Still on 14nm (But better) - Rocket lake - S to come with PCIe 4.0

[williamcll]

Last mentioned in December 2019, the new chip series will be sharing the same socket as the upcoming Comet Lake - S' LGA 1200  and will have the 20Gbps USB C connection on the chipset. But it might be awhile before these CPUs come out.

Quote

The platform codenamed Rocket Lake-S is currently expected in late 2020. The rumors about the RLK-S platform have been floating around the web for a while now, but only now (thanks to our sources at Intel) we are able to provide some concrete information on a new platform.

• The new processor core architecture
  The slide confirms that Rocket Lake-S will feature new core architecture, without stating any other details. What is rumored, however, is that Rocket Lake-S is rumored to be 14nm adoption of Tiger Lake, which is using Willow Cove cores.
• 20 lanes of PCI Express 4.0
  The most important change for Rocket Lake-S is PCIe 4.0 support. Not only will the CPU have direct 4.0 lanes, but there will be 4 additional lanes for storage (x16 for GPU and x4 for NVME drive). This means that both the primary GPU and NVME storage will be attached directly to the CPU, not the PCH.
• DMI 3.0 x8
  Direct Media Interface will be upgraded to x8 link, which means doubled transfer speed compared to x4. Intel does not state the transfer speed for a new DMI connection, but the current x4 link has a transfer of 8 GT/s (3.93 GB/s).
• Intel Xe Graphics and display updates
  The slide confirms that Rocket Lake-S will benefit from Xe graphics architecture (further evidence that RKL is a Tiger Lake desktop clone?). The upgrade to Xe (presumably Gen12) will also bring HDMI 2.0b and DisplayPort 1.4a support.
• Thunderbolt 4 and USB 3.2 20G
  At CES 2020 Intel confirmed that the Tiger Lake platform will support Thunderbolt 4. This standard does not provide an upgrade over Thunderbolt 3 in terms of transfer speed (it is still 40 Gb/s). There has been a rumor that 4.0 might be using PCIe 4.0 but the slide clearly states that TB4.0 is still using PCIe 3.0.
• SGX removed
  Intel Software Guard Extensions have been removed for Rocket Lake-S. 

 

Source: https://videocardz.com/newz/exclusive-intel-rocket-lake-s-features-pci-express-4-0-xe-graphics

Thoughts: This seems like an uphill batte against AMD right here, but I might be speculating too much, I'll see what intel has to offer next month and I'll make a decision then.

[FloRolf]

You mean pcie 4.0 and not pcie x4 there m8. 

 

But I feel like no matter what intel does these days, most people will just go "eh" anyways

 

[WereCat]

So the Rocket Lake is all about platform I/O improvements but still basicaly no tangible CPU performance improvement?

[porina]

Just now, WereCat said:

So the Rocket Lake is all about platform I/O improvements but still basicaly no tangible CPU performance improvement?

The quotes mention Willow Cove, which if true, expect IPC improvements. I think Willow Cove is the one after Sunny Cove as used in Ice Lake. This CPU overall should be Zen 2-3 like, apart from still being on 14nm.

 

[lafrente]

Trying to compensate for something with that "rocket" branding.

[5x5]

Weren't Intel claiming PCIe 4.0 is usless a few months ago? Guess they're still in panic mode.

[Dabombinable]

Lol, this will kill their 14nm supply even further.

[RejZoR]

What is weird to me, can't Intel hire TSMC or GloFo to make their 7nm chips? Or is their gate design so specific they can't roll their chips to those foundries and make them do it? Or they just refuse to do it no matter what because it would be bad PR to say their own foundries can't do it and they just stubbornly refuse to ask someone else for "help"?

14nm, no matter the design or how good it behaves still, it feels like using 45nm design in 2020...

[porina]

38 minutes ago, 5x5 said:

Weren't Intel claiming PCIe 4.0 is usless a few months ago? Guess they're still in panic mode.

For the vast majority of people, PCIe 4.0 today has very little to no value. This CPU I'd guess wont be out until 2021, by which point newer PCIe may become more relevant.

36 minutes ago, Dabombinable said:

Lol, this will kill their 14nm supply even further.

Since it replaces older products, it is more a shift in production than new production.

