AMD 400-series Chipset Surfaces on PCI-SIG, PCIe 3.0 General Purpose Confirmed

[btarunr]

AMD's second-generation Ryzen processors, which debut some time in Q1-2018, will be accompanied by the company's new 400-series motherboard chipset, even though they are expected to work with existing socket AM4 motherboards based on 300-series chipsets (with BIOS updates). The 400-series Promontory chipset surfaced on the PCIe Integrators List of PCI-SIG, the standards governing body of the PCI bus (which also oversees PCIe specifications development).

The listing seems to confirm that 400-series chipset will feature PCI-Express gen 3.0 general purpose lanes. These are downstream PCIe lanes put out by the chipset, to run the various external onboard controllers on the motherboard, and usually wired to the x1 and x4 PCIe slots. The current 300-series chipset only features up to 8 PCIe gen 2.0 general purpose lanes, and that was seen as a drawback. AMD Ryzen socket AM4 processors put out additional gen 3.0 lanes besides the 16 lanes allocated to PEG (one x16 or two x8, physically x16 slots); and 4 lanes serving as chipset bus. These additional gen 3.0 lanes typically drive a 32 Gb/s M.2 slot. With 400-series chipset bringing gen 3.0 general purpose lanes, one can expect newer socket AM4 motherboards with more than one 32 Gb/s M.2 slot (one from the SoC, another from the chipset).



View at TechPowerUp Main Site

 

[iO]

The same chip as the 300 series but at least they got the PCIe 3.0 certification this time.

 

[bonehead123]

So let me get this straight...

They will release a brand "new" chipset in 2018 with a few minor spec bumps, but NOT pcie 4 config, which I was recently approved for implementation by the pcie board...what exactly is the point?

Looks like the same ole same same cash cow trick... intro a new, but crippled, chipset in Q1, get the $$ from it, then milk it again Q3 by updating and rereleasing it with the new spec...

whodathunkit

 

[Mistral]

So let me get this straight...

They will release a brand "new" chipset in 2018 with a few minor spec bumps, but NOT pcie 4 config, which I was recently approved for implementation by the pcie board...what exactly is the point?

Looks like the same ole same same cash cow trick... intro a new, but crippled, chipset in Q1, get the $$ from it, then milk it again Q3 by updating and rereleasing it with the new spec...

whodathunkit

Yeah, they should get the all latest tech right in from the first time and never release a new product refresh...
Seriously, you're trying to blow this out of proportion. Whoever really wanted to go AMD and seriously needed the extra pcie went Threadripper. This is not Intel where every new CPU requires a new chipset, it's just a refresh to give motherboard manufacturers and new users something to play with.

 

[iO]

So let me get this straight...

They will release a brand "new" chipset in 2018 with a few minor spec bumps, but NOT pcie 4 config, which I was recently approved for implementation by the pcie board...what exactly is the point?

Looks like the same ole same same cash cow trick... intro a new, but crippled, chipset in Q1, get the $$ from it, then milk it again Q3 by updating and rereleasing it with the new spec...

whodathunkit

Yea, a PCIe 4.0 chipset would make soo much sense when there are zero PCIe 4 devices on the market and the CPU only offers 4 PCIe 3 lanes to cennect it to......

 

[TheLostSwede]

So let me get this straight...

They will release a brand "new" chipset in 2018 with a few minor spec bumps, but NOT pcie 4 config, which I was recently approved for implementation by the pcie board...what exactly is the point?

Looks like the same ole same same cash cow trick... intro a new, but crippled, chipset in Q1, get the $$ from it, then milk it again Q3 by updating and rereleasing it with the new spec...

whodathunkit

PCIe 4.0 was never intended for consumer devices, it's meant for servers and datacenter applications, or more specifically for storage, networking and apparently AI applications. The next general purpose PCIe release is 5.0 which is supposed to arrive in Q2 2019, so if you want a new PCIe spec on your motherboard, you're going to have to wait until 2020. This does by the way, apply to Intel as well, although we might get some HEDT platform from Intel with support for it, since the chipsets and CPUs are shared with the Xeon platform from a design perspective, unless Intel downgrades them to PCIe 3.0 that is...

