Phison Max14um and the State of Gen 5
Вставка
- Опубліковано 21 кві 2024
- Wendell is once again testing the limits on the progression of the future, this time with the Max14um from Phison. How much will this change the industry?
********************************
Check us out online at the following places!
bio.link/level1techs
IMPORTANT Any email lacking “level1techs.com” should be ignored and immediately reported to Queries@level1techs.com.
-------------------------------------------------------------------------------------------------------------
Music: "Earth Bound" by Slynk
Edited by Autumn - Наука та технологія
Dear editor, you read Wendell thought flow so well that it's as if he himself edited the video. The perfectly sequenced jump cuts between tangential and main lecture, for example. Quality work, in fact, is there a way I could buy you a beer?
Someday... Gen5 NVMe on Gen5x1 slots will be an economical option.
(Just as Gen4 drives do 'fine' on Gen4x1)
Gen5x1 is (Near) bandwidth equivalent to Gen3x4, and there's few drives that can *sustain* saturating that.
Other than chipset/switch added latency, you still get the benefits of fast access times, small transactions, etc. Plus, theoretically less load (heat) on the Controller.
Wendell is the kind of guy I want so much success for
Imho we need a lot more 1-lane SSDs with higher capacity than we need 10+GB/s speed SSDs that need active cooling and a heatsink. This is getting ridicolous, only people in desktops can use that
Second this. Think of the 5.25 expansion options you could get with that as well. I don't get why they don't release a Gen 4 drive with 2 lanes when they go to PCIE 5 etc. You'd barely have to alter anything on the drive
This will get even more clear with PCIE 6 that is already finalized and PCIE 7 that is set to be finalized next year. A single lane of PCIE 6 is equivalent to 4 lanes of PCIE 4. That is about 7.5GBps. For consumer use cases that is plenty and I think we would rather be able to connect more devices since we only have about 24 PCIE lanes.
Its really hard to expand storage in consumer systems. Not only there is not many pcie lanes there are also bifurcation limits. You're often capped at 2-3 m2 drivers if you have GPU already taking one slot. And sata ssd's are going extinct not to mention they're same price as m2 nvme drive but deliver 1/5th of what even old pcie3.0x4 drive does on sequential transfers : /
We seriously need cheap active PCIE4.0 @ x8 card that would bifurcate to multiple x2 links for m2 nvme drives.
@@chaosfenix I expected to be there with 5.0, as single lane = 3.0x4 where most of us were first introduced to NVMe, but rarely could cap it.
@@ChrispyNut Yeah I have been thinking this for a while as well but I started getting more vocal about it when companies started slapping big heatsinks on SSDs with PCIE 4.0 and now that they are expecting use to put stupid little fans on top of those heat sinks with PCIE 5.0 I am practically fuming.
Leve1Linux video about SR-IOV and now a juicy storage vid.. i'm getting over my Monday blues :D
Question: could we get U.3 (SFF-TA-1001) to replace SATA on motherboards? Controller as part of the southbridge?
It seems insane we're limited to 6x sata at 500MB/s each when in the same space you can have 4x U.3 based drives at 7880MB/sec each
OR 16 drives using x1 each while still being nearly 4x faster than SATA, Instead we get M.2 which is 2x larger for 1 single drive!
This blows my mind.
MCIO may slowly move in.
SlimSAS would be cool too...
@@MazeFrame We'll probably get a single SlimSAS and the board makers will pat themselves on the back for a job well done.
MCIO would be better... but again we'd get one and thats it.
I want (as a server noob) a compact connector that does NVME over PCIE, U.2 and U.3 do that. My thoughts are...
-SATA *so* old now and even slower than most USB its time to leave it like we did with IDE (maybe 1 for legacy).
-PCIE6 and PCIE7 use different signalling to PCIE4 and PCIE5.
...so it seems to me we're at a natural pivot point.
Then once PCIE6> with the new signalling we see wider adoption of U.3 (U.4?).
Currently theres no point having an ATX rig unless you have threadripper because you don't have the PCIE lanes to run a Quad M.2 card on mainstream.
This would mean we don't have to do that.
The B650I aorus could trade the rear M.2 slot for U.3 giving us...
