Tuesday, June 17th 2025
Intel "Nova Lake-S" Platform and 8,000 MT/s Memory Configuration Leaks
Intel appears set to retain its familiar 48 PCIe lanes while delivering a significant memory speed boost with the upcoming "Nova Lake-S" desktop platform. Hardware leakers have shared fresh details indicating that Intel will allocate 24 PCIe 5.0 lanes directly from the CPU, with an additional four DMI Gen 5 lanes connecting to the chipset. The chipset itself will provide eight PCIe 5.0 lanes alongside sixteen PCIe 4.0 lanes, matching the total lane count of current-generation parts. For storage and peripherals, the new platform will support eight SATA 3.0 ports along with a wide assortment of USB options. Fourteen USB 2.0 ports will cover legacy devices, while five USB 3.2 ports at 20 Gbps, 10 at 10 Gbps, and 10 more at 5 Gbps. The CPU's PCIe lanes can be configured in multiple ways to suit different use cases: one 16-lane slot paired with two 4-lane connections, two 8-lane slots plus two additional 4-lane links, or a four-by-four configuration supplemented by two extra 4-lane connections.
Memory support is where Nova Lake-S will shine. Whereas the current Arrow Lake-S lineup handles DDR5-6400 natively and may reach 9200 MT/s when overclocked, Nova Lake-S will offer 8000 MT/s out of the box in a one-DIMM per channel, single-rank configuration. This enhancement is expected to provide a performance boost for data-intensive tasks and games, where the rumored 52-core top-of-the-line SKU will need all the possible bandwidth to feed so many cores. A new LGA 1954 CPU socket and motherboard design will be required, although it remains unclear whether the improved memory speed stems from a redesigned on-die integrated memory controller or advanced motherboard trace layouts and power delivery systems. Partners have received preliminary specifications, but no engineering samples or motherboard prototypes have appeared in public benchmarks or regulatory filings as of yet.
Sources:
@jaykihn #1, @jaykihn #2
Memory support is where Nova Lake-S will shine. Whereas the current Arrow Lake-S lineup handles DDR5-6400 natively and may reach 9200 MT/s when overclocked, Nova Lake-S will offer 8000 MT/s out of the box in a one-DIMM per channel, single-rank configuration. This enhancement is expected to provide a performance boost for data-intensive tasks and games, where the rumored 52-core top-of-the-line SKU will need all the possible bandwidth to feed so many cores. A new LGA 1954 CPU socket and motherboard design will be required, although it remains unclear whether the improved memory speed stems from a redesigned on-die integrated memory controller or advanced motherboard trace layouts and power delivery systems. Partners have received preliminary specifications, but no engineering samples or motherboard prototypes have appeared in public benchmarks or regulatory filings as of yet.


23 Comments on Intel "Nova Lake-S" Platform and 8,000 MT/s Memory Configuration Leaks
Since previous Intel chipsets had lanes that could be allocated dynamically I'm assuming it's either 8x PCIe 5.0 or 16x PCIe 4.0 from the chipset. As well as 8x SATA or 8x PCIe 4.0 and either 5x USB 3.2 20G or 10x USB 3.2 10G just like in Z890.
Otherwise there wouldn't be enough bandwidth in DMI?
nice to get more lanes otherwise though
Nova Lake
Razer Lake
LGA 1954 Socket (PCIe 5.0 Certified)
Reminds me of my current MSI MEG Z790 ACE MAX motherboard is (PCIe 4.0 Certified) while my 14900KS is (PCIe 5.0 Certified) hybrid setup.
Intel 12th gen Alder Lake had PCIe 5.0 back in 2021
Chipset always a hybrid gen behind the CPU.
AMD AM6 (PCIe 6.0 Certified) vs LGA 1954
Cheers
The 10 ports USB3.x are included in the 14 ports USB2.0, too. So it's 14 ports of USB alltogether, 4 of which can be only USB2.0, 10 can be 5G or 10G (just depends on what the motherboard vendor "unlocks") or linked together for 20G.So did Intel for USB3.0 on X58/P55 and P65/Z68, for USB3.1 on Z170/Z270/Z370 and for USB3. Gen2x2 on Z390/Z490.
As for AMD chipsets it's not "a mixture of 4.0 and 3.0 lanes" but "4.0 lanes depending on 1 or 2 chipsets and SATA exclusive with 3.0 lanes". Both companies have flexible lanes in their solutions.
AMD only has a Gen4x4 connection right now, even though B650(E)/B850(E)/X870 output 8 lanes Gen4 and with X670(E)/X870E it's 8 lanes on the second chipset, which connects to the firs by Gen4x4 for 12 lanes in total. Intel has up to 14 lanes Gen4 connected via Gen4x4 on it's B860 and even with Z890, there are 24 lanes Gen4 connected via 8 lanes Gen4. There is always much more bandwith on the PCH than between PCH and CPU.
If Z990 is connected to the CPU just by Gen5x4 but has 8 lanes Gen5, you can still connect two Gen5-SSD to your CPU and access each at full speed seperately, just not simultaneously. You can not (or better should not) build a RAID0 out of two Gen5-SSD connected to the CPU and two connected to the chipset, but one would work.I don't understand what you mean. Promotory21 has 12 lanes Gen4 (4 always being used for connection to the CPU or the other chipset) and 4 lanes Gen3. Maybe in AMD's diagram's, only the Gen3-lanes can be used as SATA-ports, but I'm pretty sure I have seen plenty of boards where Gen4-lanes have been used for SATA.