 

[Aquinus]

They will release a brand "new" chipset in 2018 with a few minor spec bumps, but NOT pcie 4 config, which I was recently approved for implementation by the pcie board...what exactly is the point?

Why? There really isn't a huge push to drive more with less right now and there are no devices that use it. The thing with PCIe 3.0 is that it's already a lot of bandwidth per lane and it scales horizontally to 16 lanes for the consumer and in some cases 32 lanes for certain kinds of servers. For consumers though, if you need more bandwidth, you can get a platform with more lanes and that's part of the reason why they exist. NVMe is already fast with 4 PCIe lanes and it's not like GPUs are struggling using 8 when we talk about PCIe 3.0 so, I'm not really sure what the advantage would be given the cost.

Being certified means that all of the circuit characteristics are to spec as opposed to "good enough to operate." If anything this would be a step towards PCIe 4 but, there really is no reason to go full tilt towards it.

 

[cucker tarlson]

Yea, a PCIe 4.0 chipset would make soo much sense when there are zero PCIe 4 devices on the market and the CPU only offers 4 PCIe 3 lanes to cennect it to......

It's not about pci-e 4.0 GPUs but more about twice the bandwidth of those 16 lanes off the CPU so you can run SLI setups on mainstream mobos and multiple nvme drives that won't be bottlenecked by 4 PCH 3.0 lanes. Good luck ever pushing mainstream platforms to catch up with HEDT with that snarky attitude. You want two GPUs and three nvmes ? Forget about running them off a $300 CPU and a $120 mobo despite the CPU like Ryzen 1700 actually being almost as fast as TR 1900X. Pay twice as much for the same 8c/16t 4GHz CPU but on a twice as expensive X399 mobo.

 

[R0H1T]

It's not about pci-e 4.0 GPUs but more about twice the bandwidth of those 16 lanes off the CPU so you can run SLI setups on mainstream mobos and multiple nvme drives that won't be bottlenecked by 4 PCH 3.0 lanes.

Except Ryzen isn't held back by the PCH, until the (any) CPU itself gets PCIe 4.0 you'd be hard pressed seeing a chipset that supports it.

 

[cucker tarlson]

Except Ryzen isn't held back by the PCH, until the (any) CPU itself gets PCIe 4.0 you'd be hard pressed seeing a chipset that supports it.

Really ? Then why are most X370 mobos using only one pci-e 3.0 x4 m.2 nvme and the other one is only pci-e 3.0 x2 ?

 

[newtekie1]

Wait, the chipset is only going to provide 4 PCI-E 3.0 lanes? Seriously, why even bother? AMD seriously needs to step up their chipset game. I know they are trying to keep costs down, but this is just ridiculous. The high end chipset should provide at least another 16 PCI-E 3.0 lanes for the devices on the motherboard to use. Hell, AMD would be better off just throwing their chipset in the garbage and buying an off the shelf PCI-E Switch chip. There are PLX chips on the market that will take the 8 extra lanes from the CPU, and output usable PCI-E 3.0 lanes. You could have another x8(x16 Physical) slot, 3 x4 3.0 NVMe ports, and still have a ton of lanes left over for extra NICs and Wireless controllers, the sound chip, etc. AMD's chipset is its weakness right now, and the new 400 series doesn't look much better.

 

[Aquinus]

Wait, the chipset is only going to provide 4 PCI-E 3.0 lanes? Seriously, why even bother? AMD seriously needs to step up their chipset game. I know they are trying to keep costs down, but this is just ridiculous. The high end chipset should provide at least another 16 PCI-E 3.0 lanes for the devices on the motherboard to use. Hell, AMD would be better off just throwing their chipset in the garbage and buying an off the shelf PCI-E Switch chip. There are PLX chips on the market that will take the 8 extra lanes from the CPU, and output usable PCI-E 3.0 lanes. You could have another x8(x16 Physical) slot, 3 x4 3.0 NVMe ports, and still have a ton of lanes left over for extra NICs and Wireless controllers, the sound chip, etc. AMD's chipset is its weakness right now, and the new 400 series doesn't look much better.

They really should just stick with the minimal SoC on the CPU and let motherboard manufacturers decide to add additional chipsets or PCIe switches. Developing chipsets is probably not very profitable for AMD and isn't really worth the effort however AMD should be sensitive to shifting this burden to motherboard manufacturers.

