demise of multi GPU.
6:30 The reason why the Gen 5 drive won in the first round but lost in the second when opening PCMark 10 is because after opening PCMark 10 once, PCMark 10 is saved into the RAM as cache. So in the second run, the drives were not used, instead Windows loaded PCMark 10 from the cache instead of the drive, which is why the Gen 5 drive seemed slower in the 2nd round.
@@pabloespinoza9046 How are you so confidently wrong 😭 "By default, Windows caches file data that is read from disks and written to disks. This implies that read operations read file data from an area in system memory known as the system file cache, rather than from the physical disk. Correspondingly, write operations write file data to the system file cache rather than to the disk, and this type of cache is referred to as a write-back cache. Caching is managed per file object." - Windows app development documentation
SSD storage is interesting. A few years back I worked in a lab and we had a very niche use case where we needed maximum sustained throughput. This was when the 970pro was the top of the line. We had a high speed camera setup that collected data from an experiment. It was not for particularly insane fps, but high FPS for rather insane lengths of time to capture an entire run. We'd saturate PCIe 3 for several minutes. Basically filling an entire 1TB drive per experiment run, one camera per drive and up to three cameras. At the time, the pro series used actual MLC, not TLC they now advertise as "3 bit mlc". This meant that they didn't have the cache overflow performance hit and could happily fill at full speed. Once to save a buck, one of my colleague's got some EVOs. They did not understand the cache behaviour and just saw that the drives had the same performance specs since w didn't need longevity. Needless to say they did not work. After a few seconds of video the collected fps would tank as in the plot in this review. Seeing that chart reminds me why I remain bitter about the way drives are marketed to this day 😂.
I really liked the explanation as well as the animation for the plane passengers to explain how ssds function. I think explanations that anyone can understand help to get a lot more people into tech when it doesn't seem so intimidating. Great video!
I'm glad you included some discussion about the PCIe lanes because this does seem to be a point of stagnation. We haven't seen much development on the lanes in years other than upgrades to the PCIe generation. Though when it comes to drive performance, I would have loved to see you dust off a PCIe 3.0 Optane drive and see how its random performance compares. I suspect it would still be able to keep up with the Gen 5 drive in some tests.
Its because chip and boards manufactors are foucsed on a single chip instead of breaking up that chip set and getting softward to work with dozens or 100s of chips and thousands of lanes of traffic. In the future you will find 100s if not thousands of tiny chips on a board that will function vastly faster than anything we have today. The ability to build out a chip set for a specfice task rather than using a clock speed to deligate tasks is what will speed up the process. Today we store the OS on the main drive when the OS should be flashed on to a chip that super fast. If you get what I mean. There is no reason this can not happen.
no need to, here is tested fact. optane, the 3dxpoint only not the hybrid, would have crushed any other m2 in all metrics except sequential speeds and keep it's performance rock stable no matter how much time the tests took. that's why the enterprise loved it. too bad intel axed it due to being complex and expensive to make.
@@marvinmallette6795 The PCIe generation advances double the base throughput each time, but that is just for straight transfer speeds. Depending on what you are doing with the data you transfer across, it might not be able to keep up with the speed of the link. That's why Linus pointed out how random I/O performance on those SSDs hasn't improved all that much.
Another possible test you could do for drives is verifying files of a game. I feel like that's something people actually do semi-regularly, and it would be good to know if there's an actual real world difference there. I imagine it would just be the same as sequential read/write but you never know!
@@Crecross yeah right this exact time right now nobody is buying it, maybe in the future when gen 5 is the standard people will want testing of the product already done
In my experience, and after confirming many times that all my drives are working properly, there's no difference in any normal PC scenario or any gaming scenario between an 870 EVO SATA SSD and a Gen4 Aorus drive, I have 2 identical drives of each, I got the second Gen4 thinking stuff will actually make use of it, and I decided to just buy 2 bigger SATA drives and they're running everything flawlessly
It entirely depends on how many threads Steam is spawning to do verification. I don't think Steam would try to "Stress test" the system during verify task. So it might not be a valid test.
I think you made a strong point of how useful the new per-lane efficiencies of PCIe Gen5 are, technology and cost-wise. What you didn’t address is what are the next steps (if any) that the industry might be considering for making a leap in random read/write speeds and/or cache dimensioning. Definitely a good subject for a follow up video. See what’s out there. How to resolve the baseline problem of random read/write speeds not improving at all.
Although you are correct, i.e. it's not really improving, there is also the issue of increased layers etc. The NAND is getting more complicated and dense, which actually makes it harder to keep up the same speeds. So there is some improvement, but it is sacrificed in the course of the development.
That airplane seat analogy is really good. Very accurate description of the page read/flash/write cycle. Whoever in the writing team came up with that should take home extra office supplies this week.
I appreciate Buildzoid's characterization of these developments as being really exciting for "file transfer enthusiasts". And that snark is coming from someone who devotes large amounts of time to making spec numbers slightly higher.
This idea that sequential reads dont matter for gaming is gonna age poorly. People really dont understand what DirectStorage is doing and how developers will not just build their games differently, but will actually structure their data on the drives differently in order to take advantage of sequential read capabilities better.
@@maynardburger It's not even that foreign of a concept and was basically in practice until SSDs came into vogue and firmware rearranging made it impractical. Moving the head with discs was a very slow operation so as much data as possible was arranged sequentially. Some hard drive based games still do this but especially past titles had that one big file with everything in it. While fragmentation defeated the purpose, if you had that big file stored contiguously, the game was basically optimized for sequential access, at least so in far they took the optimization internally.
@@goblinphreak2132 that's a really stupid remark. Those games didn't really stutter because of drive speeds, but because of their unoptimized states and memory leaks
@@zeNUKEify even if it does come as a feature most people could just play games without it. I don't think it is a necessary feature for pc but that's just me.
@@georger5558 Wouldn't say decades. Remember how computers were 20 years ago? Much has changed. -Not only 2003 HDDs were extremely slow and low capacity (120GB, 50MB/s sequential was totally fine for the next 5 years), HDDs in general are now obsolete as a main drive even with the much better speeds compared to 20 years ago, much better used as an archive drive. -Single core CPUs were still king in 2003 with much slower frequencies. Now we have more than double the frequency with 15+ more cores. -GPUs back then were good for 1024x768 gaming. Now we have 3840x2160 gaming with much more demanding games. - OS only supported 3.3GB of RAM, with 64-bit support only being added later, supporting more RAM. - 2GB of RAM was totally fine for the masses, with more than that being of interest for professional use. 400MHz was the norm. Now we have 6000MHz. - PCIe was just coming out, with just 4GB/s on 16x. Now we have 63GB/s on 16x. Those random reads are a big problem that will need to be addressed in the future, and consoles now requiring a very fast NVMe SSD to work as intended, I see software in general get much more demanding when it comes to storage. 500MB/s SATA won't cut it in a few years.
@@georger5558 Never say never, new gens of hardware and software might have been relatively kind to storage drives, but they aren't excluded from Moore's law.
