Honestly what I'd prefer is to see lots of devices using half the number of lanes while achieving the same bandwidth as today. You'd immediately get twice as many lanes to work with. Pcie 4.0 x8 would be the same as x16 on pcie 3.0. You'd now have another 8 lanes of pcie 4.0 to play with. You'd only need 2 lanes to get nvme speeds that require 4 lanes today. Only super fast nvme drives are pushing the limits on pcie 3.0 x4 anyway. The i/o capabilities would be nuts for a mainstream desktop if they could implement it as such. I realize though that all said peripherals would need to be pcie 4.0 or they won't get that double bandwidth per lane. That's a while away.
Limiting a GPU to 8x is easy - just plug in a second device, and the mainboard will split it. Some even support 8/4/4 splits (at least Intel boards, don't remember seeing an AM4 with that, but thats not a AM4 limitation, so board makers could make it), so you can put a duo NVMe SSD card in one slot, and whatever fancies you in another.
AMD's X570 chipset effectively gives you that. It has 4 PCIe 4.0 lanes to the CPU, and 16 PCIe 4.0 lanes for devices. You can't run them all at full speed, but you can run two SSDs at PCIe 3.0 x 4 speed.
The links to the devices hanging off of the X570 chipset are 4.0 links though, so you could put a 4.0 SSD on it and get max bandwidth out of it if you weren't using bandwidth on any other devices... I imagine DMA between devices on the x570 chipset also doesnt' suffer since it shouldn't have to go through the slow 4x link thought that doesn't have much application outside of perhaps RAID and server applicaitons maybe?
That shouldn't be done at a device level. If the device can utilize 4 PCIe 4.0 lanes, it will work fine in a number of configurations including two lanes and PCIe 3.0. Why gimp the actual device itself? If you want a device that's slower, there will always be low-end options, but lane reduction should be handled at the board level. Which as the others have already said, board manufacturers already configure splits for users with more devices. That's a better way of handling it.
Oh and what's "nuts" and "super fast" today is mainstream in a year or so. :D
What makes you think that the lower cost M.2 SSDs will not based on two PCIe 4.0 lanes, just like they are based today on two PCIe 3.0 lanes? The transition to PCIe 4.0 allows both to double the bandwidth of the low cost M.2 SSDs and to double the bandwidth of the higher cost M.2 SSDs (or rather the bandwidth of the SSDs *can* double, since for the first 3 years at least they will be nowhere close to saturating the bandwidth of their x2 or x4 PCIe link). In plenty of cases half the lanes will be used for the same bandwidth, not just with M.2 SSDs but with x8 and x16 PCIe slots and devices. The transition to PCIe 4.0 (and eventually to PCIe 5.0) allows that flexibility.
If this site makes you so unhappy, why are you here and commenting on articles?
There are plenty of other sites (ex: krebs) that focus on security vulnerabilities if that's a topic that interests you.
And FWIW, PCIE 4.0 is not an AMD exclusive, it's a standard that's been around for a while. The latest rumors I've seen suggest that Intel may skip over PCIE 4.0 and go straight to 5.0.
I find it weird that you bring up pcie 4.0 standard to a person who wasnt claiming otherwise. Getting back on point. IT IS ALSO WEIRD that anandtech didnt post a single thing about a whole new collection of serious flaws. And also, intel isnt skipping pcie 4.0.Its coming next year in various products. https://bit.ly/2JEryKJ
PCIE 4.0 will NOT be short lived. 5.0 likely won't be on boards for a few years at least. You forget that product development takes time. This SSD and others like it were likely under development for over a year, maybe over 2 years. Manufacturers are ramping up PCIE 4.0 products as we speak. I expect PCIE 5.0 to come in 2022 or 2023 at the earliest.
I brought up the standard because the OP was making it sound like PCIE 4.0 was an AMD advantage. And why haven't they reported on the new flaws that came to light this week? Maybe because their small staff is busying getting themselves to Computex, which is going to be much more interesting.
Well, it is an AMD advantage - they will have it on their mainstream desktop platform this year and Intel not. How big of an advantage this is is another question, but being able to buy a system that supports new standards now is a plus.
Well ,its is an amd advantage isnt it ? The fact that intel will have pcie 4/5 sometime in the future doesnt really have anything to do with it for now. As for second part, they are constantly adding a lot less important things (like the news we are commenting below).So. Thats not really the case that they are SO BUSY.
It sounds like Intel is pushing PCI 4x to older 14nm product while PCI 5 us reserved for newer 10nm products - but there is two things I find interesting about this - 1, Link is only for server products and usually those come later than others 2. Intel 10nm process has been supposedly moved up. My case IceLake will support 5 and not 4
Yes I seen that rumor, but it will be interesting how this folds out and curious why Intel might skip 4.0 and go directly to 5.0. What it sounds like AMD is rushing 4.0 and Intel needs something in 5.0 to fit there up and coming technology needs.