1 pcie5x16 gpu (CPU)
1 M.2 pcie5x4 (CPU)
1 M.2 pcie4x4 (CPU)
2 U.3 pcie4x4 (chipset)*
*which can feed 2 drives upto 7.88GB/s or 8 drives upto 1.97GB/sec.
today.
The real issue here is the extreme limitation of PCIe lanes on consumer CPUs. When NVMe drives use lanes four at a time, and your GPU takes 16...that's very little room for "lots of drives". Most consumer chips have two dozen or less lanes total, plus a few slower ones through whatever chipset you have on the mobo.
U.3 is basically a different connector for PCie x4 so on most consumer platforms you don't have enough pcie for more than 1 or 2 of those without doing a clowncar and connecting them all to chipset that goes through a pcie x4 bottleneck to the CPU.
I'm 100% fine with allowing full retar mode bifurcation and go down to 16 x1 ssds on a single x16 slot instead of x4x4x4x4 so only 4 drives, but we also need SSDs that save money by using a single lane too
Thanks Wendel, you answered a question that was giving me trouble figuring out
I just want 2-lane Gen 5 M.2 slots/drives so that we can get Gen 4 speeds and either preserve lanes or reduce motherboard costs.
That said, I dream of the drives that are fast enough to give me smooth playback on super dense volumetrics in blender.
Those actually exist now - Samsung 990 EVOs.
CPUs are being released so fast I am behind in my building.
Yeah same here. I think I'm finally gonna give my am4 system it's last update with a 5950x here soon. I work in audio so that chip would be good for years and years to come. Lol
Amazing video. Once again.
I'm glad this was mostly a "Talking Head" video as I just returned from an eye exam with full dilation...
I wish Optane would return with even faster random access and stuff
same here :/
The optane p1600x and p5800x are still CPU bound in random performance. With overclocking on my 13th gen CPU I'm getting 475MB/s qd1 random reads on my optane p1600x. Insane, puts the optane in the video in an entire class below itself at that point
I would say PCIe3 x4 is good enough for 90% of consumers. Every office workload is completed in a fraction of a second.
Pcie 2.0 x4 is good for me.
Which is why I expected 5.0x1 to become a thing, but 🤷♀
The rise of SLC over DRAM has been disappointing…
they do two different things, no SSD ever had 10+ GB of DRAM write cache
@@marcogenovesi8570 I wouldn't mind both, tiered cache
@@marcogenovesi8570my man, I meant for caching.
DRAM > NAND
but the industry can't afford to sell things that last forever.
and no i don't care about the data retention, if i'm using an SSD or ramdisk i run rsync and back up to a hard disk because i have a slightly unreasonable distrust of solid state storage
@manitoba-op4jx yes cuz everyone runs rsync and whatever
I want to see how enterprise drives act in a gaming situation. Or on a daily workload Are they better?
Hey Level1, what temps can we expect from current Gen5 Ssd's while passively cooled, for example, in my Slot 1 of my Aorus Master B650E motherboard with adequate cooling.
The highest SD and Cfast cards are in Cinema Cameras. Blackmagic, Arri, Red. They only need a certain speed, even if it is 12K, 16 bit. Each camera has compression on-the-fly, so only a certain amount of speed is needed.
did you found any CXL 1.1 or CXL 2.0 devices tu run over PCIE Gen 5 ?
love storage videos, am practically only here for them, recently got my hands on some P5800X drives for an insane price, still can't believe i have them here at home
That PCIE to Nvme adapter card is also "sold" with a gen5 + gen4 slot by Asus.
But only on the old version of the Z790 Hero board...for whatever reason.
64 gig, 4 drives with 8 gig cache each. Rollin withouta colon
So if low latency is what is needed someone should make a PCIe gen 5 x4 to 4x PCIe gen 3 x4 riser, with an m.2 to MCIO adapter a gen 5 m.2 slot could provide full bandwidth to four gen 3 Optane drives, in either raid 0 or 10.
Is that ASUS PCIe Gen5 x8-to-only-1xM.2 adapter maybe something influenced by Intel where there is no PCIe Bifurcation for the 16 PCIe Gen5 lanes on their current socket 1700 platform?
Yep you've nailed it exactly
Sucks balls.
