Monday, August 7th 2017

Intel "Coffee Lake" Platform Detailed - 24 PCIe Lanes from the Chipset

Intel seems to be addressing key platform limitations with its 8th generation Core "Coffee Lake" mainstream desktop platform. The first Core i7 and Core i5 "Coffee Lake" processors will launch later this year, alongside motherboards based on the Intel Z370 Express chipset. Leaked company slides detailing this chipset make an interesting revelation, that the chipset itself puts out 24 PCI-Express gen 3.0 lanes, that's not counting the 16 lanes the processor puts out for up to two PEG (PCI-Express Graphics) slots.

The PCI-Express lane budget of "Coffee Lake" platform is a huge step-up from the 8-12 general purpose lanes put out by previous-generation Intel chipsets, and will enable motherboard designers to cram their products with multiple M.2 and U.2 storage options, besides bandwidth-heavy onboard devices such as additional USB 3.1 and Thunderbolt controllers. The chipset itself integrates a multitude of bandwidth-hungry connectivity options. It integrates a 10-port USB 3.1 controller, from which six ports run at 10 Gbps, and four at 5 Gbps.
Other onboard controllers includes a SATA AHCI/RAID controller with six SATA 6 Gbps ports. The platform also introduces PCIe storage options (either an M.2 slot or a U.2 port), which is wired directly to the processor. This is drawing inspiration from AMD AM4 platform, in which an M.2/U.2 option is wired directly to the SoC, besides two SATA 6 Gbps ports. The chipset also integrates a WLAN interface with 802.11ac and Bluetooth 5.0, though we think only the controller logic is integrated, and not the PHY itself (which needs to be isolated for signal integrity).

Intel is also making the biggest change to onboard audio standards since the 15-year old Azalia (HD Audio) specification. The new Intel SmartSound Technology sees the integration of a "quad-core" DSP directly into the chipset, with a reduced-function CODEC sitting elsewhere on the motherboard, probably wired using I2S instead of PCIe (as in the case of Azalia). This could still very much be a software-accelerated technology, where the CPU does the heavy lifting with DA/AD conversion.

According to leaked roadmap slides, Intel will launch its first 8th generation Core "Coffee Lake" processors along with motherboards based on the Z370 chipset within Q3-2017. Mainstream and value variants of this chipset will launch only in 2018.
Sources: VideoCardz, PCEVA Forums
Add your own comment

119 Comments on Intel "Coffee Lake" Platform Detailed - 24 PCIe Lanes from the Chipset

#76
bug
boeI get that 3.0 is faster than 2.0 but since all my equipment is 3.0 I need more lanes and even if they made it 4.0 my 3.0 equipment wouldn't perform any faster with insufficient lanes.
Yes, it would. You'd split one 4.0 into several 3.0 lanes and you'd be good.
newtekie1That is the beauty if DMA, it allows devices to talk directly to each other with very minimal interaction with the CPU or System RAM.
I know, but I never figured out how these PCIe complexes are built. If, like you say, they use DMA to talk to each other "directly", that would be great. And I suspect that's what's going on, but I never confirmed it.
Posted on Reply
#77
nemesis.ie
There is some very interesting "lane management" going on with the AMD X470 platform, check out the videos from L1 tech using the two on-board M.2 slots - one "steals" lanes from the slots, but it is pretty zippy.

More lanes is nice to have though and Threadripper is really the only game in town at that level/price.
Posted on Reply
#78
newtekie1
Semi-Retired Folder
boeI get that 3.0 is faster than 2.0 but since all my equipment is 3.0 I need more lanes and even if they made it 4.0 my 3.0 equipment wouldn't perform any faster with insufficient lanes. It seems I still don't understand the situation though. Let's say I got a 8700k CPU. On intel's website - it says I have 16PCIe lanes - some web sites it says there are 28 and some say 40. Some people are talking about CPUs having more that are for the motherboard. I don't know how many more there are for the motherboard which go to the slots (if any) and how do I know they aren't used up for resources like USB ports, onboard sata and raid controllers, on board sound cards, on board wifi, on board nic ports. My guess is those alone might be using at least a dozen pcie lanes. So again unless some manufacturers tell us how many lanes are available fixed or otherwise for the slots it seems like a crap shoot at best.
That is why it is important that the chipset provide 24 lanes. The total Intel platform provides 40 lanes. 16 are attached directly to the CPU(or more specifically the northbridge inside the CPU die). The other 24 come from the PCH(southbridge) chip on the motherboard, this chip is then linked to the CPU with a PCI-E x4 link. But that x4 link only becomes an issue when transferring from a storage device to system memory(opening a program/game).