Pro tip, use a software like primocache that will use an allocated amount of ram as a ram disk to cache all your SSD's/HDD data for a few minutes while programs read and write to that data randomly. Once all the data is done being utilized in either read or write applications it will either sit there on the ram ready to be accessed again or be written to the disk sequentially! Automatically or manually.
Linus' explanation of stacks using plane seats is so easy to understand, and interactive, wish he could do some more of those explanation sections in the Vids.
This video once again reminds me of how sad I am that Intel discontinued its 3DXP SSDs. They were prohibitively expensive, but the technology addressed all of NAND's shortcomings in a way that genuinely had me excited for the for the future of storage innovation.
I think that right now even a PCIe gen 3 drive is enough for most situations including gaming, would been interesting to see a drive from that generation thrown in
Really goes to show that Intel Optane was already the pinnacle of Nvme drives... It's so unfortunate that they are no longer actively developing Optane drives. We already had the random Reads/Writes with Optane as well as insanely low latency all noticeable in everyday use like booting Windows or starting programs. It's a shame really that it never had a chance to become cheap and adopted by the market.
I think the sequential speeds of Gen 4 drives are more than enough, but it's the random speeds that we need now. Not much improvements since Gen 3 drives and unfortunately there's no other competition (Optane went away too soon sadly) to motivate manufacturers to do so.
It was directed at the enterprise market, and unfortunately the slim cost differences (not to mention the implementation/compatibility hassles) were what ultimately led to the abandonment of 3D XPoint/Optane alltogether. It was a huge hassle to use if you weren't locked down in the Intel ecosystem which definitely kept most consumers away from the start. I have no doubt it would have gotten more attention if they just marketed it to gamers with the plug-n-play hybrid nand/optane drives and made the pure optane drives not have to have special software to use them.
Optane also maintained its performance consistency pretty much indefinitely, too---it wouldn't exhaust a cache and drop lower on performance---it'd just keep on trucking. It also wouldn't wear out either. Would have loved to see something like the P5800X make its way down to the retail space--those random write and read speeds are just insane.
Having worked in laptop/desktop repair, the OEM implemtations for Optane acceleration were disastrous and I'm glad Optane is long gone. The Optane+HDD RAID0 configurations failed way too often and caused massive issues trying to save Windows installations. The less common straight Optane M.2 SSDs also seemed to fail frequently. I do wish Intel could have kept improving it on, but Intel seems to follow the Google strategy of killing new things off quickly.
I'd love to see U.2 format hit consumer level hardware. While the m.2 form factor is nice for it's simplicity and compactness, I'd love to see large capacities for cheaper that would still run on pcie instead of SATA.
U.2 already failed on consumer hardware as did SATA express, most consumer level SSDs end up in laptops so the economies of scale were with M.2 on day 1.
u.2/u.3 are sas connectors re pinned.. there is no reason on consumer level. maybe oculink looks like it... especially if they can pinout a single gen5 m.2 to 4x gen3 x1 somehow.. really they just need to put more chips on the m.2 how cheap they are but the m.2 just too small... 4tb m.2 is rare 8tb would be ideal but cost ridiculous when a 2 tb is $80 u fig a 8tb should be 320.. or less only 1 controller instead of 4... but the cheapest 8tb is $1k+ so it sux these 2tb fill up fast
@@DigitalMoonlight ... which is why he said "would love to see" instead of "am expecting" or a similar expression. Optane remains the king of consumer SSDs, and it's a shame Intel couldn't get enough funds to sustain development.
I want to see these same tests but with an Optane P5800X drive. That will show why big numbers are nice, but small numbers (latency) really matter when it comes to day to day stuff.
The random reads and writes being so slow is due to crystal disk mark not being configured correctly. The problem is that it is *CPU limited* with just 1 thread due to CPU overhead when submitting I/O tasks. Raise thread count and the random performance will be *a lot* better. For reference a 1 TB WD Black PCIe 4.0 SSD can hit 3.8 GB/s random 4kB reads and 1 GB/s random 4kB writes on Windows 10 with a R9 3900X as long as you configure crystal disk mark to use 24 threads. Sequential tests likely do not have this issue due to each I/O task being much bigger resulting in lower CPU overhead from submitting them.
I got myself a 4TB Samsung 990 recently for my new PC setup. My PC board is Gen 5 capable so I imagine I'll be looking at this SSD (or better) in the future because, you know, copying an entire 4K movie from one drive to another in 2 seconds is better than waiting an excruciatingly long 4 seconds.
So is no one gonna talk about how Linus is so refined in class, as a man of culture, that he naturally holds up his pinky finger, when he has an SSD in his hand 🫡
I wish someone would do a Gen4/Gen5/Optane comparison with direct storage. Optane may be incredibly slow on read/write vs. Gen4/Gen5, but random performance it is still king. I would be curious to see if Optane's ability to grab assets from disparate portions of the storage would help more than just raw sequential speed.
Haven't Optane been pulled from the market even in enterprise spaces? I heard Intel stopped their production because of poor performance of that product line to costs to market.
Dunno if it's just me but picture quality looks really good in this video. Color grading, focus is on point, nice looking depth of field. 12:45 It's looking extra good to me today, good job guys !
So little side note if you slow down the video at 0.25 and go around 6:11-6:14 you see the crucial t did boot edge up a hair faster but it was slower when compared to the 990 at fully rendering all of edges application’s
@@Gurj101 The only resource genuinely used more from poor optimization is storage. Both primary and secondary. It actually takes some effort and forethought to get the movement and processing of data to use more of the hardware available.
One handy trick I’ve been using since SSDs first arrived was to leave 5-10% of the drive unformatted. That way if my drive is ever ‘full’ according to the OS, it actually has some spare unused space, ensuring SLC caching and TRIM can always work properly. Helps keep your older gen 3 and gen 4 SSDs feeling snappy.
6:35 The moment you load something from the SSD, it automatically goes into RAM cache. Loading twice will never be comparing the SSD, it'll only compare the RAM.
I'm still on a regular old SATA SSD. Upgrading to a fancy drive that plugs directly into my motherboard is my next upgrade. I upgraded to a nice i5-12600K and I am looking forward to the nicer speeds of M.2, but for now, my trusty old 870 EVO does the trick just fine. At least I got rid of all my spinny drives... :D
Yep, using Primocache, CrystalDiskMark read/write scores are easily more than double the Gen5 scores using my Gen4 M.2 (and random 4k are about 10 times faster). It sounds like the big benefit of Gen5 drives are the onboard cache, which can make use of Gen5 speeds. But you can get that using Gen3 or 4 (or even SATA SSDs and hard drives) by using something like PrimoCache (assuming you have enough RAM set aside for the cache to do the task before hitting the actual uncached drive speed). But then, as the video says, it's not all about the speed.
ปีที่แล้ว +7
It would be interesting to see the difference to a Gen3 SSD. Most people upgrading their PC now likely own a Gen3 SSD. It would be interesting to know if it is worth upgrading or if a 2TB Gen 3 is still fine. Tasks would be standard daily use like MS office, browsing, editing photos and mainly gaming.
