Alderlake + Windows 10 stability with SolidWorks

Hi, I just wanted to ask if anyone could tell me from experience how stable Alderlake is on Windows 10? I’m almost due for an upgrade (currently use an i7-4790/Quadro M2000), however a couple of programs I use aren’t supported on Windows 11 (SolidWorks). While my current system isn’t fast it is incredibly stable - which is of utmost importance to me.

If there are concerns with Alderlake/Windows 10, I’ll likely build a 10850k based system. It’s a tough pill to swallow though, as a 12700k is currently a few bucks cheaper, faster and will give be supported/relevant for longer.
Thanks for your thoughts!

Hey welcome to the forum!

I used a 12700k for a week before returning it due to other reasons with an RTX A6000 but ran Civil3D and Infraworks on the 12700k and it ran like a charm on Windows 11 Enterprise Vs my old 10900 system, would honestly grab the 12th gen atm due to it being faster and well cheaper.

There is a YT channel who is a AD Inventor Expert and has had no issues at all, albeit running windows 11.

Thanks, appreciate the info!

That’s a nice GPU, I have my eye on an RTX A4000. Did the A6000 run well?

Yup the A6000 was fantastic, but sold it since hardly used it since I use my A4000 a lot more, and ordered a Mac studio with the money I got from the sale, as I want to try Mac out for CFD and other things.

“Not supported on 11” usually means “we havent gotten around to testing that yet, it should work fine but no promises”.

If you are worried, just install 11 and double check all apps you are worried about for a couple of weeks.

Experience-wise i have no stake in this thread, but I’m wondering why all the replies are telling the OP to use Windows 11 when he specified that his intended program, Solidworks, does not support Windows 11? Solidworks website as of Dec 2021 even specifically says don’t use it.

" Windows 11 is not currently supported or recommended for use with SOLIDWORKS . We are expecting support from a later service pack of SOLIDWORKS 2022 but we do not have firm details at this time, we advise you to hold off from upgrading to Windows 11 for now."

This may seem a bit salty of a reply, but this is really not helpful to anyone current or future reading of this. Granted @F1N gave some competing software options that work on Windows 11, the OP wants to know if anyone has attempted Alderlake on Win10 with Solidworks.

2 Likes

That changes the layout of the land considerably, yes. I recommend to go Windows 11 in most cases because there really is no reason to hold off. Of course some exceptions exist, like this one, but for 99% of the cases it is fine.

Reason I recommend it is because you will need to upgrade to Windows 11 before 2026 regardless (if you still run Windows by then, which is a 90%+ probability despite whatever you say about Linux). Better to take the hit now than later.

Pretty much a worst case performance dip of around 5%, should be no difference otherwise, so Alder Lake on Windows 10 should be fine. :slightly_smiling_face:

True, but Solidworks does say an update is coming this year for Windows 11. So until then, still need an answer on the OP’s question, he’s got plenty of time :slight_smile:

2 Likes

The answer is pretty much, yes you can run Windows 10 with no issues on Alder Lake, albeit a 5% performance hit. So buying Alder Lake now while running Windows 10 and waiting for Windows 11 Solidworks fixes, and then upgrade once the fixes are complete should be viable.

I have no idea whether Solidworks specifically will work with Windows 10 on Alder Lake though, but I see no reason why it shouldn’t, unless Dassault Systemes explicitly warns against it.

Were there any indications that Alderlake was producing actual errors, other than possible bad scheduling? I can’t seem to recall any reports of that so I am not seeing why buying Alderlake would be bad. Until Solidworks gets an Windows 11 update, just use Windows 10 and live with less than optimal performance. In my opinion a better decision than putting money in an older CPU.

2 Likes

I think that by now everyone knows that 12th gen Intel chips - Alder Lake - is a heterogenous compute environment with two different types of cores: hyperthreaded performance cores that can run at higher clock speeds and more parsimonious efficiency cores which are not hyperthreaded. The different levels of processor contain different numbers and combinations of P and E cores.