Yea intel limits it to x8x8 :(
I'm sure they could use an extra chip so you could use 2 drives sharing the x8 bandwidth but it would be more expensive
@@Level1Techshow are both of your comments from a month ago!?!?
My initial comment wasn’t worded well; what I meant was why TF are they wasting a PCIe x8 AIC for only a single x4 M.2 SSD. They should have either just used a x4 AIC design or kept the x8 PCB for two and not just one M.2 SSD, which could be usable with platforms that do PCIe Bifurcation (and Intel doesn’t have such a platform in the consumer market at the moment), but then ASUS would have practically released something “great” for AM5 and Intel might have not liked that. Running multiple slots with x8/x8 has nothing to do with PCIe Bifurcation, that’s just regular muxing. It seems S. 1700 Intel CPUs are incapable of PCIe Bifurcation in general.
What I'd like to know is how the latency for HMB drives is these days? Is that something you have to worry about or can you go DRAM-less in most cases?
Ive gotten my optane p1600x to 475MB/s qd1 random reads and 365MB/s qd1 writes. Almost double the performance of your optane it's crazy
I'm gonna laugh when we get to the point M.2 drives run so hot their heatsinks make them as big as 2.5" drives, defeating the whole purpose of the form factor.
We are already at that point
I'm just going to use Gen 4 until they have decent heatsinks and no fans
@@chronometer9931 Even Gen3 is still sufficient. There is no need to get above Samsung EVO plus level performance for 99.99% of use cases. I would much rather have stability and longevity than speed, especially since almost no software even handles multiple GBps except for direct file transfers (most videographers are transferring from SD cards or still using old usb/thunderbolt protocols, making it even more moot).
Well, even though he didn't appear, despite my expectations ... Hi, Allyn o/
Creating an explicit ram drive in Windows is still dark magic requiring 3rd party apps where its 1 command in Linux. No hoping that Windows caches the read. Then you can just configure apps to use that.
And DDR5 capacity is increasing nicely. It's a pricey option, but a potentially viable one, depending on your use case.
Primo Cache is probably the best for windows. Also SSD cache
Although I am on AM5, I still have PCIe 4 NVMEs in my machine. Not sure when I will upgrade to gen 5.
Speaking for myself, whenever it becomes the cheapest $/TB. Looking at pcpartpicker right now, it oscillates between gen3 and gen4, so it's probably going to be a while.
Can you do a exotic ram disk video? Something where I’m importing gigabytes of Revit files and thousands of assets and billions of polygons in a 3D environment
so... even the fastest gen 4 drives do not suffer from these issues? Like, the chips/controller/etc are actually designed to work at those speeds of 5-7.5GB/s?
I hope someone discovers something way faster than nand flash instead of just adding faster pcie generations. Ssd's seem to be required for an os drive at this point, but sustained real world performance of any ssd I've used is not very impressive when moving lots of data.
I mean, there *is* still Optane. Technically. You can still buy it new, even if Intel isn't making it anymore, and it's expensive. That'll get you the performance you're craving. Otherwise, nothing stops you from putting a sizable enterprise SSD into your desktop system at home, and while it won't have the latency of Optane, it'll still be plenty fast, even when it fills up. Ofc, the downside is price, but you do get what you pay for.
Try some enterprise server oriented SSDs if you haven't, they're made for sustained workloads. Most consumer/client drives, even high end ones are only good for shallow, short burst loads. You might get ever-so-slightly less max scores on artificial benchmarks with enterprise ones, but they're a gajillion times more consistent in both throughput and IOPs.
Optane gang. I can't ever go back to a regular SSD.
I have PTSD from small fans . The quality and bearings is usually terrible because it's typically a 3-cent-made-in-China fan, which means after a year or two you will hear this high pitch frequency rattle from them. This also goes for any monitors with fans in them, just a tip.. avoid them at all costs if you can. The only small fans that are acceptable are the high pressure server fans, but we all know how loud they are..
Waiting to see Samsung release their own Gen 5 Pro SSD but would be amazing to see them release their Z-NAND SSD though.
I used to think the PCIe versions were a marketing meme until I upgraded from PCIe 3 to 4
Are these Gen 5 better than Intel Optane for daily use?
No optane is way better. For daily use gen 5 is barely better than gen 4. Not only are most workloads random in the real world, but the sequential ones are still done at such small block sizes that optanes latency also makes it faster sequentially in the real world too lmao.