For anything you listed and even more, gen 3 is more than fine. Gen 4 and 5 are only good if you need to handle lots of storage and files every day or have VMs or something.. MS office runs fast even on Sata.. smaller games like Fortnite won't benefit from anything above 2 gb/s, and even bigger games like GTA or COD will do more than good with 4gb/s
Some Benchmarks are surfaced with latest games using direct storage feature which showed there was no big difference in loading speed between Gen3 Gen4 + Gen5 NVMe drives.
Got a reply from Crucial on your tweet saying they'll drop in May. Wondering if anyone else is going to drop sooner, I'd really like a Gen 5 M.2 for my new build I'm doing.
Maybe I should finally get an SSD for my PS5. If they're getting this much faster (at least theoretically), the Gen 4 SSDs should hopefully be getting cheaper.
@@ShadowSlayer1441 Gen 3 NVME drives are getting absurdly cheap too, if you aren't worried about crazy speeds. I picked up a 1tb Teamgroup drive for $52 a little while ago, and it works beautifully in my i3 laptop where I'm not concerned about speeds faster than a SATA drive. No DRAM cache but I've been daily driving a drive without a cache for years without issue.
@@smileyguyz yea, and honestly there is barely any noticeable difference in speed between the gens...it's like oh with the top of the line this program loads up in 2 seconds...on this old gen 3 it loads up in 3.5 seconds....i havea gen 4 mobo with a Corsair MP600 pro 2tb... 7100 read/6800 write and i feel it's overkill atm.
honestly what would probably make more sense in a compatability standpoint would be an gen5 x4 slot into a gen 3x16 slot converter. on the motherboard. Not everything is actually wired to just reduce the lane count and still run, but if a converter could take the data from a gen5 x4 slot and send it to a gen3 x16 slot, we could get 4 gen3 drives in raid 0 that will achieve the relative same performance on sequentials, and quadruple the random read/writes. But tbh if you're really looking for best day to day use case, get a bunch of sata MLC SSD's and raid 0 them because the random read/writes on good MLC drives runs circles around any TLC drive
PCI 5.0 1x lane is the same as PCIe Gen 3 4x which, if you ask me, enough for the absolute vast majority of general gamers.Would be more useful for a lot of storage too. 4x 4tb on PCIE 5.0 1x each, would theoretically be the same speed as all four of them running on 16 lanes of PCIE gen 3, with 4x lanes for each.
I really hope the labs deep dive into DirectStorage memory-to-memory feature. It allows the game devs to create a managed RAM-drive that contains strategically cached ***still compressed*** assets (not decompressed), so same as a DirectStorage SSD. Except that a DDR4 3200 RAM-drive will be far faster than any fancy super expensive PCIe drive, even with the game engine hammering it also. This memory-to-memory feature will ultimately help console performance, and I presume lower end PC hardware as well. It'll be interesting times for open world games.
It is another case of software being optimized for the worst case scenario. It is only been relatively recently that we have seen more games take advantage of regular SATA or gen 4 SSD because for years old consoles where holding back developers from implementing new tech. This is kinda the same thing, over the next 5+ years we will see more and more of these drives become part of the eco system, the next series of consoles will likely (hopefully) adopt them and then as older tech becomes less common more and more software/games can use awesome new things.
As you have the means, maybe you could make a video where you make a large ram drive, install the OS and games on it, and then bench it against a gen4 ssd. Showing how much hdd speed matters (or doesn't) with what would be a drive that is completely removed as any possible bottleneck?
What good are those sequential speeds? My Optane PCIe is still crushing these much newer SSDs in the segment where it matters most. Until they improve there, I'm not really interested, got me another two Optanes before they became unavailable!
I am more excited about fibre-optic implementation in cpus as a external data bus, registers, and cache. the problem now is designing a transistor like system for light
@@TurtleFootMining yeah I know. It's probably going to be 20 years before there is going to be consumer grade chips. Even then it's going to be longer before they can be upgradeable
@@alen2937 depending on the use of the diode, whether as a transistor or a signal receiver from the edb ( I doubt the board will be fibre optic), that will determine life expectancy of said Cpu.
Holy crap did a reviewer give gamers real meaningful information on how drives work and what the SEQ numbers mean to real world usage? Well done Linus!
I had a memory and storage intensive modeling task that the previous person at my work was running on an external HDD... Each file was 10-15 gb. Took 20 min to load run and save an interation. I installed a 4 TB Crucial pcie gen 4 m.2 drive. 4.8 GB/s or so. Took a day to transfer everything onto it from the external. Now the same simulations take about 45 seconds... HUGE time savings. I'll definitely go for faster drives as they come out fur future systems. Depending on the application it can be a game changer.
This is a great video that highlights how one needs to evaluate their workflows and where the bottlenecks are before running out and purchasing the newest hot thing. I was smiling when they striped two drives as I have also striped some of my drives. As long as you don't use spinning rust for data archival, striping is a valid configuration for getting the most out of your drives.
What I'm really wondering is if the random speed is actually mainly bottlenecked by NTFS, like it's common with pretty much everything else. Would love to see similar benchmarks done on Linux.
Comparing google image results for KDiskMark results conducted on Linux to CrystalDiskMark results conducted on Windows, the results don't seem too different at a glance.
I just thought of an idea for how to make your graphs clearer and easier to interpret quickly. I always get confused looking at the graphs flashing by quickly in the video, especially when you switch between lower is better and higher is better. (and yes I know you are supposed to pause to interpret the graphs). But what if you color-coded the graphs for which is a better result, like gold, silver, bronze. So when the graphs are flipping through quickly I can look at the color and see that it wins in that test very quickly and follow along with the video. Just an idea. Anyway good video as always.
Hey, ISA was already hotswappable. I hot swapped in a ISA soundcard on my Pentium 60 when I couldn't be bothered to shut it down first. Worked like a charm. Hot swap support is more of a recommendation.
For everyday people, not people getting Epyc, Threadripper, or Xeon, PCIE lanes are often hard to come by. You buy a board and the amount of lanes you get rarely allow you much headroom for even PCIE 4 if you include an GPU at x16. My board, granted it's AM4 only has one additional x4 PCIE 4 available and it's not even a low end board. We really need to be allowing more PCIE lanes way more than we currently do.
3:40 Thanks for reminding the Samsung Firmware issues. I hope that this company, that I love, will nevzr ever favor speed over data integrity. Fortunately I make backups but it's always a disaster to lose data.
i love that i used to wait an hour for my PC to warm up before i could use it and now a 0.5 sec increase in program load time is being called significant lol
I was also excited when I bought the Crucial T700 Gen.5, very fast, but at a great price: when you install it automatically the speed of your PCI exp 5.0 graphics card from 16x goes to 8X. For me this is inconceivable. Better 2 x 990 Pro in Raid 0 and PCI at 16x.
cool. from what you described about the cache optimaization , RAID 0 might be beneficial to maintain higher speeds for longer or at least lower the avg time to transfer files per drive in raid
12:48 Ah, this was interesting and informative for me, so basically, if I understand it, increasing the transfer speeds is simpler and therefore cheaper to do than adding more lanes that's the focus at the moment until we reach a point where adding lanes becomes simpler than making it faster at a certain point. Very cool. I never thought about that
Just about a month ago I replaced the one that came with my about 2-3 years old laptop with a new one, Gen3. The reason is the laptop's own port is of that generation. I guess it is just fine.