[RejZoR]

Actually, people are underestimating the importance of PCIe 4.0 and newer with insane bandwidth. Future of games is where graphic cards will be able to fetch resources directly from system on per-need basis and not entirely rely on VRAM. Sure, VRAM will still play a huge role because only VRAM can be fast enough for the framebuffer and keeping frame times low, but there is no need to store all textures in memory. Something we actually don't do anymore in most cases (Texture Streaming), but we could further expand on that, making environments even bigger and richer looking. And only way to achieve that is to push PCIe speeds to such high levels that it would basically be indistinguishable between CPU, RAM and PCIe slots in terms of speeds. SSD's will again be the slowest storage medium in this whole deal, but still usable at speeds of 4,5 GB/s and beyond. Oh and we need to get this to as many systems as possible, so game devs take advantage of it. If only select few thousands systems have this, they won't bother.

[porina]

4 minutes ago, RejZoR said:

What is weird to me, can't Intel hire TSMC or GloFo to make their 7nm chips?

Presuming you mean TSMC or Samsung. GoFlo are pretty much out of the race.

 

4 minutes ago, RejZoR said:

Or is their gate design so specific they can't roll their chips to those foundries and make them do it? Or they just refuse to do it no matter what because it would be bad PR to say their own foundries can't do it and they just stubbornly refuse to ask someone else for "help"?

The level of redesign required is not trivial. The same argument can be made for, if 10nm is broken, why not just make it on 14nm? It really isn't something that could be done quickly. If you wanted to do it at all, it needs to be planned from the start. Deciding part way through, by the time they get it working it'll be obsolete. You might then argue, they could hedge bets for next generation product. But that's on the presumption of a high likelyhood of failing to recover from the process internally. It is looking ever more like 10nm is going to be the forgotten generation and Intel hopes to recover competitiveness at 7nm.

 

2 minutes ago, RejZoR said:

Actually, people are underestimating the importance of PCIe 4.0 and newer with insane bandwidth. Future of games is where graphic cards will be able to fetch resources directly from system on per-need basis and not entirely rely on VRAM. Sure, VRAM will still play a huge role because only VRAM can be fast enough for the framebuffer and keeping frame times low, but there is no need to store all textures in memory. Something we actually don't do anymore in most cases (Texture Streaming), but we could further expand on that, making environments even bigger and richer looking. And only way to achieve that is to push PCIe speeds to such high levels that it would basically be indistinguishable between CPU, RAM and PCIe slots in terms of speeds.

Take some mid range cards, like a 2070 Super or 5700XT both with 448 GB/s VRAM bandwidth. 16x 4.0 is around 32GB/s. System ram peak bandwidth at dual channel 3600 is about 56GB/s. Maybe once we get to PCIe 5.0 with DDR5 that could start to be more interesting.

 

2 minutes ago, RejZoR said:

Oh and we need to get this to as many systems as possible, so game devs take advantage of it. If only select few thousands systems have this, they won't bother.

Similar arguments could be made for different features. I'd rather have RT as standard than 4.0 as standard.

[TVwazhere]

What I find interesting is the removal of USB2.0.... For the ports I dont care, but what about internal headers? A LOT of fan and RGB controllers on the market are using 2.0 motherboard headers for software control of things (not saying it's the best or ideal) but that will limit the use of some products.

 

Not saying it's totally bad though. Whens the last time anyone's truly used a USB2.0 flash drive willingly? And we do need to phase out old unused parts of the board at some point (Old PCI slots)

[porina]

1 minute ago, TVwazhere said:

What I find interesting is the removal of USB2.0.... For the ports I dont care, but what about internal headers? A LOT of fan and RGB controllers on the market are using 2.0 motherboard headers for software control of things (not saying it's the best or ideal) but that will limit the use of some products.

I'd guess it would still be possible for mobo manufactures to add 2.0 style connectors even if they're connected to native 3.0 controllers, if they feel there is still a need for it in the market at the time. There already exists adapter cables that go between mobo and elsewhere 2.0/3.0, in either direction so it isn't like you can't work around it.

[RejZoR]

You can convert internal USB headers between 2.0 and 3.0. I've had one converted because my case didn't have USB 3.0 or mobo back then, not sure right now. I just got an adapter and plugged it in. It was a trivial thing to do and it worked.