8:10 - ??? Maybe poor support for UMA, Apple has it... but you don't feel the same benefit as if Windows did (assuming parts choice).
Mini-me vent with a connector to cool the m.2 disk, awww so cute and so funny. Having PTSP flasbacks of the failing mobo southbridge mini vent full of dust whining noise.
But the bigger question is if you aren't doing some specialized task that needs batshit amounts of I/O do you even need Gen 5 given how fast Gen 4 is? It reminds me of the current core wars on the desktop, how many normal folks are going to come up with enough work to max out an6 core, much less some 20 core monster, and how many are going to be able to even tell between a decent Gen 4 and Gen 5 without a stopwatch?
Yeah I wish they would use the increase in bandwidth for reducing the number of lanes needed. Just make gen5x2 M.2 slots, it as fast fast as gen4x4.Now you can finally give me back some of my PCIe slots - or put down double the amount of M.2 slots, so can build storage arrays if that's what you want. If you need fast, just use multiple drives in a RAID0.
@@Hugh_I Hell I'm running Windows on a Gen 3 as I found with its nice fat DRAM cache when I benched compared to my Gen 4 I honestly couldn't tell without looking at a benchmark screen so I just kept my Gen 4 for games.
So I agree 110%, just give me less lanes on Gen 5 so I can have quad NVMEs without needing to use a PCIe slot or buying a batshit insane "prosumer" board.
Yeah the gen is just irrelevant. My gen 3 optane p1600x beat my flagship gen 4 drive in load times relatively significantly (about 15-30%) in every single game/app I tested. 4x worse sequential speed btw, what a useless metric 😂.
It's not even necessarily that the normal gen 4 ssds are so fast, it's literally that gen 5 has improved so little where it counts.
Hell the 990 pro still has the best latency out of any nand drive despite being gen 4 and like 2 years old. Gen 5 isn't even the lowest latency 😂
Wouldn't the speed of the SD card be the bottleneck here, no matter how much data you're transferring?
Still hardly utilizing all my pcie3 speeds. Don't tell me they made two more when i wasn't looking
I'm hesitating on getting a Crucial T705 2TB SSD non Heatsink, ive got a GIGABYTE X670E AORUS MASTER rev.1.x Mobo which has a big M.2 Heatsink on the top M.2 PCIe 5.0 slot. My PC case is a COOLER MASTER COSMOS C700M with 3x 140mm front panel intake fans and 1x 140mm back panel exhaust fan. My GPU is a vertically mounted with a riser cable close to the case side panel PowerColor RX 7900 XTX Red Devil Limited Edition. My CPU is a Ryzen 7950X3D with a 420mm AiO ARCTIC Liquid Freezer III ARGB and my RAM is 128GB 4x 32 G.SKILL Trident Z5 RGB White 6400MHz CL32-39-39-102 1.40V XMP. My current storage is 3x 1TB M.2 NVMe SSDs + 1x 512GB HDD.
Should i get a Crucial T705 2TB? 🤔
The Crucial T705 is slightly faster than the Crucial T700
282.63 isn't it 3DXPoint level of speed?
The use case is not for 'consumers' its always about "pushing the limits" of what can be done, so eventually someone 'might' find a use case like why bother adding a screen to a phone or making a computer portable (laptop), at some point everybody producing the parts hopes someone 'else' will make them more profitable.
Why are drives sooooo expensive? Feels like $/TB has gone up in the last 12 monrhs, and people selling "refurbished" drives are taking the piss... more than half the price of new.
Latency over peak sequential output anytime anyday !
i have yet to see a software loads at more than 1-2GB a sec
PCI Gen 5 just isn't ready for prime time.
I don't think it is a problem with the PCIE standard but how it is implemented. Vendors are applying the lane speeds to physical connection standards that are 21 years old at this point. Personally I would love to see these connection speeds but with interfaces that were more versatile like I laid out in my comment.
PCIE 5 appears to have a lot more signalling issues than PCIE 4. A lot of manufacturers have had to shorten traces to maintain signal integrity. It definitely brings about a lot of engineering issues that don't seem to have been worked out reliably yet
@@brandonrattray8439 Yep, those traces are antennas, and you need to be *really* careful about impedance matching. Digital signals are hitting analog domain problems.