Most manufacturers make it pretty clear where the PCI-E lanes are coming from, or it is pretty easy to figure it out. The 16 lanes from the CPU are supposed to only be used for graphics. The first PCI-E x16 slot is almost always connected to the CPU. If there is a second slot PCI-E x16 slot, then almost always the first slot will become an x8 and the second will be an x8 as well, because they are sharing the 16 lanes from the CPU. The specs of the motherboard will tell you this. You'll see something in the specs like "single at x16 ; dual at x8 / x8" Some even say "single at x16 ; dual at x8 / x8 ; triple at x8 / x4 / x4". In that case, all 3 PCI-E x16 slots are actually connected to the CPU, but when multiple are used, they run at x8 or x4 speed.

Any other slot that doesn't share bandwidth like this, is pretty much guaranteed to be using the chipset lanes and not the ones directly connected to the CPU.
bugI know, but I never figured out how these PCIe complexes are built. If, like you say, they use DMA to talk to each other "directly", that would be great. And I suspect that's what's going on, but I never confirmed it.
I remember back in the days when the CPU had to handle data transfers, it was so slow. Does anyone else remember the days when burning a CD would max out your CPU, and if you tried to open anything else on the computer, the burn would fail? That was because DMA wasn't a thing(and buffer underrun protection wasn't a thing yet either).
nemesis.ieMore lanes is nice to have though and Threadripper is really the only game in town at that level/price.
Threadripper isn't a perfect solution either. In fact, it introduces a new set of issues. The fact that Threadripper is really a MCM, and the PCI-E lanes coming from the CPU are actually split up like they are on two different CPUs, leads to issues. If a device connected to one CPU wants to talk to another, it has to be done over the Infinity Fabric, which introduces latency. And it really isn't that much better than going over Intel's DMI link from the PCH to the CPU. It also had issues with RAID, due to the drives essentially being connected to two different storage controllers, but I think AMD has finally worked that one out.
Posted on Reply
#79
bug
newtekie1I remember back in the days when the CPU had to handle data transfers, it was so slow. Does anyone else remember the days when burning a CD would max out your CPU, and if you tried to open anything else on the computer, the burn would fail? That was because DMA wasn't a thing(and buffer underrun protection wasn't a thing yet either).
Yeah, UDMA was a big deal back in the day. However PCIe is different. It's a hierarchical structure. And idk whether leaves can talk to each other without going all the way to the root. The engineer in me says you don't come up with a tree-like structure unless you want the branches to be able to work on their own. But the same engineer needs a spec/paper that says that is indeed the case.
Posted on Reply
#80
nemesis.ie
@newtekie, I remember CPUs being maxed for many tasks that are considered "light" today. Including installing whopping 32MB hard drives into PCs back in 1988. ;)
Posted on Reply
#81
boe
newtekie1That is why it is important that the chipset provide 24 lanes. The total Intel platform provides 40 lanes. 16 are attached directly to the CPU(or more specifically the northbridge inside the CPU die). The other 24 come from the PCH(southbridge) chip on the motherboard, this chip is then linked to the CPU with a PCI-E x4 link. But that x4 link only becomes an issue when transferring from a storage device to system memory(opening a program/game).

Most manufacturers make it pretty clear where the PCI-E lanes are coming from, or it is pretty easy to figure it out. The 16 lanes from the CPU are supposed to only be used for graphics. The first PCI-E x16 slot is almost always connected to the CPU. If there is a second slot PCI-E x16 slot, then almost always the first slot will become an x8 and the second will be an x8 as well, because they are sharing the 16 lanes from the CPU. The specs of the motherboard will tell you this. You'll see something in the specs like "single at x16 ; dual at x8 / x8" Some even say "single at x16 ; dual at x8 / x8 ; triple at x8 / x4 / x4". In that case, all 3 PCI-E x16 slots are actually connected to the CPU, but when multiple are used, they run at x8 or x4 speed.

Any other slot that doesn't share bandwidth like this, is pretty much guaranteed to be using the chipset lanes and not the ones directly connected to the CPU.
So you say it is clear but it isn't clear to me - maybe I'm just obtuse or maybe my pants are too big. If I have a 16x video card, 8x raid card and 8x nic - all of which are PCIe 3.0 it sounds like my max speed for anything other than my 16x video card is 4x and I don't know if I have even 4x to spare for the other 2 cards as I don't know how many PCIe lanes are used by the onboard sound, raid, sata, wifi, usb ports, etc.