[CarlBar]

2 hours ago, porina said:

The level of redesign required is not trivial. The same argument can be made for, if 10nm is broken, why not just make it on 14nm? It really isn't something that could be done quickly. If you wanted to do it at all, it needs to be planned from the start. Deciding part way through, by the time they get it working it'll be obsolete. You might then argue, they could hedge bets for next generation product. But that's on the presumption of a high likelyhood of failing to recover from the process internally. It is looking ever more like 10nm is going to be the forgotten generation and Intel hopes to recover competitiveness at 7nm.

 

Also Intel has a huge amount of raw production capability compared to TSMC and Samsung put together. They could never fab enough chips for intel even if intel had designs ready to drop in.

[Sauron]

if they add a few more +s they'll get 15nm

[LAwLz]

Lots of great features on that motherboard/CPU.

Hardware accelerated AV1 decoding.

2.5Gbps ethernet.

PCIe 4.0 (and more lanes than previous Intel processors).

 

 

Let's hope the CPUs to go along with it are good too.

 

 

3 hours ago, 5x5 said:

Weren't Intel claiming PCIe 4.0 is usless a few months ago? Guess they're still in panic mode.

I can't find any source that Intel said anything like that. Do you have one?

[schwellmo92]

49 minutes ago, CarlBar said:

 

Also Intel has a huge amount of raw production capability compared to TSMC and Samsung put together. They could never fab enough chips for intel even if intel had designs ready to drop in.

That is completely incorrect. TSMC has more than 3 times the capacity than Intel, and then Samsung has even more. Intel isn’t even in the top 5 for capacity.

[porina]

1 hour ago, schwellmo92 said:

That is completely incorrect. TSMC has more than 3 times the capacity than Intel, and then Samsung has even more. Intel isn’t even in the top 5 for capacity.

However much capacity they have, it's how much free capacity they have at what node, that counts. It'll have to be 7nm class or better to be a worthy upgrade over their own 14nm. We also know TSMC are pretty much booked up in their existing capacity at the better nodes. Intel would have to out-bid everyone else to get part of it in future. This isn't a matter of if they can or not, but if it is worth it for them. I think the cost of doing so is simply too high.

[The King of the Undead]

whose keeping track of the exact amount of pluses?

[Taf the Ghost]

6 hours ago, porina said:

The quotes mention Willow Cove, which if true, expect IPC improvements. I think Willow Cove is the one after Sunny Cove as used in Ice Lake. This CPU overall should be Zen 2-3 like, apart from still being on 14nm.

 

It does appear to be a semi-backport of Icelake to 14nm.

 

As a note, Rocket Lake is Intel's 2021 Desktop product line. Comet Lake, is starts the official announcement phase in April, is just a true Skylake Refresh...again.

[Taf the Ghost]

12 minutes ago, The King of the Undead said:

whose keeping track of the exact amount of pluses?

Not even Intel, as even they gave up. Though we should probably start using 14b through 14z or something, because there's actually different library options at all of the pluses. 

[CarlBar]

5 hours ago, schwellmo92 said:

That is completely incorrect. TSMC has more than 3 times the capacity than Intel, and then Samsung has even more. Intel isn’t even in the top 5 for capacity.

 

I should have added the qualifier but didn't want to confuse some people by potentially getting over specific. I was specifically talking about processor grade node fabbign capacity. The overwhelming majority of the worlds fab capacity for semi-conductors is in NAND and RAM manufacturing. Cut things down to processes and fab facilities capable of handling processors and you all but eliminate most of the non-intel players. Cut it down to high performance processes suitable for desktops and laptops and Intel runs away with a ridiculous amount of the available fab capacity. Mostly because thats virtually all intel's fabs do. I've really got 10x or more the surface area, (and thus in very loose terms fab capacity worth of), of my CPU die in GDDR, DDR, and SSD modules in my current system for example. But none of the stuff used to make that could make the processor. 

[thechinchinsong]

Wonder why they removed SGX...

[schwellmo92]

Just realised this now after coming back to the thread. Wasn’t the Xe graphics supposed to consume most of the die size for their future chips? Couple that with them still using 14nm here and we may be in for some quite large die sizes.