Are they worth it? As it all things, it depends. I have found that my file access needs are not very significant. When I went from a spinning platter drive to a SATA SSD in 2017, it was a game changer. When I went from that same SATA SSD to a PCIe Gen4 SSD this year... not really noticing any measurable improvement. But after 5+ years I felt the old SSD's time would soon be up so it was a worthwhile replacement at any rate.
Grab a PCI-e Gen 5 Hardware Raid card and put 4x of those NVME's on there in a RAID 0 configuration. as long as you have the proper motherboard and cpu you should have enough lanes on x16 for full bandwidth!
I just bought the 990 pro in 2TB size for just over $100 (I had a $40 samsung rebate). The cheapest I'm seeing the crucial for is in the low $300s. Pretty sure I made a good choice lol If you have money to burn, go for the crucial, but from all of the reviews I've seen on it, the samsung is the performance per dollar king right now since their prices just came WAY DOWN.
On the most current motherboard is Gen5 m.2 port shared with GPU, so it means that GPU will switch to 8x instead 16x. Would be interested to see difference here, it if actually slow down game? But in reality what we need is gen4 m.2 which is capable to reach max gen 4 speed in random access, it will have much much bigger impact, than this.
The B650 chipset is just taking advantage of what Linus said at the end. My motherboard has 1 gen5 nvme slot and 2 gen4 nvme slots, I can forgo sata drives thanks to this.
thanks... i was thinking that the reason to upgrade was to get 5en 5 drives.. but maybe not.. Raid 0 seemed to massively increase random reads so was totally worth it !
Wow, this is the best real life explanation of how SSD work. Fresh SSD is like empty jet, every seat is free. Once it's near full, people need to get out so others can get in.
I used a 1st Gen Threadripper's high PCI-Express lane count for this very reason - 3 M.2 drives & a PCI-Express SSD on an add in card to get performance & storage space that was not really possible any other way at the time (without significantly higher cost)
Insane speeds are nice, but I would prefer moderately fast NVMes with huge capacities, like 12TB, which are actually affordable. Even high capacity SATA SSDs still force you to rob a bank...
I could appreciate the use of half the lanes. I have an PCIe SSD from 2015 and I'm only able to use 2 SATA devices. I used to think my motherboard was defective.
That intro was... interesting...
Yeah, it was really cringe.
No it was stupid like other idiocracy USA is in....
@@GakkariThat’s why I’m subscribed. For the cheesy moments
Thats one way to put it
beautiful.
I think it would be great if you guys would color code your graphs depending on if higher or lower is better - I feel like that would make them much easier to read at a glance.
This comment needs more likes. Get on it!
Seconded. Maybe a little logo on the left or right for lower/higher or something, so you don't halve the available colors for big graphs, but that would make them much easier to read, especially just watching through the video rather than pausing to read each one in detail.
Third-ed.
Fourthed but hoping not colour-blinded!
So true considering they show their graphs for like only 2-3 seconds.
I loved the final bit about PCI 5.0, it's the most important message of the video. Desktop is starting to get the same PCI lane benefits of server-class platforms of a few generations ago, in a much more slim package. PCI 6.0 might be the tipping point of too much bandwidth, where we come out with novel ways of using it all.
I would be excited to see 2x lanes for GPU shrouds connecting over USB4 or 5, imagine how small we can get the power of a steam deck to be in another 5 years. The use wouldn't be for 16x slots, and maybe at some point 8x slots become the norm for desktop GPUS
@@endmymisery3623 8x 4.0 GPUs are already becoming the norm for the mid and lower end lineup of cards which is really good for anyone who might need a heck lot of PCIE cards for home servers. hoping that CPUs and mobos also support for 32+ lanes on consumer chips because can benefit a lot from that also.
@@Blackdune00006 8 lane GPUs aren't in any way an advantage over x16 lane cards. It only reduces flexibility. It hurts performance when installing the cards in a previous pcie gen motherboard.
It only applies sometime in the future where all devices attached via CPU lanes are gen 5. An RTX 4090 in a gen 5 slot that's been reduced to 8 lanes will work with gen 4 x 8 bandwidth.
Also add in cards for more drives are expensive. E.G a 4x m.2 nvme Gen 4 adapter.
That and motherboard manufacturers have stopped building boards with split x8 x8 CPU lanes for the most part with the demise of multi GPU.
@@justincase2312 Lower-end chipsets like B760 have like half the pci-e lanes available as Z790 so I think that's what he means
6:30 The reason why the Gen 5 drive won in the first round but lost in the second when opening PCMark 10 is because after opening PCMark 10 once, PCMark 10 is saved into the RAM as cache. So in the second run, the drives were not used, instead Windows loaded PCMark 10 from the cache instead of the drive, which is why the Gen 5 drive seemed slower in the 2nd round.
Disabling superfetch would prevent this, correct?
But then, why did the pci-4 system open it quicker at all?
you don't develop software right? cause that's not how it works
@@pabloespinoza9046 Since you have all the wisdom, why not share it instead if being sarcastic.
@@pabloespinoza9046 How are you so confidently wrong 😭
"By default, Windows caches file data that is read from disks and written to disks. This implies that read operations read file data from an area in system memory known as the system file cache, rather than from the physical disk. Correspondingly, write operations write file data to the system file cache rather than to the disk, and this type of cache is referred to as a write-back cache. Caching is managed per file object." - Windows app development documentation
SSD storage is interesting. A few years back I worked in a lab and we had a very niche use case where we needed maximum sustained throughput. This was when the 970pro was the top of the line.
We had a high speed camera setup that collected data from an experiment. It was not for particularly insane fps, but high FPS for rather insane lengths of time to capture an entire run. We'd saturate PCIe 3 for several minutes. Basically filling an entire 1TB drive per experiment run, one camera per drive and up to three cameras.
At the time, the pro series used actual MLC, not TLC they now advertise as "3 bit mlc". This meant that they didn't have the cache overflow performance hit and could happily fill at full speed.
Once to save a buck, one of my colleague's got some EVOs. They did not understand the cache behaviour and just saw that the drives had the same performance specs since w didn't need longevity. Needless to say they did not work. After a few seconds of video the collected fps would tank as in the plot in this review.
Seeing that chart reminds me why I remain bitter about the way drives are marketed to this day 😂.
Very interesting
I really liked the explanation as well as the animation for the plane passengers to explain how ssds function. I think explanations that anyone can understand help to get a lot more people into tech when it doesn't seem so intimidating. Great video!
Meh, for me it felt a bit too simplified to the point of uselessness, I don't think it helps you understand what's _actually_ going on
doesn't explain the use or need of these speeds. You can't jam many people in a F-22.
I'm glad you included some discussion about the PCIe lanes because this does seem to be a point of stagnation. We haven't seen much development on the lanes in years other than upgrades to the PCIe generation. Though when it comes to drive performance, I would have loved to see you dust off a PCIe 3.0 Optane drive and see how its random performance compares. I suspect it would still be able to keep up with the Gen 5 drive in some tests.