I appreciate your help on clarifying this as I'd really like to know. I can pick a processor and specific motherboard if it helps you to give me an answer that helps me go in the right direction. I use my PC for gaming, work and as a HTPC. I have 150TB of storage and often transfer it to my other PC which has 100TB of storage so I hammer my nic and raid controller about once per week -usually about 4TB of transfer with updates and changes.

Getting a clean PCIe answer has been as challenging for me as finding out when I can actually use my frequent flyer miles.
Posted on Reply
#82
nemesis.ie
You could look at a motherboard with built-in 5 or 10Gbe interface and put a similar NIC in the other PC as an add-in board? Then just connect them with a crossover cable. Use the original port for internet access.

Some of the Asrock boards have an Aquantia 5 or 10Gbe as well as a 1Gbe (or two) on them.

Once the NIC saturates it doesn't matter how fast the drives on either end are or what lanes they are connected to.
Posted on Reply
#83
cdawall
where the hell are my stars
boeSo you say it is clear but it isn't clear to me - maybe I'm just obtuse or maybe my pants are too big. If I have a 16x video card, 8x raid card and 8x nic - all of which are PCIe 3.0 it sounds like my max speed for anything other than my 16x video card is 4x and I don't know if I have even 4x to spare for the other 2 cards as I don't know how many PCIe lanes are used by the onboard sound, raid, sata, wifi, usb ports, etc.

I appreciate your help on clarifying this as I'd really like to know. I can pick a processor and specific motherboard if it helps you to give me an answer that helps me go in the right direction. I use my PC for gaming, work and as a HTPC. I have 150TB of storage and often transfer it to my other PC which has 100TB of storage so I hammer my nic and raid controller about once per week -usually about 4TB of transfer with updates and changes.

Getting a clean PCIe answer has been as challenging for me as finding out when I can actually use my frequent flyer miles.
For server grade tasks maybe you should look at server grade products. Both amd and Intel offer products that easily fulfill those needs with plenty of pcie lanes.
Posted on Reply
#84
boe
nemesis.ieYou could look at a motherboard with built-in 5 or 10Gbe interface and put a similar NIC in the other PC as an add-in board? Then just connect them with a crossover cable. Use the original port for internet access.

Some of the Asrock boards have an Aquantia 5 or 10Gbe as well as a 1Gbe (or two) on them.

Once the NIC saturates it doesn't matter how fast the drives on either end are or what lanes they are connected to.
I use 1 1GB nic for the internet, 1 1gb nic for configuring switches and firewalls and the remainder 4x 10gb ports as a poor mans 10gb switch to go to my other PCs.
Posted on Reply
#85
newtekie1
Semi-Retired Folder
bugYeah, UDMA was a big deal back in the day. However PCIe is different. It's a hierarchical structure. And idk whether leaves can talk to each other without going all the way to the root. The engineer in me says you don't come up with a tree-like structure unless you want the branches to be able to work on their own. But the same engineer needs a spec/paper that says that is indeed the case.
It has to go back to the root, but in the case, the root is the PCH(Southbridge) or the CPU(Northbridge on die). There are two roots, and those two roots can talk to each other over the DMI link between the CPU and PCH.
boeSo you say it is clear but it isn't clear to me - maybe I'm just obtuse or maybe my pants are too big. If I have a 16x video card, 8x raid card and 8x nic - all of which are PCIe 3.0 it sounds like my max speed for anything other than my 16x video card is 4x and I don't know if I have even 4x to spare for the other 2 cards as I don't know how many PCIe lanes are used by the onboard sound, raid, sata, wifi, usb ports, etc.

I appreciate your help on clarifying this as I'd really like to know. I can pick a processor and specific motherboard if it helps you to give me an answer that helps me go in the right direction. I use my PC for gaming, work and as a HTPC. I have 150TB of storage and often transfer it to my other PC which has 100TB of storage so I hammer my nic and raid controller about once per week -usually about 4TB of transfer with updates and changes.

Getting a clean PCIe answer has been as challenging for me as finding out when I can actually use my frequent flyer miles.
It is going to vary between different boards. Some motherboards wire all the x16 slots to the CPU, this means they will have to split the 16 lanes for those slots. Others will only have the first two slots connected to the CPU, and others still only connect the first to the CPU.