From what I can tell PCI 5.0 is designed to improved IO bandwidth significantly likely meaning 8 lane SSD will have bottleneck issues with other versions. It also suppose to do with lower power Who knows this might mean that 4 lane PCI 5.0 graphics card can do what 16 lane card does today. Intel probably wants it for IO bandwidth but following update on spec indicates also that it does this at lower power
Spec indicates it will be finalize this year 2019 - so that could mean in products in Intel future timeline, AMD will likely be following with PCI 5.0 update in the future.
it's going to be a long time before PCIE 5.0 devices are on the market (devices meaning motherboards and add-in cards). PCIE 4.0 was ratified on June 8, 2017. We are just now beginning to see a trickle of products 2 years later. It takes time to develop new products. PCIE 5.0 was just ratified this year. That means that 2022 or 2023 we'll see CPUs and products take advantage of it. We MIGHT see something in 2021, but not before then.
Yes, it's really interesting - both Fudzilla and Anandtech did not mention Zombieload even once - afaik not even in the tweets. This is a bit strange since this was important tech news and pretty much every other site reported on it.
Even the fact that Google disabled HT for Intel based Chrome books was not mentioned here, or how Apple said that in order to be fully protected you will need to disable HT in addition to the patches.
To be fair the May update implements Retpoline and fast import, which will help users regain much of their lost performance. How much will vary, and I'd love to see some tests. With that being said I'm really looking forward to seeing Zen 2 chips hit the market. The only really interesting news coming out of Intel recently is rumors about their discrete graphics.
The consumer market isn't ready for that yet. Prices still need to come down a lot. Going to 4TB on a M.2 SSD is possible with current chips, but it generally requires moving from DDR3 to DDR4 DRAM and from 256Gb to 512Gb NAND in order to fit everything on the card. That means a 4TB drive is pretty different from lower capacity versions, and isn't something that can easily or cheaply be added to existing product lines.
They're waiting for sales to drop on the other drives and then, as we know, they'll drop the new items. It nothing to with the market being ready... it's to do with making the most £££/$$$.
What's noteworthy, yet is rarely if ever mentioned, is that these numbers are at QD32. Meaning, any normal/typical user won't ever see even half this type of performance. I wish the focus went to QD1 numbers as a priority, with QD>1 provided only as a secondary reference, or an afterthought. Maybe in a format something like this:
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
42 Comments
Back to Article
Ej24 - Friday, May 24, 2019 - link
Honestly what I'd prefer is to see lots of devices using half the number of lanes while achieving the same bandwidth as today. You'd immediately get twice as many lanes to work with. Pcie 4.0 x8 would be the same as x16 on pcie 3.0. You'd now have another 8 lanes of pcie 4.0 to play with. You'd only need 2 lanes to get nvme speeds that require 4 lanes today. Only super fast nvme drives are pushing the limits on pcie 3.0 x4 anyway. The i/o capabilities would be nuts for a mainstream desktop if they could implement it as such. I realize though that all said peripherals would need to be pcie 4.0 or they won't get that double bandwidth per lane. That's a while away.nevcairiel - Saturday, May 25, 2019 - link
Limiting a GPU to 8x is easy - just plug in a second device, and the mainboard will split it. Some even support 8/4/4 splits (at least Intel boards, don't remember seeing an AM4 with that, but thats not a AM4 limitation, so board makers could make it), so you can put a duo NVMe SSD card in one slot, and whatever fancies you in another.All that needs is PCIe4 everything, as you say.