Its because chip and boards manufactors are foucsed on a single chip instead of breaking up that chip set and getting softward to work with dozens or 100s of chips and thousands of lanes of traffic. In the future you will find 100s if not thousands of tiny chips on a board that will function vastly faster than anything we have today. The ability to build out a chip set for a specfice task rather than using a clock speed to deligate tasks is what will speed up the process. Today we store the OS on the main drive when the OS should be flashed on to a chip that super fast. If you get what I mean. There is no reason this can not happen.
@@kameljoe21 There are lots of reasons that doesn't happen.
no need to, here is tested fact. optane, the 3dxpoint only not the hybrid, would have crushed any other m2 in all metrics except sequential speeds and keep it's performance rock stable no matter how much time the tests took. that's why the enterprise loved it. too bad intel axed it due to being complex and expensive to make.
@@marvinmallette6795 The PCIe generation advances double the base throughput each time, but that is just for straight transfer speeds. Depending on what you are doing with the data you transfer across, it might not be able to keep up with the speed of the link. That's why Linus pointed out how random I/O performance on those SSDs hasn't improved all that much.
@@marvinmallette6795 You can literally run a 3080 on PCIE 3.0 x16 slot, even 3090s suffer like 5% reduction. Just search up some tests.
Another possible test you could do for drives is verifying files of a game. I feel like that's something people actually do semi-regularly, and it would be good to know if there's an actual real world difference there. I imagine it would just be the same as sequential read/write but you never know!
Doesn't really matter if no one is going to buy it 🥺
@@Crecross yeah right this exact time right now nobody is buying it, maybe in the future when gen 5 is the standard people will want testing of the product already done
that more so has to do with your CPU since it's hashing all the files
In my experience, and after confirming many times that all my drives are working properly, there's no difference in any normal PC scenario or any gaming scenario between an 870 EVO SATA SSD and a Gen4 Aorus drive, I have 2 identical drives of each, I got the second Gen4 thinking stuff will actually make use of it, and I decided to just buy 2 bigger SATA drives and they're running everything flawlessly
It entirely depends on how many threads Steam is spawning to do verification. I don't think Steam would try to "Stress test" the system during verify task. So it might not be a valid test.
I think you made a strong point of how useful the new per-lane efficiencies of PCIe Gen5 are, technology and cost-wise.
What you didn’t address is what are the next steps (if any) that the industry might be considering for making a leap in random read/write speeds and/or cache dimensioning.
Definitely a good subject for a follow up video. See what’s out there.
How to resolve the baseline problem of random read/write speeds not improving at all.
intel optane already solved it. But it was too expensive
Although you are correct, i.e. it's not really improving, there is also the issue of increased layers etc. The NAND is getting more complicated and dense, which actually makes it harder to keep up the same speeds. So there is some improvement, but it is sacrificed in the course of the development.
@@ilovehotdogs125790 Optane was alien technology, it's a shame it died
It's called Optane!
My first attempt at messing with ram disks a couple of years ago wasn't as fast as this. Crazy how far mass storage has come
That airplane seat analogy is really good. Very accurate description of the page read/flash/write cycle. Whoever in the writing team came up with that should take home extra office supplies this week.
I appreciate Buildzoid's characterization of these developments as being really exciting for "file transfer enthusiasts". And that snark is coming from someone who devotes large amounts of time to making spec numbers slightly higher.
Buildzoid is a snark enthusiast first and foremost.
This idea that sequential reads dont matter for gaming is gonna age poorly. People really dont understand what DirectStorage is doing and how developers will not just build their games differently, but will actually structure their data on the drives differently in order to take advantage of sequential read capabilities better.
@@maynardburger It's not even that foreign of a concept and was basically in practice until SSDs came into vogue and firmware rearranging made it impractical. Moving the head with discs was a very slow operation so as much data as possible was arranged sequentially. Some hard drive based games still do this but especially past titles had that one big file with everything in it. While fragmentation defeated the purpose, if you had that big file stored contiguously, the game was basically optimized for sequential access, at least so in far they took the optimization internally.
@@goblinphreak2132 that's a really stupid remark. Those games didn't really stutter because of drive speeds, but because of their unoptimized states and memory leaks
I feel like 3gen drives will be good enough for me for at least a decade and they can easily be passively cooled.
Assuming direct storage gaming doesn’t take off
@@zeNUKEify direct Storage benefits really drop off past Gen 3
@@zeNUKEify even if it does come as a feature most people could just play games without it. I don't think it is a necessary feature for pc but that's just me.
@@georger5558 Wouldn't say decades. Remember how computers were 20 years ago? Much has changed.
-Not only 2003 HDDs were extremely slow and low capacity (120GB, 50MB/s sequential was totally fine for the next 5 years), HDDs in general are now obsolete as a main drive even with the much better speeds compared to 20 years ago, much better used as an archive drive.
-Single core CPUs were still king in 2003 with much slower frequencies. Now we have more than double the frequency with 15+ more cores.
-GPUs back then were good for 1024x768 gaming. Now we have 3840x2160 gaming with much more demanding games.
- OS only supported 3.3GB of RAM, with 64-bit support only being added later, supporting more RAM.
- 2GB of RAM was totally fine for the masses, with more than that being of interest for professional use. 400MHz was the norm. Now we have 6000MHz.
- PCIe was just coming out, with just 4GB/s on 16x. Now we have 63GB/s on 16x.
Those random reads are a big problem that will need to be addressed in the future, and consoles now requiring a very fast NVMe SSD to work as intended, I see software in general get much more demanding when it comes to storage. 500MB/s SATA won't cut it in a few years.
@@georger5558 Never say never, new gens of hardware and software might have been relatively kind to storage drives, but they aren't excluded from Moore's law.
Pro tip, use a software like primocache that will use an allocated amount of ram as a ram disk to cache all your SSD's/HDD data for a few minutes while programs read and write to that data randomly. Once all the data is done being utilized in either read or write applications it will either sit there on the ram ready to be accessed again or be written to the disk sequentially! Automatically or manually.
Linus' explanation of stacks using plane seats is so easy to understand, and interactive, wish he could do some more of those explanation sections in the Vids.
I think they had a Tech Quicky about this years ago already
I wonder if he or one of the writers came up with it :D
Stacking seats from FloatPlane
@@BastianHodapp The result is the same...a really GOOD explanation...
I wish he had CCNA and other CyberOp Certs explained his way, I would purchase them in a heartbeat.
I love how this video went into SO MUCH INFO!!! it wasn't just a "this is how it performs. The end" it really went into details. Thanks!
Indeed it was really informative and detailed and very much relating!
This is why I picked up a couple of Optane 905p drives recently; the latency and random Q1T1 rates are better than anything other than a ramdrive.
optane's still alive??
@@Amfibios Got mine from Newegg; might still be some left.
This video once again reminds me of how sad I am that Intel discontinued its 3DXP SSDs. They were prohibitively expensive, but the technology addressed all of NAND's shortcomings in a way that genuinely had me excited for the for the future of storage innovation.
