Intel's Arrow Lake CPUs throttle PCIe 5.0 SSD speeds, tests reveal

Alfonso Maruccia

Posts: 1,800   +542
Staff
Editor's take: It's no secret that Intel is in a tight spot. The company that invented the microprocessor doesn't seem to understand how to make great products anymore, and users are suffering significant performance issues as a consequence.

Intel changed its approach to silicon manufacturing with Meteor Lake CPUs, moving from a monolithic approach to a disaggregated, tile-based design. However, the significant technology switch is clearly providing a headache or two; reviewers and power users can't stop discovering new performance issues affecting Chipzilla's latest Core Ultra processors.

The Arrow Lake CPUs, which Intel introduced on October 24, 2024, under the Core Ultra Series 2 brand, proved to be slower than Raptor Lake processors (Core i9-14900K). They are also much slower than the 9800X3D, which is why AMD sales are soaring right now. But wait, there's more: Intel Core Ultra 200 CPUs are even slower when it comes to driving the fastest SSD units available on the market.

The SSD Review discovered the issue while testing PCIe 5.0 SSDs. These drives should theoretically be able to reach a 14GB/s data rate, but the actual benchmark results were falling short of that target. The storage-focused website experienced the problem with the Samsung 9100 Pro and Micron 4600 SSDs, pairing the two drives with both high-end Raptor Lake (Z790) and Arrow Lake-compatible motherboards (Z890).

The Z790 board was able to achieve speeds of 14.3GB/s with both SSDs, while the same storage units weren't able to exceed 12.3GB/s when paired with Arrow Lake CPUs and motherboards. The reviewer also tested the drives through a PCIe add-in storage card (Asus Hyper M.2), discovering that the issue mostly affects the bandwidth available through the CPU's PCIe lanes. The expansion card provided higher sequential I/O rates, while random-performance figures were still lacking compared to Z790 motherboards.

The issue isn't limited to specific motherboard brands. Asus and ASRock confirmed and replicated the performance downgrade in their labs, and they also explained why the issue is actually happening: Core Ultra 200 CPUs experience higher latency to the I/O tile that feeds the processors' Gen5 PCIe lanes.

Intel also confirmed The SSD Review's discovery. The company said that PCIe lanes 21 to 24 feeding the Gen5 port "may" exhibit increased latency compared to PCIe lanes 1 to 16 because the Arrow Lake design uses a longer die-to-die data path. Intel has tried to improve gaming performance on its chiplet-style CPUs with a few firmware (microcode) updates. But we don't see how the Santa Clara corporation will be able to fix the newly discovered issue with fast storage drives without a significant readjustment of the CPU design.

Permalink to story:

 
Not quite a correction, but there is debate to be had on the invention of the first microprocessor::
The first to be argued to be invented was by Boysel of Autonetics in the form of the D200 in 1967. The first mass produced one was the Garrett AiResearch MP944 made in 1970 that wasn't declassified until 1998 due to it being central to the F-14. Viatron Computer Systems coined the term Microprocessor. What Intel was first to was the single chip microprocessor and mass market adoption.
 
This actually is a worse blow than many think... while the actual SSD performance will largely be irrelevant to most users, the Arrow Lake CPUs were SUPPOSED to hold their own in productivity - while they failed at gaming...

The only users who actually require the top performance from Gen 5 SSDs will be productivity users. Sorry Intel, no luck for you this generation...
 
You want to throttle fine just don't hide that fact from the consumer that thinks his/her devices are running at full speed 100% of the time. What is the point of buying a more expensive PCIe 5.0 device when it will function closer to a PCIe 3.0 or 4.0 device most of the time.
 
Intel's chiplet era feels like watching someone try to reinvent a car engine by gluing parts from five different engines together and wondering why it's stalling at every red light. Great in theory, but if your PCIe lanes are bottlenecking flagship SSDs, you've missed the "performance" part of high performance.

Meanwhile, AMD’s just sitting back watching Intel trip over their own architecture like it’s a reality show.
 
Intel's chiplet era feels like watching someone try to reinvent a car engine by gluing parts from five different engines together and wondering why it's stalling at every red light. Great in theory, but if your PCIe lanes are bottlenecking flagship SSDs, you've missed the "performance" part of high performance.

Meanwhile, AMD’s just sitting back watching Intel trip over their own architecture like it’s a reality show.
Switching to a new design always involves a few hiccups… we saw that with AMDs initial Zen offerings… but Intel is late to the party… it took AMD a few years to recover and now they’re comfortably on top. I’m not sure Intel has the time to do the same… but… here’s hoping, as competition is a good thing.