Intel worked closely with Microsoft on the “Thread Director” (aka Hardware Feedback Interface in Intel literature) which allows the kernel to know in real time how the cores are performing with regard to efficiency and execution (performance). The Windows 10 kernel does not have this functionality (I’ve seen reports that it may get back-ported in 2H but don’t know for certain). The kernel needs this functionality in order to run smoothly, making sure the right cores are working on the appropriate work load. It’s coming to the Linux kernel in the next release (5.18) but older Linux kernels will have the same inefficiency as Windows 10 because the kernel doesn’t understand that cores are now not all the same.

There are reported issues with Alder Lake systems running smoothly with Windows 10. The issues are usually solved with a move to Windows 11. However, we’re in the first phase of a new architecture and the hardware and software, from kernels to apps - games can have significant issues depending on whether they assume a heterogenous compute environment too - are not completely aligned and won’t be until everyone is on the same page regarding the new Intel processor architecture.

This is something I’ve struggled with - stay with homogenous compute environment, i.e. Ryzen (/Threadripper) or older Intel, or roll the dice on Alder Lake and the upcoming releases of Intel processors and PCH chipsets - Raptor Lake, Meteor Lake, Arrow Lake, Lunar Lake, etc.

Intel are going all in on heterogenous cores per their road map, so these issues - just teething problems hopefully - will be solved. I certainly hope so because I just built an Alder Lake system and I’m intending to run Linux which until the next kernel release has really preferred symmetric cores.

Anyone running Windows on an Alder Lake system needs to be on Windows 11. Software vendors are likely to take some time to ensure that their products work seamlessly on Alder Lake. It’s a little bit shocking that almost six months into Alder Lake we have these issues. It’s especially surprising to me that Intel apparently didn’t work as closely with the Linux community on getting Thread Director/HFI into the kernel by Alder Lake release day last November, even more surprising when Intel’s own variant of Linux outperforms the official builds on Alder Lake (those results can be found on Phoronix (phoronix.com))

1 Like

Thanks everyone for your input.

As mentioned, stability is important to me and I’d prefer not to be the one figuring it out. Especially if I discover issues outside of the return period of the hardware I purchase. In addition to SolidWorks, I also use Lightspeed trader where stability is more important.

So, this is all why I’m considering 10th gen as it’s a known quantity, but it doesn’t feel great buying an older generation (for the same price as a current generation) when I’m trying to get the most from my money in terms of longevity. If 10th gen were significantly cheaper the choice would be easier.

I will eventually upgrade to Win 11, but I see no reason to be an early-ish adopter when Win 10 is still supported (from a security standpoint) and most programs are still written for it. From what I’ve read, Win 11 still needs to be sorted out.

The ~5% performance difference isn’t a factor in my decision, both processors are much faster than my 4790.

@AliasInWendellLand: if you could elaborate on the issues you mentioned I’d appreciate it - it sounds like you’re talking about something other than just a drop in performance?

2 Likes

Here’s a good illustration of the issue on Windows 10: Poor Alder Lake performance when building Firefox due to bad scheduling

The issues with correct scheduling has been described in various Level1Techs, Gamers Nexus and Hardware Unboxed videos amongst others, but since they’re usually embedded in a collection of other issues/reports I don’t have a signal example at my fingertips, sorry (a search in those channels should return something to a similar effect of the above link).

The Windows kernel, and the Linux kernel for that matter, are designed for a Symmetric Multi Processor environment, back when there was only one core per processor, but you could have a load of processors in a departmental server, e.g. Until Alder Lake that now venerable design has worked perfectly, first with single-core multi-processor architectures, and then multi-core single-processor architectures to which there was a seamless migration thanks to the foresight and brilliance of hardware and software engineers at the time.

Personally I’m still not convinced that the hybrid Alder Lake model is the right choice, but Intel are going for it full-speed, and ARM have been there for a while - they invented big.LITTLE after all.

“We’re not in SMP Kansas anymore, Toto.”

1 Like

I’ve heard general commentary about scheduling issues from the channels you mentioned, but it didn’t sound as concerning as the information in the link you provided - thanks for sharing it. It does sound like disabling the e-cores is a valid work around, which I’m guessing would not drop performance to below 10th gen levels.

While I haven’t made a decision, I’m better informed so that’s a win. Here’s to hoping a 10850k or 11900k goes on a steep sale soon.