Chaitanya - Sunday, May 26, 2019 - link
X470 based Asus Crosshair is the only Am4 mainboard that allows for 8/4/4 pci-e splitting for GPU and SSD Raid.PixyMisa - Saturday, May 25, 2019 - link
AMD's X570 chipset effectively gives you that. It has 4 PCIe 4.0 lanes to the CPU, and 16 PCIe 4.0 lanes for devices. You can't run them all at full speed, but you can run two SSDs at PCIe 3.0 x 4 speed.cb88 - Saturday, August 10, 2019 - link
The links to the devices hanging off of the X570 chipset are 4.0 links though, so you could put a 4.0 SSD on it and get max bandwidth out of it if you weren't using bandwidth on any other devices... I imagine DMA between devices on the x570 chipset also doesnt' suffer since it shouldn't have to go through the slow 4x link thought that doesn't have much application outside of perhaps RAID and server applicaitons maybe?Alexvrb - Saturday, May 25, 2019 - link
That shouldn't be done at a device level. If the device can utilize 4 PCIe 4.0 lanes, it will work fine in a number of configurations including two lanes and PCIe 3.0. Why gimp the actual device itself? If you want a device that's slower, there will always be low-end options, but lane reduction should be handled at the board level. Which as the others have already said, board manufacturers already configure splits for users with more devices. That's a better way of handling it.Oh and what's "nuts" and "super fast" today is mainstream in a year or so. :D
Santoval - Sunday, May 26, 2019 - link
What makes you think that the lower cost M.2 SSDs will not based on two PCIe 4.0 lanes, just like they are based today on two PCIe 3.0 lanes? The transition to PCIe 4.0 allows both to double the bandwidth of the low cost M.2 SSDs and to double the bandwidth of the higher cost M.2 SSDs (or rather the bandwidth of the SSDs *can* double, since for the first 3 years at least they will be nowhere close to saturating the bandwidth of their x2 or x4 PCIe link).In plenty of cases half the lanes will be used for the same bandwidth, not just with M.2 SSDs but with x8 and x16 PCIe slots and devices. The transition to PCIe 4.0 (and eventually to PCIe 5.0) allows that flexibility.
sorten - Friday, May 24, 2019 - link
Can't wait until Sunday when I can read about the hardware that going in my next PC build.Freeb!rd - Saturday, May 25, 2019 - link
D!tto.Slash3 - Friday, May 24, 2019 - link
Link to press release: https://www.gigabyte.com/Press/News/1690BigMamaInHouse - Saturday, May 25, 2019 - link
AMD getting ultra-fast nvme PCIeGen4 while Intel loosing SSD storage performance after MDS Patches .supdawgwtfd - Saturday, May 25, 2019 - link
MDS patches? What are you talking about???Haven't read anything about them here at Anandtech?
shabby - Saturday, May 25, 2019 - link
AnandTech doesn't like badmouthing its favourite sponsor, it didn't even post the latest vulnerabilities from intel... I'm shocked.supdawgwtfd - Saturday, May 25, 2019 - link
What's funny is they time after time claim to be unbiased...But they don't report these important security issues???
Sure... No bias here. No favouritism. 100% balanced journalism here!
sorten - Saturday, May 25, 2019 - link
If this site makes you so unhappy, why are you here and commenting on articles?There are plenty of other sites (ex: krebs) that focus on security vulnerabilities if that's a topic that interests you.
And FWIW, PCIE 4.0 is not an AMD exclusive, it's a standard that's been around for a while. The latest rumors I've seen suggest that Intel may skip over PCIE 4.0 and go straight to 5.0.
RaV[666] - Saturday, May 25, 2019 - link
I find it weird that you bring up pcie 4.0 standard to a person who wasnt claiming otherwise.Getting back on point.
IT IS ALSO WEIRD that anandtech didnt post a single thing about a whole new collection of serious flaws.
And also, intel isnt skipping pcie 4.0.Its coming next year in various products.
https://bit.ly/2JEryKJ
HStewart - Saturday, May 25, 2019 - link
One more thing Tom's Hardware indicated that PCI 5.0 is expected this year and 4.0 will be short lived.https://www.tomshardware.com/news/pcie-4.0-5.0-pci...
eek2121 - Saturday, May 25, 2019 - link
PCIE 4.0 will NOT be short lived. 5.0 likely won't be on boards for a few years at least. You forget that product development takes time. This SSD and others like it were likely under development for over a year, maybe over 2 years. Manufacturers are ramping up PCIE 4.0 products as we speak. I expect PCIE 5.0 to come in 2022 or 2023 at the earliest.amnesia0287 - Sunday, May 26, 2019 - link
Pcie 5.0 is listed on multiple roadmaps for 2020/2021RaV[666] - Saturday, May 25, 2019 - link
I dont know where you read that there. I dont see it.Also, intels recent roadmap puts ice lake SP with PCIE 4 next year.Yes.FOUR.
sorten - Saturday, May 25, 2019 - link
I brought up the standard because the OP was making it sound like PCIE 4.0 was an AMD advantage. And why haven't they reported on the new flaws that came to light this week? Maybe because their small staff is busying getting themselves to Computex, which is going to be much more interesting.Irata - Saturday, May 25, 2019 - link
Well, it is an AMD advantage - they will have it on their mainstream desktop platform this year and Intel not. How big of an advantage this is is another question, but being able to buy a system that supports new standards now is a plus.RaV[666] - Saturday, May 25, 2019 - link
Well ,its is an amd advantage isnt it ? The fact that intel will have pcie 4/5 sometime in the future doesnt really have anything to do with it for now.As for second part, they are constantly adding a lot less important things (like the news we are commenting below).So. Thats not really the case that they are SO BUSY.