One example is the Gigabyte Aorus X870 Elite WiFi (look at the block diagram in the manual). 4 lanes Gen4 are shared between one M.2 and one x4-slot, 2 lanes Gen3 are used for one x2-slot, the two other lanes for LAN and WiFi. That leaves 4 lanes Gen4, which are used for 4xSATA.Sometimes it doesn't matter. AFAIR, USB3.0 and SATA 6Gbps only reached their full potential when integrated into the chipset, even when controllers at that time used Gen2x2 for connection to chipset, which should have been enough for both. Meanwhile, there was no measurable difference with USB3.1 speed between Z370 with ASM3142 and Z390.
With USB4, we will not know until USB4v2/TB5 120Gbps controllers are available, because the limiting factor will always be on the external device. TB4 controllers were always only Gen3x4 and only now because USB4-controllers are Gen4x4 can we reach the full potential. But how can we measure if TB4 integrated into Intel CPUs from Tiger Lake upwards or USB4 integrated into AMD Rembrandt upwards reach higher performance than ASM4242, if the only thing we can connect to it to benchmark are SSD connected via ASM2464PD(X) @Gen4x4?
The cost of these new boards nowadays aren't a joke compared before..mind you, I was disappointed on getting the X870E Hero just to see the I/O options I have and its quirky..its not something we end users get to pay for
I agree that this way of adding USB4 is more costly and reduces I/O-options. The latter is of no big concern to me, since absolutely als B650/X670 have used those four lanes from CPU the ASM4242 gets connected to for an additional M.2, which I have no use for. This led to many X870(E)-boards offering 1-2 additional M.2 Gen5x4, shared with the x16-slot, which was much rarer in X670.
I choose my Asus Prime X670E-Pro WiFi specifically because it has two Gen4x4-slots, one of which might be used for an USB4-AIC, the other for 10GbE.
What I find slightly infuriating is that even if USB4 gets integrated into Zen6, boards released until now won't support it. Phoenix APUs have USB4 intgerated, but there is only one board that supports this on one USB-C port, all other only support up to 10G on CPU-ports. This one board shows, that it's possible, but even with 800-series refresh no vendor made use of that.For me it was 5th grade, too.
They released 7.0 already, so might be the time to update?
One clear example of such issues happening is with my own x670e ProArt. It's USB4 controller shares bandwidth with the first NVMe slot that's wired to the CPU, even though this is not listed in the manual. Having a NVMe in that slot alongside with using the USB4 and placing some load in both leads to undesired behaviors.
I was aware of such limitation for my case and simply decided to not use the first slot, given that I do use the TB4 port for my video output (4k60) + peripherals on a KVM switch.
Likely, if you use both at the same time, one Gen4x4-device and one Gen3x4-device try to access the CPU by Gen4x4 at the same time. Probably one of them is attached to the second chipset, which is connected to the first by another Gen4x4, so it has to go through both Gen4x4. That is likely causing your "undesired behaviors".
For that reason, X870(E) board use four lanes from the CPU to connect ASM4242.
Many of Asus designs for x670e have the maple ridge controller sharing bandwidth with the first NVMe slot, as noted by even TPU:
www.techpowerup.com/review/asus-rog-crosshair-x670e-hero/3.html
Yeah... USB-IF drove off a cliff called 'consistency and sound decision making' sometime a decade ago and are now a long ago burnt-out wreck at the bottom of said cliff.
I still am sure it doesn't work that way.
If lanes are shared, only one device can be used at time and you have to switch between devices in BIOS. Those IC-switches are not the same as a PLX PCIe-switch that works like an ehternet-switch.
If the JHL8540 isn't connected to the chipset, then there would be 4 lanes Gen4 unaccounted for, which makes no sense and seems especially unlikely when there are two lanes shared beteween one Gen4x2-slot and one Gen4x4 M.2.
Furthermore, I have only ever seen lanes shared between slots and devices of the same PCIe-generation. In above example, lanes are Gen4 on both the x2-slot and the M.2. On my Asus Prime X670E-Pro, two lanes Gen3 of a M.2 Gen3x4 are shared with 2xSATA. On others, Gen3-slots are shared with M.2 Gen3. M.2 shared with the x16-slot are always Gen5, only unshared M.2-slots connected to the CPU are Gen4.
Now, the JHL8540 is Gen3x4, while the M.2_1 is Gen5x4.
I still think it's because the the JHL8540 is attached to chipset, possibly to the second chipset.I agree Inconsistency is a problem. But concerning technical specification, I see very little to criticize.
I'm not a fan of the USB-C plug for every usage, because it's so small and flimsy and some of mine have been been just by the weight of the cable. It has no locking mechanism either.
But ask yourself this: How many devices truly need more than 5Gpbs? How many of those will you attach at the same time to one computer or laptop? How many of those need PD? How many displays will you attach via USB-C to one device?
Honestly, for most customers even 10G is overkill, let alone 40G+. There are very little use cases, in which this speed saves you enough time to warrant the cost. I would not attach any displays via USB-C on my desktop computer, because a displayport cable stays in place much more secure and doesn't bend so easily and I can't really power many displays with the level of PD I get from most USB-C ports with OD (60-100W, most of the time less).
Every laptop should have at least one USB-C with 10G+DP+PD or USB4/TB4, better two, or an additional one with 20G without DP and PD. On Desktop computers, I don't see the point in DP and PD on USB-C and just need like 1x20G front and back and 2x10G on the back. I still need much more USB-A.
The hardware to make all ports have all the features of USB4/TB4 would make boards insanely expensive.
I'm not sure where TPU got their info on the Hero since it's not listed in the manual either.Even if it were, it make no sense for load on it + load on the NVMe that's connected to the CPU to cause issues.
I current have 1 NVMe on the 2nd CPU slot, and 2x NVMes on the chipset (along with 3x SATA devices), as well as my 2x GPUs in x8/x8 from the CPU. Even when driving those components hard, I have no issues.
Now if the 1st nvme slot is used, I get drops in bandwidth and my video output starts to have issues as well. Go figure.