3 Likes

If stability is utmost importance, then I would go for a machine that supports ECC, rather than a brand-new platform that has reports of issues running with DIMMs in all slots :wink:

I’m a solidworks user and assuming what you care about most is modeling performance and not Sim or PV, single core performance is what matters. Disabling the E cores and sticking to Windows 10 is what I am going to do for my next solidworks machine… I’m just waiting for the W680 boards to come out.

1 Like

From what I’ve read, it appears to be an issue with some kits only when running XMP. 2x16GB will suit my use case with headroom.

ECC briefly entered my mind, but I don’t have enough knowledge about server level hardware and systems, not to mention the additional cost. My 4790 / M2000 / H97 system has served me well so I’m comfortable staying with consumer grade hardware. If you’re hinting at AMD Ryzen w/ECC, I’m not convinced AMD is stable enough for me. While SolidWorks may be written with both AMD and Intel in mind, Lightspeed trader is more likely to lean towards Intel.

@twin_savage - thanks for the input. Yes, mostly modeling performance but there’ll be some simulation mixed in there. Even with e-cores disabled, I’m sure the performance will be more than adequate. W680 looks intriguing, I’m sure you’ll have a great machine on your hands.

My take is in the worst case scenario you disable E cores in the BIOS and then you have essentially an 11th gen equivalent, albeit one that is more performant (performance cores gave a ~30% uplift in some single-threaded tests over the previous gen).

I’m hoping and expecting kernel devs, in concert with Intel engineers, to ensure that this new world of Asymmetric Multi Processing works as well or better than the previous SMP architecture.

Of course, if Intel manages to get their lithography down to AMD levels and maybe reduce the number of transistors on a chip then that will mitigate the power consumption problem that has resulted in P and E cores, and then we go back to familiar and effective SMP, as the kernel prefers.

Intel are going all-on on P & E core asymmetry, but really it’s a work-around for their relatively awful inefficiency. So it may only be a stop-gap measure. It’s a little desperate I think. Not sure what Linus (Torvalds) has to say on the subject.

1 Like

On second thoughts we’re probably saddled with P & E cores forever, even if Intel does reduce its lithographic dimensions thereby commensurately reducing power consumption.

ARM have had the big.LITTLE hybrid/heterogenous compute environment for some time and that’s not changing.

The cost of power is increasing as a result of climate and political (war!) considerations.

The increase in processor productivity isn’t coming from more transistors on a chip as we approach the physical limits of semiconductors - atomic-scale transistors! - but by increased parallelization and core count.

There’s no reason to not have “Efficiency” cores that consume fewer Watts as the number of processors in the world explodes.

So, my earlier comment about going back to “old-fashioned” SMP is obviously wrong and not correct for the world in which we find ourselves in the 21st century. SMP was great for the end of the 20th century when available power and global heating were not issues, and when even the most powerful departmental machine running a Windows NT kernel had 32 or fewer processors. Now we have workstations with 64 cores and 128 threads, and that’s about to double.

Presumably AMD will be joining the hybrid/AMP bandwagon soon.

1 Like

This is my opinion, not fact, but the way I see it big.little architecture doesn’t belong in x86 because of the vast depth of legacy software that x86 can reach into and is expected to run without scheduler issues; for ARM arch, this isn’t as much the case, which is why it makes much more sense there.

My feeling is that the P+E core designs are going to be like the integrated graphics intel has been embedding in there CPUs for the past ~10 years, something that originated in purpose on the mobile side of the house and took desktops along for the ride.

Notice on intel’s roadmaps that they don’t have any heterogeneous server chips in the pipeline, all the xeons are going to be pure P cores (not counting the “mobile” xeons or the socket 1700 xeons since they are derived from a consumer product).
Now down the road (>2 years) intel is supposed to make a pure E core xeon chip with sierra forest but I expect it will not be well received or perhaps it will be a very niche sku.

If power consumption was the true goal (solely speaking for desktops) then some kind of strict power policy enforcement could be orders of magnitude more effective than bothering with creating a different type of core.
That being said, I can see why P+E cores are attractive for mobile.

1 Like