 

[Aldain]

Wait, the chipset is only going to provide 4 PCI-E 3.0 lanes? Seriously, why even bother? AMD seriously needs to step up their chipset game. I know they are trying to keep costs down, but this is just ridiculous. The high end chipset should provide at least another 16 PCI-E 3.0 lanes for the devices on the motherboard to use. Hell, AMD would be better off just throwing their chipset in the garbage and buying an off the shelf PCI-E Switch chip. There are PLX chips on the market that will take the 8 extra lanes from the CPU, and output usable PCI-E 3.0 lanes. You could have another x8(x16 Physical) slot, 3 x4 3.0 NVMe ports, and still have a ton of lanes left over for extra NICs and Wireless controllers, the sound chip, etc. AMD's chipset is its weakness right now, and the new 400 series doesn't look much better.

This has gt to be the dumbest comment i have read on this site..period

It is like x399 never happened

 

[cucker tarlson]

They really should just stick with the minimal SoC on the CPU and let motherboard manufacturers decide to add additional chipsets or PCIe switches charge us even more for those features to be included only in the top tier enthusiast boards instead of getting them as the basic ones with mid-range mobos,which get more and more expensive year after year even though we're getting the same pci-e configuration as mobos had 2 years ago. Developing chipsets is probably not very profitable for AMD and isn't really worth the effort however AMD should be sensitive to shifting this burden to motherboard manufacturers.

FTFY

 

[Aquinus]

FTFY

Yeah, since it doesn't cost AMD any money to produce something that does the same thing themselves. Hitting the same bar as what a PCH-like device would provide isn't a high bar. If you're expecting premium features and every little feature, I would expect that you would have to pay for it. If you think entry level hardware and high-end hardware should be similarly priced, then you're just not being realistic because more features costs more money to produce and if you have a setup where it's not required, you're allowing the cheap to get even cheaper and the most expensive to get as expensive as the motherboard manufacturer wants to sell it for. Just remember, if they don't sell it, they don't make money. They can charge whatever they want and if it's earning them profit, they don't care but, the most elaborate boards aren't the ones that have the highest demand.

I'm trying to provoke a fight.

I fixed that for you...

 

[cucker tarlson]

Yeah, since it doesn't cost AMD any money to produce something that does the same thing themselves. Hitting the same bar as what a PCH-like device would provide isn't a high bar. If you're expecting premium features and every little feature, I would expect that you would have to pay for it. If you think entry level hardware and high-end hardware should be similarly priced, then you're just not being realistic because more features costs more money to produce and if you have a setup where it's not required, you're allowing the cheap to get even cheaper and the most expensive to get as expensive as the motherboard manufacturer wants to sell it for. Just remember, if they don't sell it, they don't make money. They can charge whatever they want and if it's earning them profit, they don't care but, the most elaborate boards aren't the ones that have the highest demand.

I fixed that for you...

So now more than a measly 4 pci-e pch lanes is a premium.....
Good grief, boards are already expensive enough without that.

 

[newtekie1]

They really should just stick with the minimal SoC on the CPU and let motherboard manufacturers decide to add additional chipsets or PCIe switches. Developing chipsets is probably not very profitable for AMD and isn't really worth the effort however AMD should be sensitive to shifting this burden to motherboard manufacturers.

They would still need the chipset to provide things like extra SATA ports and USB ports. Offloading those to 3rd party chips is probably not something motherboard manufacturers would be happy with doing on high end boards. They should just farm the design of the PCI-E switch part of their chipset out to someone like PLX, who knows what they are doing.

This has gt to be the dumbest comment i have read on this site..period

It is like x399 never happened

Wow, the X399 chipset is even worse when put in the context of HEDT. It, just like X370, only provides 8 PCI-E 2.0 lanes, which is just laughable for a HEDT chipset. The only thing saving AMD on HEDT is all the lanes from the CPUs. In fact, I'd almost be willing to bet that X399 and X370 are the same piece of silicon, but X370 has a few parts disabled.

And the worst part about it is that Threadripper basically confirms that the AM4 platform should really have 28 usable PCI-E 3.0 lanes(32 - 4 Chipset Lanes), but AMD has locked 4 lanes off for some reasons.

