News Intel 14nm Rocket Lake-S Leaked: New Core Architecture, Xe Graphics, PCIe 4.0

[InvalidError]

Even adding more cores on to 14nm isn't out of the question

TDPs are getting out of control, I'd say that's a plenty good enough reason to call it quits on throwing more cores at 14nm for the mainstream. Most desktop software isn't heavily threaded enough to benefit from it anyway.

 

[bit_user]

At a die size of ~177mm2 Intel could add all the additional logic into a 14nm chip. Will the die increase in size, yes, but the original quad core Nehalema was 260mm2.

You're comparing 45 nm to 14 nm, as if the cost of die area is constant, which I'm sure it's not. But, cost isn't even my chief concern.

 

[bit_user]

TDPs are getting out of control, I'd say that's a plenty good enough reason to call it quits on throwing more cores at 14nm for the mainstream. Most desktop software isn't heavily threaded enough to benefit from it anyway.

Agreed, although nobody seems to have told AMD that mainstream doesn't need ever more cores.
; )

It actually irks me that to get their highest-binned chiplets, I need to buy a 16-core Ryzen or 32-core ThreadRipper, when I don't need that many cores. I keep hoping they'll release a single-chiplet Ryzen with a single, top-bin chiplet.

 

[InvalidError]

Agreed, although nobody seems to have told AMD that mainstream doesn't need ever more cores.
; )

Mainstream doesn't, but the people who fall in-between mainstream and HEDT do, especially after AMD jacked up the entry-level price for HEDT so much with Zen 2.

As for having to buy a 3950X for premium chiplets, it at least used to be that the 3900X had one (very) good chiplet and one average / 3600-class chiplet. If that is still the case today and all you want is a couple of faster cores for lightly threaded workloads, then you can simply get AMD's second best in the lineup and save a few hundred bucks. Where a single chiplet variant of the same is concerned, that is kind of what the 3800X is supposed to be but falls short from being and for the price premium AMD is asking for it over the 3700X, it certainly should be.

 

[spongiemaster]

To unlock more IPC potential, they have to go to a smaller process node, as they did with Ice Lake. Otherwise, clock speeds are going to suffer, and it's hard to believe there's enough additional IPC to be had that they can make up for it and then some. We're certainly not looking at another Core 2-style revolution.

If Rocket Lake has 10% lower clocks, and a 400-500MHz drop would be pretty massive, it would still be quite a bit faster than Skylake if the IPC increase is 15-20%. A 4.5Ghz all core boost with only a 15% IPC increase would put Intel comfortably ahead of Ryzen 3000 in performance per core. It should even be enough to keep it in a close race with Ryzen 4000.

 

[spongiemaster]

TDPs are getting out of control, I'd say that's a plenty good enough reason to call it quits on throwing more cores at 14nm for the mainstream. Most desktop software isn't heavily threaded enough to benefit from it anyway.

Besides the obvious dearth of heavily threaded mainstream software, isn't the ring bus becoming more of a deterrent than anything else for adding more than 10 cores on the mainstream platform?

 

[bit_user]

Where a single chiplet variant of the same is concerned, that is kind of what the 3800X is supposed to be but falls short from being and for the price premium AMD is asking for it over the 3700X, it certainly should be.

Yeah, so I'm saying what I want is a 3850X that features their top-binned chiplet.

Maybe the pricing is what doesn't work about that. Perhaps it would land too close to the 3900X.

 

[bit_user]

isn't the ring bus becoming more of a deterrent than anything else for adding more than 10 cores on the mainstream platform?

10 is the most they've ever had on a single-ring chip. Going beyond that, they added a second ring (either half or full). That said, it does appear that Broadwell's dual-ring chips went up to 12 cores/ring.

Pics for Broadwell (server) are here: https://en.wikichip.org/wiki/intel/xeon_e5#Die_Stats

AFAICT, Sandybridge maxed out at a single ring, while Ivy Bridge scaled up to two. This shows the largest Ivy Bridge and Haswell (server) core config topologes:

https://www.anandtech.com/show/8423/intel-xeon-e5-version-3-up-to-18-haswell-ep-cores-/4​

​

Of course, the Skylake generation went to a mesh, which scales better but initial benchmarks showed that latency suffered. I don't know if Cascade Lake managed any better.

 

[InvalidError]

Yeah, so I'm saying what I want is a 3850X that features their top-binned chiplet.

Maybe the pricing is what doesn't work about that. Perhaps it would land too close to the 3900X.

That is kind of exactly what I meant when I wrote that this is what I would have expected from the 3800X for the price premium AMD charges over the 3700X. There shouldn't have been a perceived need for a 3850X. The current 3800X may not even warrant branding as a 3750X.

 

[PCWarrior]

That makes no sense. Why would they make the chipset x8 3.0, if the CPU's DMI link were x4 4.0? That just means you have to put additional logic on the board, to convert it. That's just wasting money & board realestate.
I was only talking about their DMI link, which is hard-wired between the CPU and chipset. The chipset has a switch and can downgrade all the way to 1.0. But, all of its subordinate links could remain 3.0, in the current widths, and you'd get the same bandwidth to the CPU with either a x4 4.0 or x8 3.0 DMI link.

