Wednesday, January 3rd 2024
Intel Meteor Lake P-cores Show IPC Regression Over Raptor Lake?
Intel Core Ultra "Meteor Lake" mobile processor may be the the company's most efficient, but isn't a generation ahead of the 13th Gen Core "Raptor Lake" mobile processors in terms of performance. This isn't just because it has an overall lower CPU core count in its H-segment of SKUs, but also because its performance cores (P-cores) actually post a generational reduction in IPC, as David Huang in his blog testing contemporary mobile processors found out, through a series of single-threaded benchmarks. Huang did a SPECint 2017 performance comparison of Intel's Core Ultra 7 155H, and Core i7-13700H "Raptor Lake," with AMD Ryzen 7 7840HS, 7840H "Phoenix, Zen 4," and Apple M3 Pro and M2 Pro.
In his testing, the 155H, an H-segment processor, was found roughly matching the "Zen 4" based 7840U and 7840HS; while the Core i7-13700H was ahead of the three. Apple's M2 Pro and M3 Pro are a league ahead of all the other chips in terms of IPC. To determine IPC, Huang tested all processors with only one core, and their default clock speeds, and divided SPECint 2017 scores upon average clock speed of the loaded core logged during the course of the benchmark. Its worth noting here that the i7-13000H notebook was using dual-channel (4 sub-channel) DDR5 memory, while the Core Ultra 7 155H notebook was using LPDDR5, however Huang remarks that this shouldn't affect his conclusion that there has been an IPC regression between "Raptor Lake" and "Meteor Lake."
Sources:
David Huang's Blog, Tom's Hardware
In his testing, the 155H, an H-segment processor, was found roughly matching the "Zen 4" based 7840U and 7840HS; while the Core i7-13700H was ahead of the three. Apple's M2 Pro and M3 Pro are a league ahead of all the other chips in terms of IPC. To determine IPC, Huang tested all processors with only one core, and their default clock speeds, and divided SPECint 2017 scores upon average clock speed of the loaded core logged during the course of the benchmark. Its worth noting here that the i7-13000H notebook was using dual-channel (4 sub-channel) DDR5 memory, while the Core Ultra 7 155H notebook was using LPDDR5, however Huang remarks that this shouldn't affect his conclusion that there has been an IPC regression between "Raptor Lake" and "Meteor Lake."
84 Comments on Intel Meteor Lake P-cores Show IPC Regression Over Raptor Lake?
However, I'd really like a work laptop with one of these in it. I think that corporate laptop thing is really its niche, seems perfect for that and for students. A bit too high end in price IMO for say, grandma to pay her bills with though.
EDIT: Oh BTW, AMD Ryzen ultra mobile CPU's also have a lower IPC than the desktop variants... Just throwing it out there.
Now let's see some performance per watt specs.
Now that I see this table, I get the impression that the die shrink microarchitectures were supposed to be trees and the new microarchitectures were supposed to be yellow. Also if Arrow Lake really comes out on IFS 20A then it will be the first line since Ice Lake to be on its originally intended node. It's just a few years later than once planned.
*Technically, what I'm calling IFS 7 here, Intel broke into 4ish names, 10nm, 10nm+ (renamed 10nm), 10nm++ (renamed 10SF), and 10ESF (renamed Intel 7). For simplicity I'm calling all of them IFS 7 here. Most of that table is basically a fairy tale anyway.
It seems like Intel is slowly coming to grips with the E-cores taking the fore front role and P cores having more of a back seat role. It's still going to take a bit of transitioning time, but most tasks are pretty multi-thread aware today and systems perform a lot more leisurely multi-tasking these days versus systems a decade ago.
The most notable thing to me is the graphics improvement and the SoC island looks like a promising addition. I think if you're on the fence though wait on the follow up generation because I'm sure they will iron out improvement upon it.