What CPU and board are you looking at?
Posted on Reply
#86
nemesis.ie
boeI use 1 1GB nic for the internet, 1 1gb nic for configuring switches and firewalls and the remainder 4x 10gb ports as a poor mans 10gb switch to go to my other PCs.
So you are using a 4 port 10Gbe card in each machine? What model, that sounds expensive. If not, I might want a couple. ;)
Posted on Reply
#87
bug
newtekie1It has to go back to the root, but in the case, the root is the PCH(Southbridge) or the CPU(Northbridge on die). There are two roots, and those two roots can talk to each other over the DMI link between the CPU and PCH.
Yes, that's the part I'm missing: confirmation that the PCH root can handle things on its own (when both parties are connected to the PCH root, of course) and it doesn't have to go to the CPU.
Posted on Reply
#88
boe
I was considering something like the 8700k combined with the Z370 Taichi Asrock motherboard -never used an Asrock before. I'm not married to any particular brand of motherboard or chipset (although I've had a number of MSI boards die so I don't think I'd want another one of those - I replace with gigabyte boards that have lasted 3 times as long and are still running) - I typically use gigabyte and might consider the Z370 AORUS Gaming 7.

I'm definitely going with the
NVidia 1180 Video card (probably asus strix)
X710-T4 network card
9460-16i raid controller
I'm also using a butt load of usb devices and the onboard nvme if it matters.
I will occasionally use wifi just for testing - not frequently.
nemesis.ieSo you are using a 4 port 10Gbe card in each machine? What model, that sounds expensive. If not, I might want a couple. ;)
In my new machine I'll have quad port that I'll dual port trunk to two machines. In my old machines I have the old intel dual 10g nics -ports (x540). I only use them for backups so I'll have 20gbit which is more than enough as my raid will max out well before that. It exceeds 10gbit but I don't know by how much as currently I'm only connected at 10g.
Posted on Reply
#89
nemesis.ie
So why not look at e.g. the Asrock X470 (+ Ryzen 2xxx) with a 10Gbe on board and 2 x M.2s that you can RAID for >6GB/s? Or as mentioned, threadripper? Is it a budget issue?

What NICs are you using?
Posted on Reply
#90
boe
nemesis.ieSo why not look at e.g. the Asrock X470 (+ Ryzen 2xxx) with a 10Gbe on board and 2 x M.2s that you can RAID for >6GB/s? Or as mentioned, threadripper? Is it a budget issue?

What NICs are you using?
The ryzen - even ryzen 2 have less FPS for gaming than the Intel 8700k (ryzen 2 about 10% slower in gaming with a 1080ti - not sure how much it will be with the 1180. I don't do any video editing or photoshop so I don't care if it is 4 core or 400 core). I use it for gaming as well so I want the best gaming performance. X710-T4
Posted on Reply
#91
newtekie1
Semi-Retired Folder
boeI was considering something like the 8700k combined with the Z370 Taichi Asrock motherboard -never used an Asrock before. I'm not married to any particular brand of motherboard or chipset (although I've had a number of MSI boards die so I don't think I'd want another one of those - I replace with gigabyte boards that have lasted 3 times as long and are still running) - I typically use gigabyte and might consider the Z370 AORUS Gaming 7.

I'm definitely going with the
NVidia 1180 Video card (probably asus strix)
X710-T4 network card
9460-16i raid controller
I'm also using a butt load of usb devices and the onboard nvme if it matters.
I will occasionally use wifi just for testing - not frequently.
So on the Z370 Taichi, the three PCI-E x16 slots all run off the CPU. So, with your setup, your graphics card will get an x8 link in the top slot, the NIC and RAID cards will get x4 links in the other two slots. Everything else on the motherboard runs off the 24 lanes from the chipset.

