Intel Raptor Lake-S CPU-attached NVMe Storage Remains on PCIe Gen4

[AleksandarK]

Intel is preparing to launch its next-generation desktop platform codenamed Rocket Lake-S. According to the presentation held by Intel today in Shenzen, China, we have official information regarding some of the platform features that Raptor Lake is bringing. Starting with memory support, Raptor Lake is still carrying the transitional DDR4 and DDR5 support, as the full swing towards DDR5 is still in progress. Unlike the previous generation Alder Lake, which brought DDR5-4800 support, Raptor Lake's integrated memory controller can drive DDR5 modules with a 5600 MT/s configuration. As DDR4 support remains, it is limited to 3200 MT/s speed.

Interesting information from the leaked slide points out that support for CPU-attached NVMe storage remains PCIe Gen4. While AMD will provide an AM5 socket with CPU-attached NMVe storage on PCIe Gen5 protocol, Intel is taking a step back and holding on to Gen4. The CPU is outputting 16 PCIe Gen5 lanes on its own. Motherboard vendors for the upcoming 700-series boards for Raptor Lake can still provide a PCIe Gen5 NVMe slot; however, it will have to subtract eight Gen5 lanes from the PCI Express Graphics (PEG) slot and route them to NVMe storage. As our testing shows, this will affect GPU's performance by a few percent. AMD's upcoming AM5 platform has no such issues, as the CPU provides both the PEG and CPU-attached NVMe storage with sufficient PCIe Gen5 bandwidth.


View at TechPowerUp Main Site | Source

 

[ratirt]

It would seem, with the PCIe lanes, this is a cost cut sacrifice. Is it because Intel wants to stay competitive with price?

 

[Patriot]

It would seem, with the PCIe lanes, this is a cost cut sacrifice. Is it because Intel wants to stay competitive with price?

Power, they need that power allotment for CPU.

 

[AlwaysHope]

At this stage of the evolution with PC games, a PCIe 5 drive is utterly useless. The devs still haven't optimised for PCIe 4 yet. Storage drive reviews on here make that case obvious.
I welcome rapid lake & AM5, along with RX 7000 & Nvidia's 4000 series coming out before year end... what a time to be alive!

 

[ZetZet]

If this helps them cut prices and DDR4 is still good enough with Raptor Lake I don't see much benefit in going with AMD, unless you upgrade every 1-2 years and want to bet on AM5 early on. Intel should be winning the value battle.

 

[Deleted member 24505]

Sure there is enough here to get the AMD fans hooting and jeering at Intel as usual now on TPU. I'm looking forward to trying one of these in my 690 board.

 

[Bomby569]

I would sacrifice a useless feature for less cost at this point, mobo prices are getting ridiculous and 99% of users don't even use full gen 4 speeds let alone 5. I have a gen 4 capable board and happy with a gen 3 nvme

 

[HenrySomeone]

Sure there is enough here to get the AMD fans hooting and jeering at Intel as usual now on TPU. I'm looking forward to trying one of these in my 690 board.

They don't have much else at their disposal when their chips are once again well behind core-for-core. They might get closer with Zen4 vs. Raptor Lake, but next year's Meteor Lake is what I'm expecting to be a very painful blow...

 

[docnorth]

For me the most interesting information is the native support of 5600 MT/s DDR5 speed, like the early rumors suggested. If this can be achieved with the standard primary latency and voltage, cl40 and 1,1v respectively, it will be a big leap for DDR5.

 

[zlobby]

No way, Jose!

 

[Aretak]

At this stage of the evolution with PC games, a PCIe 5 drive is utterly useless. The devs still haven't optimised for PCIe 4 yet.

Devs haven't "optimised" for a 3.0 NVMe drive yet. You'll get exactly the same performance out of a SATA SSD for gaming purposes. That will only change when DirectStorage titles start arriving, the first of which seems to be Forspoken in October.

 

[dont whant to set it"']

Classic Intel:cheaping out. long live the quad intel core.

 

[Deleted member 24505]

Classic Intel:cheaping out. long live the quad intel core.

How is it cheaping out? because AMD will have near useless PCIe 5 GPU slot and storage control when there will be no PCIe 5 video cards and maybe only very few and very expensive PCIe 5 NVME's. well done AMD

 

[GerKNG]

"Intel is preparing to launch its next-generation desktop platform codenamed Rocket Lake-S"

 

[ncrs]

How is it cheaping out? because AMD will have near useless PCIe 5 GPU slot and storage control when there will be no PCIe 5 video cards and maybe only very few and very expensive PCIe 5 NVME's. well done AMD

Since AMD sockets are longer lasting than Intel sockets, historically of course, it makes sense that the former are more future proof. It also might be that AMD doesn't want a repeat of the PCIe 4.0 on older motherboards fiasco.

 

[Chrispy_]

Even the 3090Ti doesn't need all of its lanes, and the only reason you'd realistically want more bandwidth on the SSD is for DirectStorage where the SSD is talking to the GPU. In that instance, balancing out the bandwidth between SSD and GPU is the only way to alleviate any bottlenecks and at the moment the horribly inbalanced 16 lanes for GPU and 4 lanes for SSD makes no sense in that use case.