I'd like to see what Intel could do with a modern unlocked dual core and respectable binning. It would also be neat in that scenario to see them insert a larger iGPU and make it designed more as a buget minded HTPC/APU. It would be pretty great at it and low power.
If that happens, that would be an even worse launch than first gen P4
I wonder what is the real reason for this? One hypothesis could be that these P cores were a result of two development projects, one going really well, above expectations (Alder lake), and one having some problems (Meteor lake). That combined with possible delays of Meteor lake resulted in the lower performance in comparison with Alder/Raptor lake P core.
Or is something in Meteor lake P core simply broken and they could not make it to work in time?
Given that a uArch takes about 5 years start to finish, it should be no surprise that shifting to tiled design introduced a small performance penalty. This is Raptor Lake adapted to a tiled package and a die shrink.
There are four wins for Intel here. The new node, improved iGPU, inclusion of NPU, and the shift to tiled packaging. The user sees benefit from the iGPU, NPU, more cores, and the uArch of the E-cores is improved - so much improved multi-core performance.
Clearly, not so much benefit in single thread since the P-cores are likely raptor lake based.
People looking for big single thread bumps are looking for Arrow Lake.
The 'original' plan from the earlier post (no Raptor Lake here) :
- new node brought compute regression in IPC. This is disappointing. It's not surprising that Intel cancelled desktop Meteor Lake. In 370 tests, Phoronix found that Phoenix 7840U is on average 28% faster, which is hugely embarrassing for Intel. Let's see whether any BIOS update brings less humiliation in compute workloads. Meteor Lake 155H is actually shockingly slow considering node shrink. It is what it is...
www.phoronix.com/review/intel-core-ultra-7-155h-linux
- NPU: Intel has never shared performance numbers of this AI engine. At best, it might perform as last year's XDNA, around 10 TOPS. In a few weeks, XDNA2 on Hawk Point will bring 16 TOPS
- iGPU is 8% faster on average in compute workloads (Phoronix big testing) than RDNA3 on 7840U. This is 1024 sp. (VEs) on 155H vs 768 sp. on Phoenix.
So, iGPU has 33% more silicon, but only 8% higher performance. Good catch-up, but not that impressive at all, which will be visible once Strix come out with 16CUs and 1024 sp.
- gaming is not clear. We need to see comprehensive, like-to-like tests in dozens of titles, with the same RAM, preferably on similar laptops.If you are a taxpayer, would you approve of this behaviour and then have Meteor Lake 155H delivered on your lap being over 20% slower in CPU compute than 7840U?What would you say if any other company delivered a brand new CPU with IPC regression on a new node?
Notebookcheck has it about 15-20% faster overall in their performance tests vs AMD 7840U.
CPU performance ratings :
The higher end AMD 7840U (15W) = 69.2
155H (28W) = 81.7
The 45W range 7840HS = 83.6
The 45W range 13700H = 84.8
The 155H isn't the top end ML, 165H is. Quite likely a 165H would match or beat the 45W parts above if they tested it.
Yes it's not a 15W chip, but neither is it a 45W chip. But it performs *very* close to those 45W range chips.
Ref:
www.notebookcheck.net/MSI-Prestige-16-B1MG-laptop-review-From-Core-i7-Xe-to-Core-Ultra-7-Arc.785587.0.html
www.notebookcheck.net/Lenovo-Yoga-Pro-7-14-G8-laptop-review-AMD-Zen4-isn-t-automatically-better.787943.0.html
On the other hand, what you wrote certainly does not sound good. Taking two steps forward and one step back is worse than taking three steps forward, that is for sure.
BTW Intel delivered very good Alder lake CPUs with IPC significantly better than AMD 5000 CPUs in 2021. I guess the next significant improvement comes with Arrow lake late this year.
Is having nothing new really exciting for three years in this very complex industry I described above a disaster, or something that simply can happen? At least in desktop they kept adding cores and increasing frequencies, improving performace (at the cost of energy efficiency) in this period.