I think that right now even a PCIe gen 3 drive is enough for most situations including gaming, would been interesting to see a drive from that generation thrown in
@ZaHandle Gen 3 nvme drives are way too cheap (and also easier to install) than sata or hdd too not be bought though
Would love to see some more real world/game comparisons, but with more setups like hdd vs sata ssd vs gen 3/4/5 nvme
Really goes to show that Intel Optane was already the pinnacle of Nvme drives... It's so unfortunate that they are no longer actively developing Optane drives.
We already had the random Reads/Writes with Optane as well as insanely low latency all noticeable in everyday use like booting Windows or starting programs.
It's a shame really that it never had a chance to become cheap and adopted by the market.
I think the sequential speeds of Gen 4 drives are more than enough, but it's the random speeds that we need now. Not much improvements since Gen 3 drives and unfortunately there's no other competition (Optane went away too soon sadly) to motivate manufacturers to do so.
It was directed at the enterprise market, and unfortunately the slim cost differences (not to mention the implementation/compatibility hassles) were what ultimately led to the abandonment of 3D XPoint/Optane alltogether. It was a huge hassle to use if you weren't locked down in the Intel ecosystem which definitely kept most consumers away from the start. I have no doubt it would have gotten more attention if they just marketed it to gamers with the plug-n-play hybrid nand/optane drives and made the pure optane drives not have to have special software to use them.
I still use a 905p in all my systems. Had zero reason to change them out for anything else.
Optane also maintained its performance consistency pretty much indefinitely, too---it wouldn't exhaust a cache and drop lower on performance---it'd just keep on trucking.
It also wouldn't wear out either.
Would have loved to see something like the P5800X make its way down to the retail space--those random write and read speeds are just insane.
Having worked in laptop/desktop repair, the OEM implemtations for Optane acceleration were disastrous and I'm glad Optane is long gone. The Optane+HDD RAID0 configurations failed way too often and caused massive issues trying to save Windows installations. The less common straight Optane M.2 SSDs also seemed to fail frequently. I do wish Intel could have kept improving it on, but Intel seems to follow the Google strategy of killing new things off quickly.
I'd love to see U.2 format hit consumer level hardware. While the m.2 form factor is nice for it's simplicity and compactness, I'd love to see large capacities for cheaper that would still run on pcie instead of SATA.
U.2 already failed on consumer hardware as did SATA express, most consumer level SSDs end up in laptops so the economies of scale were with M.2 on day 1.
u.2/u.3 are sas connectors re pinned.. there is no reason on consumer level. maybe oculink looks like it... especially if they can pinout a single gen5 m.2 to 4x gen3 x1 somehow.. really they just need to put more chips on the m.2 how cheap they are but the m.2 just too small... 4tb m.2 is rare 8tb would be ideal but cost ridiculous when a 2 tb is $80 u fig a 8tb should be 320.. or less only 1 controller instead of 4... but the cheapest 8tb is $1k+ so it sux these 2tb fill up fast
@@DigitalMoonlight ... which is why he said "would love to see" instead of "am expecting" or a similar expression. Optane remains the king of consumer SSDs, and it's a shame Intel couldn't get enough funds to sustain development.
I want to see these same tests but with an Optane P5800X drive. That will show why big numbers are nice, but small numbers (latency) really matter when it comes to day to day stuff.
M.2 Drives (of any gen, on the consumer side) do not have the extended ground pins that would allow them to be hot swappable.
optane will be faster in everything except sequential speeds.
The random reads and writes being so slow is due to crystal disk mark not being configured correctly. The problem is that it is *CPU limited* with just 1 thread due to CPU overhead when submitting I/O tasks. Raise thread count and the random performance will be *a lot* better. For reference a 1 TB WD Black PCIe 4.0 SSD can hit 3.8 GB/s random 4kB reads and 1 GB/s random 4kB writes on Windows 10 with a R9 3900X as long as you configure crystal disk mark to use 24 threads. Sequential tests likely do not have this issue due to each I/O task being much bigger resulting in lower CPU overhead from submitting them.
I got myself a 4TB Samsung 990 recently for my new PC setup. My PC board is Gen 5 capable so I imagine I'll be looking at this SSD (or better) in the future because, you know, copying an entire 4K movie from one drive to another in 2 seconds is better than waiting an excruciatingly long 4 seconds.
So is no one gonna talk about how Linus is so refined in class, as a man of culture, that he naturally holds up his pinky finger, when he has an SSD in his hand 🫡
So posh 😂
@@ArniesTech Ahh yes, so distingushed 😄
He's just letting the other Victorians know he is available for the night.
@@-Buckarooz Victorians see this video and be like "Pinky army, we ride at dawn!!!" 😄
Aha! Now I know why he always drops shit.
You should revisit the idea of a RAM disk in 2023, would be interesting to see how if it performs better for random reads and writes
Very badly it performs unfortunately for random reads and writes.
I wish someone would do a Gen4/Gen5/Optane comparison with direct storage. Optane may be incredibly slow on read/write vs. Gen4/Gen5, but random performance it is still king. I would be curious to see if Optane's ability to grab assets from disparate portions of the storage would help more than just raw sequential speed.
Haven't Optane been pulled from the market even in enterprise spaces? I heard Intel stopped their production because of poor performance of that product line to costs to market.
you can actually get optane drives that’ll max out pcie gen4x4 (P5800x)
@@snowwsquire As I mentioned aren't they stopped being actively produced though?
Why use optane when you can use DDR5 and as much real RAM as you need and more?
@@snowwsquire you can. If you have Linus cash. Us mere mortals cannot :(
Dunno if it's just me but picture quality looks really good in this video. Color grading, focus is on point, nice looking depth of field. 12:45 It's looking extra good to me today, good job guys !
just you
So little side note if you slow down the video at 0.25 and go around 6:11-6:14 you see the crucial t did boot edge up a hair faster but it was slower when compared to the 990 at fully rendering all of edges application’s
That aeroplane seat animation is the best analogy I’ve seen for how SSDs work 😊
Entirely agrees, even the flight attendant crew being your controller was VERY neat information to relations.
Nice, I hope more apps and games will start using this performance.
I hope that Doom II will load in 8 ns
actually they will just make worse software to increase the resources utilised
@@Gurj101 The only resource genuinely used more from poor optimization is storage. Both primary and secondary. It actually takes some effort and forethought to get the movement and processing of data to use more of the hardware available.
One handy trick I’ve been using since SSDs first arrived was to leave 5-10% of the drive unformatted. That way if my drive is ever ‘full’ according to the OS, it actually has some spare unused space, ensuring SLC caching and TRIM can always work properly. Helps keep your older gen 3 and gen 4 SSDs feeling snappy.
will this make it last longer tho ?
Or just look how full your drive is lol
They do that by default?
Just format all of it and don't fill up more than 90% lmao
SSDs do that out of the box.
6:35 The moment you load something from the SSD, it automatically goes into RAM cache. Loading twice will never be comparing the SSD, it'll only compare the RAM.