HStewart - Saturday, May 25, 2019 - link
It sounds like Intel is pushing PCI 4x to older 14nm product while PCI 5 us reserved for newer 10nm products - but there is two things I find interesting about this - 1, Link is only for server products and usually those come later than others 2. Intel 10nm process has been supposedly moved up. My case IceLake will support 5 and not 4Korguz - Saturday, May 25, 2019 - link
" Intel 10nm process has been supposedly moved up " moved up from what ?? its already 4 YEARS late.Opencg - Saturday, May 25, 2019 - link
Somebody call the whambulance! Another side channel attack that requires the attacker already compromised your computer.HStewart - Saturday, May 25, 2019 - link
Yes I seen that rumor, but it will be interesting how this folds out and curious why Intel might skip 4.0 and go directly to 5.0. What it sounds like AMD is rushing 4.0 and Intel needs something in 5.0 to fit there up and coming technology needs.From what I can tell PCI 5.0 is designed to improved IO bandwidth significantly likely meaning 8 lane SSD will have bottleneck issues with other versions. It also suppose to do with lower power Who knows this might mean that 4 lane PCI 5.0 graphics card can do what 16 lane card does today. Intel probably wants it for IO bandwidth but following update on spec indicates also that it does this at lower power
https://pcisig.com/pci-express®-50-specificat...
HStewart - Saturday, May 25, 2019 - link
Spec indicates it will be finalize this year 2019 - so that could mean in products in Intel future timeline, AMD will likely be following with PCI 5.0 update in the future.eek2121 - Saturday, May 25, 2019 - link
Intel will adopt PCIE 4.0 first. PCIE 5.0 was JUST ratified.shabby - Saturday, May 25, 2019 - link
Amd rushing pcie4? 🤣eek2121 - Saturday, May 25, 2019 - link
it's going to be a long time before PCIE 5.0 devices are on the market (devices meaning motherboards and add-in cards). PCIE 4.0 was ratified on June 8, 2017. We are just now beginning to see a trickle of products 2 years later. It takes time to develop new products. PCIE 5.0 was just ratified this year. That means that 2022 or 2023 we'll see CPUs and products take advantage of it. We MIGHT see something in 2021, but not before then.Irata - Saturday, May 25, 2019 - link
Yes, it's really interesting - both Fudzilla and Anandtech did not mention Zombieload even once - afaik not even in the tweets. This is a bit strange since this was important tech news and pretty much every other site reported on it.Even the fact that Google disabled HT for Intel based Chrome books was not mentioned here, or how Apple said that in order to be fully protected you will need to disable HT in addition to the patches.
Phynaz - Sunday, May 26, 2019 - link
AMD is irrelevant. Get used to itAlexvrb - Saturday, May 25, 2019 - link
To be fair the May update implements Retpoline and fast import, which will help users regain much of their lost performance. How much will vary, and I'd love to see some tests. With that being said I'm really looking forward to seeing Zen 2 chips hit the market. The only really interesting news coming out of Intel recently is rumors about their discrete graphics.xstylus - Saturday, May 25, 2019 - link
Great. Wonderful. Lovely. When can I have 4TB in a M.2-2280 form factor, please?Billy Tallis - Saturday, May 25, 2019 - link
The consumer market isn't ready for that yet. Prices still need to come down a lot. Going to 4TB on a M.2 SSD is possible with current chips, but it generally requires moving from DDR3 to DDR4 DRAM and from 256Gb to 512Gb NAND in order to fit everything on the card. That means a 4TB drive is pretty different from lower capacity versions, and isn't something that can easily or cheaply be added to existing product lines.damianrobertjones - Monday, May 27, 2019 - link
They're waiting for sales to drop on the other drives and then, as we know, they'll drop the new items. It nothing to with the market being ready... it's to do with making the most £££/$$$.boeush - Saturday, May 25, 2019 - link
What's noteworthy, yet is rarely if ever mentioned, is that these numbers are at QD32. Meaning, any normal/typical user won't ever see even half this type of performance. I wish the focus went to QD1 numbers as a priority, with QD>1 provided only as a secondary reference, or an afterthought. Maybe in a format something like this:"read/write speeds 0.065(<5)/0.265(<5) GB/s"
cmdrdredd - Saturday, May 25, 2019 - link
Yeah I don't see many doing anything to get close to the potential.jabber - Sunday, May 26, 2019 - link
To be honest NVMe x4 was a big let down. I'd settle for cheap reliable drives that just pushed 1000MBps R/W with lower latency.blingon - Monday, May 27, 2019 - link
M.2 seems like a bad form factor for very fast SSDs. Sad that U.2 doesn't seem to have become more widespread.LoneWolf15 - Thursday, May 30, 2019 - link
If TLC, interested.If QLC, not in the least bit interested.