@@brandonrattray8439 In servers they just use a lot of cables, which seem to work alright.
@@brandonrattray8439 Pcie6 seems to address the signaling issues with regards to CXL3.0
Personally I think that the large heat sinks and fans just shouldn't be a thing. If it overheats too much at those speeds then we should just halve the lanes to 2 and figure out somewhere else to use those lanes. Lord knows how limited we are on consumer platforms with PCIE lanes. Ideally I would like to see an update to the actual PCIE slot standard.
It doesn't have to be exactly like this in that I don't care about the specifics like the connector type but I think I would like this architecture. It would be something like an MCIO connector that would only have 4 lanes by default. That is it. No more than that would be allowed in the connector. Each individual connector would be specced to provide power between 50-100W. I don't care the specific range. Just that it should be able to provide up to a specified power. You would still support pcie bifurcation which means you could then turn a 4 lane port into a 2x2, a 2x1x1, or a 1x1x1x1. This could be amazing for addin cards as if you wanted to add a bunch of PCIE devices you would simply assign a single pcie lane to them. Honestly not too much different here from the current spec. Here is where it would get spicy though. Part of the spec would be spacing between each individual MCIO connector. The reason for that is because you would also allow not only for the bifurcation of the slot but for the combination of the slots as well. Single mode would be the default but you could choose to combine up to 4 of the slots together as well in the bios. This would mean that you could still have devices connect to up to 16 PCIE lanes if you wanted but if you didn't then you would simply have 4 individual MCIO connectors you could direct attach to instead. It would be hugely more versatile. Also you would get your greater power delivery in that a more power hungry device using all 4 connectors would be supplied with like up to 200-400W of power directly. Sure you are going to have devices, especially GPUs, that still need additional power but that should be rare if they could work with up to 400W.
I think you could even allow some backwards capability if you made available an adapter to go from the 4 MCIO connectors and PCIE. Then you would just need to provide cheap standoffs for the screws at the back. This wouldn't be a problem for most cards but if you had a chonker like a 4090 you could have z height issues in the case. For most regular cards it wouldn't be an issue though and the issue would go away eventually as people switched to the new standard.
I like your proposal so much. The introduction of M.2 on desktop MoBos was really a step in the wrong direction. Had they just made NVMEs PCIe add-in cards for the existing slots and Mobos with a number of 4x slots become the standard, it would have been way more flexible and prevented issues that now crop up. One could easily use those lanes for either storage or other add-in cards. We also wouldn't have to deal with ridiculous tall cooler towers and tiny fans on them. Nor with space constraints on the M.2 drives themselves. Nor with those M.2 slots plastered all over the Mobo to make them fit somehow, and squeezed in between PCIe slots where you can't reach them when add-in cards are installed. Your solution would of course be even better.
@@Hugh_I Yeah I like NVME drives but honestly m.2 wasn't the best connector for it. Realistically u.2/u.3 should have been the connector by simply replacing the SATA we were used to. I think part of the issue was that we were stuck on PCIE 3.0 for so long. Now that we are getting PCIE revisions again I really wish we would take a generation or 2 to simply half the lanes when we double the speeds. I mean 512GBps for a x16 PCIE 7.0 card will just be kinda dumb when the 64GBps that GPUs currently get is overkill. If 2 lanes of PCIE 7.0 is enough for current gen GPUs then just giving everything 4 lanes would be a good idea. I get that backwards compatibility is desirable but it is to the point where it is holding us back now. We need some breaking changes.
1. Update the PCIE slot like I stated.
2. Drop the ATX PSU form factor. You can get 1100W out of a SFX PSU now. That should just be the new standard going forward. Also should be ATX12VO which should help even more as the PSU wouldn't need 3.3v and 5v Circuitry. Could probably get 1300-1500W out of the same form factor. Could even go with Flex ATX but I think most people won't like the small 40mm fan.
3. NVME/PCIE Only. NVME now has support for HDDs. SATA and SAS should be dropped entirely to simplify everything. The OS could then be designed with very fast low latency storage in mind as a requirement and we could remove any considerations for running on spinning rust besides as additional storage.
How do you improve latency on these drives? RAID 0?
RAID 0, still being managed still seems like a latency bottleneck.