 

[cucker tarlson]

They would still need the chipset to provide things like extra SATA ports and USB ports. Offloading those to 3rd party chips is probably not something motherboard manufacturers would be happy with doing on high end boards. They should just farm the design of the PCI-E switch part of their chipset out to someone like PLX, who knows what they are doing.



Wow, the X399 chipset is even worse when put in the context of HEDT. It, just like X370, only provides 8 PCI-E 2.0 lanes, which is just laughable for a HEDT chipset. The only thing saving AMD on HEDT is all the lanes from the CPUs. In fact, I'd almost be willing to bet that X399 and X370 are the same piece of silicon, but X370 has a few parts disabled.

And the worst part about it is that Threadripper basically confirms that the AM4 platform should really have 28 usable PCI-E 3.0 lanes(32 - 4 Chipset Lanes), but AMD has locked 4 lanes off for some reasons.

Those 64 lanes off the CPU are more than enough to compensate, though I''m not sure if you can have a bootable OS nvme drive running off CPU lanes instead of PCH lanes.

I'd take 64 PCI-E 3.0 lanes off the CPU with fewer lanes off the chipset over 28 CPU lanes like on 7820X with more lanes off the chipset.

 

[Aquinus]

They would still need the chipset to provide things like extra SATA ports and USB ports. Offloading those to 3rd party chips is probably not something motherboard manufacturers would be happy with doing on high end boards. They should just farm the design of the PCI-E switch part of their chipset out to someone like PLX, who knows what they are doing.

If they were going to do this, I would want at least another 4 lanes worth of bandwidth if it's going to switch 16 or more lanes because switching only 4 lanes worth of bandwidth alone will choke very easily and would only take a single NVMe device to saturate it. I would agree with you if and only if they doubled the bandwidth to the chipset.

Those 64 lanes off the CPU are more than enough to compensate, though I''m not sure if you can have a bootable OS nvme drive running off CPU lanes instead of PCH lanes.

Modern hardware will boot a single NVMe drive from whatever PCIe complex it's attached to. The question is will the drive be detected at boot time. Most modern hardware probably will. Even my 6 year old board, with a PCIe to M.2 adapter would detect it. The question would be about RAID, which I would argue is a non-issue because booting from NVMe RAID doesn't have a whole lot of benefits when you can mount a NVMe RAID device after the machine has started the kernel.

 

[cucker tarlson]

Do you have the pci-e to m.2 adapter plugged into a pci-e slot that runs off the chipset or the cpu ? It's nice to able to utilize the extra cpu lanes on the 3930K for a pci-e 3.0 x4 drive, but frankly in 2018 we should be able to get that on an entry level chipset like b320 or b350, x370 should provide at least two of those, not to mention its successor.

 

[Aquinus]

Do you have the pci-e to m.2 adapter plugged into a pci-e slot that runs off the chipset or the cpu ?

Neither, I'm still using SATA RAID 0 for the time being but, even if I did, the PCH isn't an option because the only PCIe lanes the chipset is attached to is the 1x slots and they're shared with WiFi and eSATA on my board. If I could though, I wouldn't use the PCH because it's DMI 2.0 which would constrain me to the equivalent of PCIe 2.0 x4, which is fine for SATA RAID-0 but, is inadequate for a good Samsung NVMe drive, forget sharing it with other devices.

 

[notb]

"AMD's second-generation Ryzen processors, which debut some time in Q1-2018, will be accompanied by the company's new 400-series motherboard chipset, even though they are expected to work with existing socket AM4 motherboards based on 300-series chipsets (with BIOS updates)."

They are expected?

They will release a brand "new" chipset in 2018 with a few minor spec bumps, but NOT pcie 4 config, which I was recently approved for implementation by the pcie board...what exactly is the point?

The point is: to release a new chipset - possibly fixing some of 300-series issues.
PCIe 4.0 wouldn't be utilized right now. They're keeping this for future generations. We will most likely have at least one more AM4 gen - a good place to put a novelty like that.

Except Ryzen isn't held back by the PCH, until the (any) CPU itself gets PCIe 4.0 you'd be hard pressed seeing a chipset that supports it.