Maybe im missing the point but what is the practical use of gen 5/gen 4 drives outside creators moving huge files around
I'm still on a regular old SATA SSD. Upgrading to a fancy drive that plugs directly into my motherboard is my next upgrade. I upgraded to a nice i5-12600K and I am looking forward to the nicer speeds of M.2, but for now, my trusty old 870 EVO does the trick just fine. At least I got rid of all my spinny drives... :D
Not sure if M2 can actually be called an upgrade, at least for gaming
@I killed that beard guy yeah, most definitely
Going from Sata to M.2 is an upgrade in itself but afterwards it's so fast you can't notice an important difference
@I killed that beard guy Oh that's plenty good 👍
i actually still use my old spinny drive cause i am too lazy to transfer all the documents and other stuff on there over to a new drive
Nice video, Linus/LAB! I use Primocache with my HDDs. It uses RAM to Cache with about 90% hit rate. It works great!
Yep, using Primocache, CrystalDiskMark read/write scores are easily more than double the Gen5 scores using my Gen4 M.2 (and random 4k are about 10 times faster). It sounds like the big benefit of Gen5 drives are the onboard cache, which can make use of Gen5 speeds. But you can get that using Gen3 or 4 (or even SATA SSDs and hard drives) by using something like PrimoCache (assuming you have enough RAM set aside for the cache to do the task before hitting the actual uncached drive speed). But then, as the video says, it's not all about the speed.
It would be interesting to see the difference to a Gen3 SSD. Most people upgrading their PC now likely own a Gen3 SSD. It would be interesting to know if it is worth upgrading or if a 2TB Gen 3 is still fine.
Tasks would be standard daily use like MS office, browsing, editing photos and mainly gaming.
For anything you listed and even more, gen 3 is more than fine. Gen 4 and 5 are only good if you need to handle lots of storage and files every day or have VMs or something.. MS office runs fast even on Sata.. smaller games like Fortnite won't benefit from anything above 2 gb/s, and even bigger games like GTA or COD will do more than good with 4gb/s
OMG if that wasn't the best description of sequential vs. random access reads/writes I've ever encountered
Some Benchmarks are surfaced with latest games using direct storage feature which showed there was no big difference in loading speed between Gen3 Gen4 + Gen5 NVMe drives.
Got a reply from Crucial on your tweet saying they'll drop in May. Wondering if anyone else is going to drop sooner, I'd really like a Gen 5 M.2 for my new build I'm doing.
Maybe I should finally get an SSD for my PS5. If they're getting this much faster (at least theoretically), the Gen 4 SSDs should hopefully be getting cheaper.
Gen 4 nvme ssds are getting sooo cheap
@@ShadowSlayer1441 Gen 3 NVME drives are getting absurdly cheap too, if you aren't worried about crazy speeds. I picked up a 1tb Teamgroup drive for $52 a little while ago, and it works beautifully in my i3 laptop where I'm not concerned about speeds faster than a SATA drive. No DRAM cache but I've been daily driving a drive without a cache for years without issue.
just get things when you need them, no point trying to predict the future
If your going to get a one of these Gen 5 drives, you won't see any benefit since the PS5 is limited to like I believe its 5500 mbs.
@@smileyguyz yea, and honestly there is barely any noticeable difference in speed between the gens...it's like oh with the top of the line this program loads up in 2 seconds...on this old gen 3 it loads up in 3.5 seconds....i havea gen 4 mobo with a Corsair MP600 pro 2tb... 7100 read/6800 write and i feel it's overkill atm.
5:00 CGP Grey would like to have a word
thought the exact same thing 😅
honestly what would probably make more sense in a compatability standpoint would be an gen5 x4 slot into a gen 3x16 slot converter. on the motherboard. Not everything is actually wired to just reduce the lane count and still run, but if a converter could take the data from a gen5 x4 slot and send it to a gen3 x16 slot, we could get 4 gen3 drives in raid 0 that will achieve the relative same performance on sequentials, and quadruple the random read/writes. But tbh if you're really looking for best day to day use case, get a bunch of sata MLC SSD's and raid 0 them because the random read/writes on good MLC drives runs circles around any TLC drive
PCI 5.0 1x lane is the same as PCIe Gen 3 4x which, if you ask me, enough for the absolute vast majority of general gamers.Would be more useful for a lot of storage too. 4x 4tb on PCIE 5.0 1x each, would theoretically be the same speed as all four of them running on 16 lanes of PCIE gen 3, with 4x lanes for each.
This would made way more sense from a real life practicable standpoint and an evolution away from SATA.
I seriously hope that the lab does things differently 😂. Great Video Linus!
Poor limited Gamer Suckux. Only does some certain bs on apple's abundant trash. Steve can beg from his daddy Linus.
I really hope the labs deep dive into DirectStorage memory-to-memory feature. It allows the game devs to create a managed RAM-drive that contains strategically cached ***still compressed*** assets (not decompressed), so same as a DirectStorage SSD. Except that a DDR4 3200 RAM-drive will be far faster than any fancy super expensive PCIe drive, even with the game engine hammering it also. This memory-to-memory feature will ultimately help console performance, and I presume lower end PC hardware as well. It'll be interesting times for open world games.
?
It is another case of software being optimized for the worst case scenario. It is only been relatively recently that we have seen more games take advantage of regular SATA or gen 4 SSD because for years old consoles where holding back developers from implementing new tech. This is kinda the same thing, over the next 5+ years we will see more and more of these drives become part of the eco system, the next series of consoles will likely (hopefully) adopt them and then as older tech becomes less common more and more software/games can use awesome new things.
As you have the means, maybe you could make a video where you make a large ram drive, install the OS and games on it, and then bench it against a gen4 ssd. Showing how much hdd speed matters (or doesn't) with what would be a drive that is completely removed as any possible bottleneck?
What good are those sequential speeds? My Optane PCIe is still crushing these much newer SSDs in the segment where it matters most.
Until they improve there, I'm not really interested, got me another two Optanes before they became unavailable!
nice, im always hyped when there is a new Linus video!
I am more excited about fibre-optic implementation in cpus as a external data bus, registers, and cache. the problem now is designing a transistor like system for light
Photonics are 100x to 1000x faster than electronics. It’ll be interesting to have CPUs with 300GHz speeds 😮
Your talking 10+ years from that even being viable
@@TurtleFootMining yeah I know. It's probably going to be 20 years before there is going to be consumer grade chips. Even then it's going to be longer before they can be upgradeable
the laserf diode will dia and out goes your processor unless they make it replaceable
@@alen2937 depending on the use of the diode, whether as a transistor or a signal receiver from the edb ( I doubt the board will be fibre optic), that will determine life expectancy of said Cpu.
Holy crap did a reviewer give gamers real meaningful information on how drives work and what the SEQ numbers mean to real world usage? Well done Linus!
I had a memory and storage intensive modeling task that the previous person at my work was running on an external HDD...
Each file was 10-15 gb. Took 20 min to load run and save an interation.
I installed a 4 TB Crucial pcie gen 4 m.2 drive. 4.8 GB/s or so.
Took a day to transfer everything onto it from the external.
Now the same simulations take about 45 seconds...
HUGE time savings.
I'll definitely go for faster drives as they come out fur future systems. Depending on the application it can be a game changer.
This is a great video that highlights how one needs to evaluate their workflows and where the bottlenecks are before running out and purchasing the newest hot thing. I was smiling when they striped two drives as I have also striped some of my drives. As long as you don't use spinning rust for data archival, striping is a valid configuration for getting the most out of your drives.