As for other use cases that aren't DirectStorage, what real-world SSD scenario requires 13.5GB/s transfers in a way that will provide meaningful improvements that a human being can notice compared to what they'll get with "only" 7GB/s transfers of PCIe 4.0 x4? The only benefit I can think of right now is being able to stroke fragile egos when they see a 5-digit number in CrystalDiskmark instead of a 4-digit number

 

[TechLurker]

One underutilized advantage to AMD's setup that I wish mobo makers would do, is that mobo makers could downgrade the innate PCIe 5.0 lanes to 4.0 and effectively double the available lanes available (effectively serving like a budget mini-Threadripper in a sense), and add more I/O options to the board itself. Kind of like the old 7-10 slot PCIe mobos of the late 2000s/early 2010s when XL-ATX was a popular form factor variant and slotting in quad cards were a thing. In this case though, I'd love to see a few pro-level workstation like boards with PCIe 5.0 downgraded to 4.0 but having enough lanes to run 3-4 NVMe Cards + GPU. Then take it a step further with "budget" pro versions that downgrade some of the PCIe 4.0 to 3.0 for even more lanes, making use of the existing ecosystem of PCIe 3.0 gear to slot maybe 5-6 PCIe 3.0 NVMe cards + GPU.

 

[ppn]

PCIE 5.0 IO logic is a slightly enlarged version of PCIe 4.0. but it doesn't translate into, you get only half of the lanes or slower speed. there is just not enough space to fit it in width without overlapping with the memory controller on the right side.

I guess if the GPU can do away with only using 8 lanes, the remaining 8 can be split for the NvMe why not. Can it also do 12+4 or is it strictly 8+8.

 

[Deleted member 24505]

Same as Intel putting PCIe 5 on z690 Video card slot, was pointless and a waste. They didn't even bother having no PCIe 5 storage control. Apart from storage control, looks like AMD will make the same waste of PCIe 5

 

[Chrispy_]

PCIE 5.0 IO logic is a slightly enlarged version of PCIe 4.0. but it doesn't translate into, you get only half of the lanes or slower speed. there is just not enough space to fit it in width without overlapping with the memory controller on the right side.

View attachment 253730


I guess if the GPU can do away with only using 8 lanes, the remaining 8 can be split for the NvMe why not. Can it also do 12+4 or is it strictly 8+8.

GPUs are fine with 8 lanes - the disadvantage is margin-of-error in testing, even at PCIe 4.0
8 lanes for GPU, 8 lanes for SSD is ideal, and that frees up the 4 additional lanes that the SSD would otherwise be using if the GPU had all 16.

 

[Tek-Check]

Same as Intel putting PCIe 5 on z690 Video card slot, was pointless and a waste. They didn't even bother having no PCIe 5 storage control. Apart from storage control, looks like AMD will make the same waste of PCIe 5

I do not think it's a waste for those who would use available bandwidth in a meaningful way. On motherboards with two PCIe 5.0 slots, it's easy to put GPU into one to run at x8 and then another AIC in the second slot to run at x8 too. BIOS allows bifurcation, so the second slot can run two Gen 5 NVMe drives at full speed or four Gen 4 NVMe drives at full speed, depending on how many NVMe drives are allowed to run by BIOS. Almost a workstation set-up.

Another x4 for NVMe drive from CPU is availabel too, plus everything what is on chipset. Plenty of devices can run.

 

[chrcoluk]

As our testing shows, this will affect GPU's performance by a few percent

Well I checked the link, and your testing showed a 1% loss from gen 4 to gen 3, which I guess would be similar going from 4x16 to 4x8, so how have you translated that to a few percent loss going from 5x16 to 5x8?

Hope TPU isnt trying to overplay newer pcie gen importance.

For those who really want the epeen of a gen5 nvme, just buy a pcie adaptor and install the gen 5 drive in there, likely sub 1% affect on the GPU.

gen5 in general is a waste on consumer hardware.

 

[trsttte]

I think the Intel strategy of being the first mover in pcie5.0 with the x16 graphics slot was misguided. They were able to get a few headlines but there's simply no consumer hardware using the spec, let alone taking advantage from it. GPUs aren't able to saturate PCIe gen4 let alone PCIe gen5.

On the other hand, the AMD strategy of going for gen5 nvme first was smarter because storage is already able to saturate gen4 ports (even if there aren't pratical advantages to do so) and they went after the storage manufacturers so gen5 SSDs actually exist any time soon to take advantage of the 5.0 m.2 slots.

 

[ModEl4]

For enthusiasts level PCs (gen5 SSD costs) possibly will be an advantage for AMD when Microsoft Direct Storage with GPU decompression enabled become utilized in a meaningful way in games (although i suspect latency and random performance will play a bigger role than sequential speeds so i really don't know how gen5 SSDs will affect the experience, just like i don't know how gen5 pci-express 16X will effect the GPU decompression procedure (will Ada & RDNA3 be pci-express gen5 based?)

 

[Chrispy_]

I think the Intel strategy of being the first mover in pcie5.0 with the x16 graphics slot was misguided. They were able to get a few headlines but there's simply no consumer hardware using the spec, let alone taking advantage from it. GPUs aren't able to saturate PCIe gen4 let alone PCIe gen5.

AMD had the advantage of launching the first PCIe 4.0 platform at the same time as launching the first PCIe 4.0 graphics card.
Intel are playing chicken-and-egg with PCIe 5.0. There's nothing to really put in a PCIe 5.0 slot, but nobody else will make a consumer PCIe 5.0 device unless the slot is already released to the consumer market.

Thanks, once again, to all the paying beta testers who spent big money on a platform that will be obsolete before things that can meaningfully take advantage of that platform arrive. We'll all be rocking i5-14600 24-core processors on B860 motherboards by the time software that can take advantage of a PCIe 5.0 SSD is common enough to be worth worrying about....