I was referring to the likelihood these cpus to actually be having 24 PCIe4 lanes from which 20 are direct (16 for graphics/etc and 4 for fast storage) and 4 are destined to go to the chipset. Then those 4 PCIe4 lanes (that are destined to go to the chipset) are converted into 8 PCIe3 lanes (or 2xDMI3.0) before leaving the cpu, exactly in order for such a conversion to not be needed at the chipset side through costly additional logic (which would most likely be more expensive than the cost of more PCIe traces).

I believe that the reason Intel is not deploying DMI4.0 with Rocket lake is because probably that would require breaking backward compatibility with Z490 boards. Never before have cpus with a higher DMI version been paired with boards launched alongside cpus with a lower DMI version. It seems easier for Intel to essentially disable half of what is essentially a “dual DMI.3.0” link when a Rocket lake CPU is paired with a Z490 board. Also moving to a new version of DMI is not something that Intel has done quickly in the past. For Z77, Z87, Z97, X99 the link remained DMI 2.0 (x4PCIe2) despite PCIe3 direct cpu lanes being around for 3 generations. It should also be noted here that historically all the direct chipset lanes are whatever the DMI link PCIe lanes are. For Z77, Z87, Z97, X99 where you had DMI2.0 all the chipset lanes are PCIE2.0. Similarly for Z170, Z270, Z370 and Z390 where the DMI link is 3.0 (x4PCIe3), all the chipset lanes are PCIe3.

I would point to Sandybridge vs. Ivy Bridge. Both share the same socket (in fact, I run a i7-2600K CPU in an Ivy Bridge-capable HD77KC motherboard). With a Sandybridge CPU in it, the CPU-connected x16 slot is only 2.0, but it switches to PCIe 3.0, when the board is used with an Ivy Bridge CPU.

Your example with the 2600K in H77 boards is the exact same as using an X570 board with a 3700X and 2700X. That in no way counters my point however. Paul was referring to an article claiming that Z490 chipsets had an issue supporting PCIE4. How can people expect a motherboard to be supporting PCIe4 chipset lanes when it launches alongside a cpu that doesn’t support PCIe4? And then claim there must be a problem because such a unicorn board won’t launch! But what is more bizarre is taking this supposed ‘problem’ and using it as ‘evidence’ to infer a ‘problem’ with a product even further down the line. Thre would be a counter-point only if P68 boards (which launched alongside SandyBridge 2000 series cpus) could magically support PCIe3 chipset lanes when you put an Ivybridge 3770K in them. Only then you have precedence to expect boards launched alongside Comet lake supporting PCIE4 chipset lanes.

 

[cryoburner]

That is kind of exactly what I meant when I wrote that this is what I would have expected from the 3800X for the price premium AMD charges over the 3700X. There shouldn't have been a perceived need for a 3850X. The current 3800X may not even warrant branding as a 3750X.

It might be that they don't have enough of those top-binned chiplets to make relatively mainstream parts out of them. Even as it was, there seemed to be major shortages of the 3900X following its launch. I imagine processors in the sub-$400 range outsell those higher core-count parts by a large margin, so it might not have been practical to give the 3800X a higher-binned chiplet.

 

[InvalidError]

It might be that they don't have enough of those top-binned chiplets to make relatively mainstream parts out of them.

If AMD did not have enough decent chiplets to make the 3800X more than a 4% better chip for 30% higher MSRP, it shouldn't have launched it in the first place. That would allowed it to brand the 3900X as the 3800X and the 3950X as the 3900X. A far more sensible product stack naming scheme IMO.

 

[mcgge1360]

"leak" is right. Why are the words saying PCIe 4.0, but the graphics in the middle show PCIe 3.0?

I wonder if Intel knows 10th gen is a dude, released at least 18 months after 9th gen and is not close to 11th gen being ready. Never mind it still won't be competitive with Zen 2, nor the upcoming Zen 3. And so they are starting to get the market warmed up for 11th gen.

Competitive with Zen 2? Intel still has the market share for desktop and laptops by a big lead. Remember most AMD CPUs don't have integrated graphics, so it adds an additional cost for manufacturers. And where the most computers sell (mainstream laptops) AMD is still barely fooling around. Intel's 8th gen i5 (8250u) kinda destroys even the best mainstream (3700u) laptop r7. Nevermind the revisions like the 8265u, and the 10th gen laptop SKUs like the 1035G1. AMD needs to focus on where sales actually are: Laptops. And at this point in time they're far from it.

 

[InvalidError]

AMD needs to focus on where sales actually are: Laptops. And at this point in time they're far from it.

AMD scored design wins for a ton of formerly Intel-exclusive lineups including Microsoft's Surface. Intel's struggles with 10nm are starting to hurt and force OEMs away.

 

[bit_user]

It might be that they don't have enough of those top-binned chiplets to make relatively mainstream parts out of them. Even as it was, there seemed to be major shortages of the 3900X following its launch.

This is why I hoped they do a 3850X, as yield improves and stocks are built up.