Dang it Linus I just bought my first Gen 4 SSD this week with a PCI Express Gen 4 capable motherboard. Already obsolete 🤣
Imagine what I feel with my gen 3 motherboard that I bought for thousand of bucks not too long ago. 🙄
Fast storage makes a huge difference in VR frame times.
I assume you deliberately wrote frame times instead of frame rates correct??
(Since they're two completely different things.)
@@PSYCHOV3N0M yes the amount of ms
Raid 0 is actually excellent for gamers. I've been using it with PCIe gen 3 NVMe SSDs since 2018 and haven't had a failure yet.
That airplane seating analogy was probably the best way I've ever heard SSD writes explained. Cheers!
What I'm really wondering is if the random speed is actually mainly bottlenecked by NTFS, like it's common with pretty much everything else. Would love to see similar benchmarks done on Linux.
Comparing google image results for KDiskMark results conducted on Linux to CrystalDiskMark results conducted on Windows, the results don't seem too different at a glance.
I just thought of an idea for how to make your graphs clearer and easier to interpret quickly.
I always get confused looking at the graphs flashing by quickly in the video, especially when you switch between lower is better and higher is better. (and yes I know you are supposed to pause to interpret the graphs).
But what if you color-coded the graphs for which is a better result, like gold, silver, bronze. So when the graphs are flipping through quickly I can look at the color and see that it wins in that test very quickly and follow along with the video.
Just an idea. Anyway good video as always.
Even just having a big arrow pointing in the "is better" direction would make it easier. Color coding gets hard once you factor color blindness in.
Day 3 of asking Linus to build A 600 dolar Gaming pc in 2023
Bump
Hey, ISA was already hotswappable. I hot swapped in a ISA soundcard on my Pentium 60 when I couldn't be bothered to shut it down first. Worked like a charm.
Hot swap support is more of a recommendation.
For everyday people, not people getting Epyc, Threadripper, or Xeon, PCIE lanes are often hard to come by. You buy a board and the amount of lanes you get rarely allow you much headroom for even PCIE 4 if you include an GPU at x16. My board, granted it's AM4 only has one additional x4 PCIE 4 available and it's not even a low end board. We really need to be allowing more PCIE lanes way more than we currently do.
Windows was made to read and write in random not sequential that's the bottleneck
Too bad my internet only 8mb/s
Loved the airplane bit to explain the writing to drives, made it extremely clear!
3:40 Thanks for reminding the Samsung Firmware issues. I hope that this company, that I love, will nevzr ever favor speed over data integrity. Fortunately I make backups but it's always a disaster to lose data.
bro you cute
Tf you on?
0:01 No Linus, that's a M.2 SSD.
i love that i used to wait an hour for my PC to warm up before i could use it and now a 0.5 sec increase in program load time is being called significant lol
I was also excited when I bought the Crucial T700 Gen.5, very fast, but at a great price: when you install it automatically the speed of your PCI exp 5.0 graphics card from 16x goes to 8X. For me this is inconceivable. Better 2 x 990 Pro in Raid 0 and PCI at 16x.
cool. from what you described about the cache optimaization , RAID 0 might be beneficial to maintain higher speeds for longer or at least lower the avg time to transfer files per drive in raid
I love the “airplane seats” explanation
To be fair, the plane explanation for trim was really good
5:24 - orange blobs sneakily switching seats in the end 😅
"A hare among tortoises, a star among planets, a diamond among... roughs."
Who wrote this? That's genius.
12:48 Ah, this was interesting and informative for me, so basically, if I understand it, increasing the transfer speeds is simpler and therefore cheaper to do than adding more lanes that's the focus at the moment until we reach a point where adding lanes becomes simpler than making it faster at a certain point. Very cool. I never thought about that
me, with only old Sata SSDs and older Sata HDDs "hmmm, Gen5 drives need to really step it up if it wants to impress me"
@5:23 - Hey! #2 and #3 switched seats!
Just about a month ago I replaced the one that came with my about 2-3 years old laptop with a new one, Gen3. The reason is the laptop's own port is of that generation. I guess it is just fine.
Are they worth it? As it all things, it depends.
I have found that my file access needs are not very significant. When I went from a spinning platter drive to a SATA SSD in 2017, it was a game changer. When I went from that same SATA SSD to a PCIe Gen4 SSD this year... not really noticing any measurable improvement. But after 5+ years I felt the old SSD's time would soon be up so it was a worthwhile replacement at any rate.
On AM5 systems, you can connect 2 SSDs directly to the CPU if the motherboard doesn't have integrated Thunderbolt.
The plane seating analogy was perfect
Really love the plane boarding analogy! :) Totally stealing that
Tips are rare on this channel. Reminds me of the late night infomercials pimping junk. I mean the store is like Walmart, yet the prices are skyhigh.
Here I am with my Gen3 drives waiting a maximum of about 6 seconds for my favorite games to load. I'm good. I'll wait until I'm forced to upgrade.
Random writes and read was really good explained! Sweet, now I know how it works!
Grab a PCI-e Gen 5 Hardware Raid card and put 4x of those NVME's on there in a RAID 0 configuration.
as long as you have the proper motherboard and cpu you should have enough lanes on x16 for full bandwidth!
I just bought the 990 pro in 2TB size for just over $100 (I had a $40 samsung rebate). The cheapest I'm seeing the crucial for is in the low $300s. Pretty sure I made a good choice lol If you have money to burn, go for the crucial, but from all of the reviews I've seen on it, the samsung is the performance per dollar king right now since their prices just came WAY DOWN.
On the most current motherboard is Gen5 m.2 port shared with GPU, so it means that GPU will switch to 8x instead 16x. Would be interested to see difference here, it if actually slow down game?
But in reality what we need is gen4 m.2 which is capable to reach max gen 4 speed in random access, it will have much much bigger impact, than this.
The B650 chipset is just taking advantage of what Linus said at the end. My motherboard has 1 gen5 nvme slot and 2 gen4 nvme slots, I can forgo sata drives thanks to this.
thanks... i was thinking that the reason to upgrade was to get 5en 5 drives.. but maybe not.. Raid 0 seemed to massively increase random reads so was totally worth it !
Wow, this is the best real life explanation of how SSD work. Fresh SSD is like empty jet, every seat is free. Once it's near full, people need to get out so others can get in.
I used a 1st Gen Threadripper's high PCI-Express lane count for this very reason - 3 M.2 drives & a PCI-Express SSD on an add in card to get performance & storage space that was not really possible any other way at the time (without significantly higher cost)
I bought a $25 Crucial 500GB Gen3 SSD today from Amazon. I remember they were riotously expensive and I set them aside until today.
1:32 its dope how you got Michael J Fox to do the hand modeling here
Insane speeds are nice, but I would prefer moderately fast NVMes with huge capacities, like 12TB, which are actually affordable. Even high capacity SATA SSDs still force you to rob a bank...
That airplane analogy animation was great. Well done team
I could appreciate the use of half the lanes. I have an PCIe SSD from 2015 and I'm only able to use 2 SATA devices. I used to think my motherboard was defective.
2:59 Linus sounded like a rapper there for a second.