Why is that? The chipset (or an extension card) can support a newer PCIe revision. And this is exactly what we'll see in server gear.

 

[TheoneandonlyMrK]

"AMD's second-generation Ryzen processors, which debut some time in Q1-2018, will be accompanied by the company's new 400-series motherboard chipset, even though they are expected to work with existing socket AM4 motherboards based on 300-series chipsets (with BIOS updates)."

They are expected?


The point is: to release a new chipset - possibly fixing some of 300-series issues.
PCIe 4.0 wouldn't be utilized right now. They're keeping this for future generations. We will most likely have at least one more AM4 gen - a good place to put a novelty like that.

Why is that? The chipset (or an extension card) can support a newer PCIe revision. And this is exactly what we'll see in server gear.

Until a cpu puts out pciex4 there is no point in another chip splitting to pciex4 ,what would feed that chip enough system bandwidth for it's rated speed ,2x 16 lanes of pciex3? And for what , GPUs local cache's are getting large only next gen nvme or raided nvme might benefit in the home.

 

[newtekie1]

Those 64 lanes off the CPU are more than enough to compensate, though I''m not sure if you can have a bootable OS nvme drive running off CPU lanes instead of PCH lanes.

I'd take 64 PCI-E 3.0 lanes off the CPU with fewer lanes off the chipset over 28 CPU lanes like on 7820X with more lanes off the chipset.

Yeah, that's why I said those lanes are the only thing saving Threadripper. There are some issues with having the lanes off the CPU. NVMe drives are bootable, but NVMe RAID was tricky to get working. Supposedly AMD got it working, released the drivers, then pulled them for some reason, and I don't know if they ever re-released them. This is because the PCI-E lanes are split between two CPU dies. So Threadripper could still benefit from a better chipset.

If they were going to do this, I would want at least another 4 lanes worth of bandwidth if it's going to switch 16 or more lanes because switching only 4 lanes worth of bandwidth alone will choke very easily and would only take a single NVMe device to saturate it. I would agree with you if and only if they doubled the bandwidth to the chipset.

I don't believe it to be necessary, remember Intel uses the same x4 link to their Z270/370, and those both provide 24 PCI-E 3.0 lanes.

But, if AMD was to make a chipset that was actually worth a damn, they could then dedicate all 8 extra lanes on the CPU to the connection to the chipset, because the chipset would have enough lanes to stand on its own, and you wouldn't need PCI-E lanes from the CPU to be used for an NVMe drive.

 

[Aquinus]

But, if AMD was to make a chipset that was actually worth a damn, they could then dedicate all 8 extra lanes on the CPU to the connection to the chipset, because the chipset would have enough lanes to stand on its own, and you wouldn't need PCI-E lanes from the CPU to be used for an NVMe drive.

Absolutely, so long as the added latency is tolerable which, I think it would in a lot of cases. However, in W1zz's PCIe review, the 4 lanes off the PCH are actually slower than 4 lanes off the CPU and I suspect that's more of a latency problem than a bandwidth problem but, there are certain classes of devices where that's acceptable (GPUs are probably not one of them.) Intel's boards can have as many general purpose serial ports as they want but, you're still constrained by DMI 3.0's bandwidth. So, if you're just using one NVMe drive and a couple other SATA drives, then it's probably not enough of a bottleneck to be crippling but, it will slow you down. If you try using two of those fast Samsung NVMe drives in RAID-0, it's going to be choked by the PCH along with anything else attached to it, which (likely) includes NICs and WiFi, SATA drives, Audio, and USB bandwidth like all of those USB 3.0 ports you now get off the PCH. It all adds up if you start using a lot of them and it only gets a lot worse if a lot are used at the same time. That's my main point. Sometimes that's acceptable and sometimes it's not. If NVMe RAID is your goal, it's not. If it's just being able to add more devices and you care more about flexibility, it is.

With that said, it's my general belief that chipsets should be capable of reasonably handling all of a boards I/O needs aside from PCIe. PCIe should be there but, it's a convenience thing, not a performance thing, given bandwidth and latency but, if people expect full performance out of NVMe RAID, then the PCH (in my opinion,) shouldn't be the option because anything that demands high performance and low latency should be wired directly to the CPU.