On the Z370 AORUS Gaming 7, the top two PCI-E x16 slots run off the CPU, the bottom one runs off lanes from the chipsets. So your GPU will get an x8 link, then whatever card you plug into the second slot will get an x8 link as well, and the card in the bottom slot will get an x4 link.
Posted on Reply
#92
boe
Thanks - good to know. I did not realize that so that is very helpful to me. I hate to get a 7900x - partially because of the cost but also because I leave my computer on 24x7 and electricity is expensive here.
Posted on Reply
#93
cdawall
where the hell are my stars
boeThanks - good to know. I did not realize that so that is very helpful to me. I hate to get a 7900x - partially because of the cost but also because I leave my computer on 24x7 and electricity is expensive here.
Cpu at idle difference isn't much. The platform itself can draw more, but considering you have a whole heck of a lot running the difference won't be as much as you think.
Posted on Reply
#94
boe
You are correct. Still leaves me with about a $500 difference just to get more PCIe lanes but the gaming won't be any faster. I'm not saying you are wrong - just some how feels like Intel is asking me to bend over without even giving me breakfast in the morning. I was hoping Intel would have a new lower nm fabrication i9 out this year but that looks extremely unlikely at the moment. No idea if the new z390 chipsets would benefit me for PCIe lanes or what the or x399 does to modernize the expensive intel processor motherboard which usually seem about a generation behind most gaming PCs other than the PCIe lanes.
Posted on Reply
#95
cdawall
where the hell are my stars
Intel is asking you to buy HEDT or their server platform if you want a server number of gpu lanes.
Posted on Reply
#96
nemesis.ie
boeThe ryzen - even ryzen 2 have less FPS for gaming than the Intel 8700k (ryzen 2 about 10% slower in gaming with a 1080ti - not sure how much it will be with the 1180. I don't do any video editing or photoshop so I don't care if it is 4 core or 400 core). I use it for gaming as well so I want the best gaming performance. X710-T4
But if more cores are used better in the future, the 8 of the Ryzen will work out better in the long run and an extra 10GBe is a nice thing to have IMO. I suppose it depends on how long you plan to keep a system.
Posted on Reply
#97
boe
You may be right but I sincerely doubt it any time in the next 5 years. I only have 4 cores on my 6700k and it still beats the multicore 2600x. I think an 8700k or 9700k with 6 or 8 cores will probably bury the 2600x. However, the ryzen 3 might have their design improved enough by then that it will outperform an 8700k - not because of a googleplex of cores but because the processor is faster.
Posted on Reply
#98
bug
boeYou may be right but I sincerely doubt it any time in the next 5 years. I only have 4 cores on my 6700k and it still beats the multicore 2600x. I think an 8700k or 9700k with 6 or 8 cores will probably bury the 2600x. However, the ryzen 3 might have their design improved enough by then that it will outperform an 8700k - not because of a googleplex of cores but because the processor is faster.
It all comes down to the software you use more. Some apps junst don't multithread that easily (if at all) and those will always run better on fewer faster cores. Those that do will benefit from as many cores as you can throw at them.
At the end of the day, beyond 3D rendering and video editing few applications need tons of threads. I believe game engines only recently (past two years or so) have broken the 4 core barrier. While I can't speak for others, I know my quad core will easily get the job done for a few more years.
Posted on Reply
#99
nemesis.ie
Indeed, in gaming I'm seeing some games + some games - between the competitors: www.pcper.com/reviews/Processors/Ryzen-7-2700X-and-Ryzen-5-2600X-Review-Zen-Matures/1440p-and-4K-Gaming

It also takes around 10% faster to notice a significant difference and smoothness (consistency) is also important, the "headline" benchmark charts often to not report "perceived gameplay". If you do things like file transfers in the background while gaming, extra cores should help. IMO if you get to the point of a 10% difference but are seeing 100+fps with consistent frame times you'd be hard pressed to feel a difference outside competitive FPS gaming.

More cores is the future IMO.

If I was buying a machine now to last for up to 5 years with maybe only a graphics card swap or 3 I'd go for the better multi-threaded performance. Look what has happened with the 7700k, a year ago it was the gaming king, today I think it's "not so much" versus the more-cored CPUs. Then there are the platform considerations ...
Posted on Reply
#100
boe
You are of course entitled to your opinions and you may be right but I doubt I'll see anything on my pc that will benefit beyond 8 cores in the next 5 years (no photoshop, no video editing, not a lot of compression or code breaking and since I buy real raid controllers they do all the work - no load really on the CPU. Also the ryzen still has a PCIe slowdown that hasn't been addressed since the ryzen 1. I'm not knocking AMD - they have really come a long way in the past year for CPUs and I'd actually like to see them beat Intel as competition is great but right now I don't think they are the king but certainly closing. My first PC that I built was a 386-40 AMD so I'm not shy about getting AMD when they are the best in class.
Posted on Reply
Add your own comment
May 14th, 2024 16:57 EDT change timezone

New Forum Posts

Popular Reviews

Controversial News Posts