@@paulct91 but now aren't you in some way opening the highways, meaning the time to read/write is lower?
RAID 0 may increase your throughput and maybe max IOPS, but the minimum latency won't be better. You still need the same time to go from CPU over the PCIe bus to the drive (or multiple drives) and back. You can only make latency slightly worse, since you added some small overhead for the RAID.
Click click,,, no need my boyz, get primocache sit back and laugh
No PCIE 6.0 yet? Hopefully they stick to x4 lanes with it
Love the Fractal North White, built my latest PC in it. Got a 7950X3D, Aorus Master B650E, Aorus Master 4090, 64gb 6000mt/s CL30 Gskill Trident Z5 Neo RGB, 3 1TB 980 SSD's , 1 4TB Teamforce something SSD lol (steam libary), EVGA 1300 Supernova G something (RIP EVGA), Artic Liquid Freezer II 420 in Push/Pull (had to mod the case and do alittle pushing lol) 2 140 intakes on the bottom, 1 120 exhaust at the rear and the AIO fans pulling air in and exhausting through the case. All temps are awesome, nothing over 40 at idel or 60ish while heavy gaming, the 7950x3d will hit 80c on synthetics still though unless I massively undervolt it (in which case it's stabilizes at 74 to 78c) but then my system isn't stable at idle, and only at idle/watching youtube, will pass everything then crash 20 mins into a Linus Wan Show Rant lol.
"This isn't really a problem for AM5" Of course its not. We are the chosen ones. We are homo superior. WHERE MY X3D BOYS AT!!! LETTTSSS GOOOOOOOOOOOOO!!!
I have just got my hands on 2 Crucial T705 SSD's for my Ryzen 7000 build (RAID 0 baby). in order to keep them cool i got 2 of Corsair's M.2 Waterblocks. I tied 2 block into the CPU Loop (as one M.2 is just below the CPU) and then the 2nd block got tied into the GPU Loop (yes i run dual loops, sue me :) )
So far even under extreme benchmark loads, the drives get warm but not excessive, i'd hate to see how hot they get when air cooled
Why tho
Why not
@@camjohnson2004 can’t argue with that, carry on
Keep your gen5 garbo.
3.5" SSDs now please
no point in that cuz windows will never reach those speeds
I don't get who cares about PCIe gen 5 except... file copying enthusiasts?? It's driven the price of motherboards up for zero consumer benefit.
Is garbage arranged alphabetically
14um is a terrible name -- it makes it sound like the flash cell node is 14 micrometer, which is hilariously outdated.
not gonna lie, I hate SSDs. These speeds are useless(anything above 1GB/s) because a lot of the time, I'm not moving 1TB around so who cares? Even if I were to move an entire 100GB game from one SSD to another, at 1GB/s, it'd just take 100s(1min40 seconds), which is no biggie.
Companies keep pushing faster SSDs, leading to absolute trash that lasts about a year and needs A COOLER. Can we please start getting just, slightly slower SSDs, but that focus on lasting longer and not needing cooler with a fan?
Yeah the numbers game is really killing it. I like SSD speeds but not when they come at the cost of other things like now needing a fan header for my storage drive. As I said in my comment I would rather they cut the lanes available down to 2 or even a single PCIE lane and let us connect more devices with those lanes given we are limited to around 24 pcie lanes in total on consumer platforms.
The main reason they do is because of businness segment pushes for higher and higher speeds with a 3 year lifespan.
And the manufacturers turn around and try to sell the same stuff (or cut down designs) to consumers.
You already have an option that I think is exactly what you want: Optane. You can get a 1.5TB 905p with a U.2 PCIe controller for about $430. Very low latency (as Wendell said) and while max speed on that drive is "only" 2600MB/sec, you'll get that no matter how full your drive is, or how much you use it!
realtime video capture applications: extreme frame rates (for slow motion, 10k~1.75m fps), or fidelity (4K60 lossless 24b/32b, 1.5/2GB/s)
but it looks like Phantom cameras have moved on to just having up to 512GB of RAM to achieve those framerates.
@@RNGwhydoihavetoregis wouldn't those benefit more from latency? since they're constant writing to a file. I imagine they dump most of the data to RAM first then slowly to the storage anyway
random 4k read = trash