Right now, AMD can charge virtually whatever they want for the high end (including HEDT), and Intel can’t do a thing…

I remember buying the Intel 5960x with an “insane” 8 cores… now the 7980x (which given the 10 year gap is the closest equivalent), costs more than 5 times the amount…
 
Only a (super) Nova Lake can save Intel on desktop.

I will stick to Zen 5 for my next update and see if Nova Lake is still a fluster cluck or AMD knocks it out of the park with Zen 6 which is already confirmed as having 12 core ccd's.
 
This specific issue if taken in isolation wouldn't affect many users.
But it can't be taken in isolation as it indicates poor decisions with the overall design of their latest chips.
Also productivity users will (probably) be the only ones to notice. It would be a let down. Luckily the news is out.

Putting it all together, I mean all things to do with CPU performance including gaming, high productivity requirements etc. It's clear, that Intel really have lost the plot. AMD is making a mockery of them in more ways than just gaming.

It's tempting to feel something good about Intels embarrassment, but in reality this is not a good thing at all.

If it continues like this then AMD will dominate more and more. Not saying they don't deserve to with there excellent recent releases, vs. Intels loss of an obvious path forward. Intel can't blame AMD.

My main point here is obvious:
Most people will recognize that that competition is critical in pretty much everything.
No need to explain what happens when one large company has dominance over a particular sector.

Intel still have name recognition and anyway we are not there yet. (some may disagree, and I wouldn't argue the point.) But:

I just hope Intel get there act together in the next 2 years or so. They have a lot of work to do, and unless something new and amazing is found they won't catch up in a few months.
 
Last edited:
This specific issue if taken in isolation wouldn't affect many users.
But it can't be taken in isolation as it indicates poor decisions with the overall design of their latest chips.
Also productivity users will (probably) be the only ones to notice. It would be a let down. Luckily the news is out.

Putting it all together, I mean all things to do with CPU performance including gaming, high productivity requirements etc. It's clear, that Intel really have lost the plot. AMD is making a mockery of them in more ways than just gaming.

It's tempting to feel something good about Intels embarrassment, but in reality this is not a good thing at all.

If it continues like this then AMD will dominate more and more. Not saying they don't deserve to with there excellent recent releases, vs. Intels loss of an obvious path forward. Intel can't blame AMD.

My main point here is obvious:
Most people will recognize that that competition is critical in pretty much everything.
No need to explain what happens when one large company has dominance over a particular sector.

Intel still have name recognition and anyway we are not there yet. (some may disagree, and I wouldn't argue the point.) But:

I just hope Intel get there act together in the next 2 years or so. They have a lot of work to do, and unless something new and amazing is found they won't catch up in a few months.

Intels problem goes all the way back to it's decision to skip EUV at 10nm, not being able to do it, and getting stuck in a rut hoping that once it *finally* fixed it's node issues they could resume to get top performance through die shrinks alone. Meanwhile, AMD was forced to iterate (multiple times) due to it's node disadvantage versus Intel, and eventually ended up with a far superior design.

Lets face it: Intels overall core architecture is ancient at this point, and Intel is only now starting the process of coming up with something new. Ideally, Arrow Lake will be Intels Zen1, and they'll eventually iterate on the design and come up with something competitive.
 
Intels problem goes all the way back to it's decision to skip EUV at 10nm, not being able to do it, and getting stuck in a rut hoping that once it *finally* fixed it's node issues they could resume to get top performance through die shrinks alone.
IMO, that's only part of Intel's problem. Over the years, they've tried to pull some under-handed stuff and were caught doing so. That "stuff" they pulled was an attempt to make up for an inferior design with smoke and mirrors.

Even if Intel does manage to eek out a better design, if they keep pulling utter BS, that won't bode well for them, IMO.
 
I fully agree with gamek2, and Wiyosaya's posts, above. Mine too, more generalized.

Reason for posting is, just wow, there are so many things they have screwed up. I forgot about the "under-handed," stuff. What they were caught for.

It adds up to a right Royal FUBAR. They only have themselves to blame in almost all of this.
Can't think of anything that wasn't there fault though.

I do hope they get their act together though. No under-handed crap, a complete re-think. We will have to wait and see, but they have lost something valuable. Consumer confidence in their products.
 
Eh, if it was only a reduction from 14GBps to 12GBps max over PCIe that caused a discount on these it might be tempting, except for a) you know there won't be any discount because the normal rules of market forces broke a long time ago and b) there are still all the other things such as the competition existing and the platform being a dead end.
 
Back