Hi Everyone,

There is apparently some tech tidbits out there about the recently announced Alder gen of Intel desktop CPUs set for release this time 2021.   Two bits of news circulating is the hybrid core composition and a centralized scheduler.  

The scheduler imho could be the most compelling innovation.   This could either mean that the scheduler presents one large processor comprised of all physical cores or takes over all multi threading distribution.  Software would no longer need to be coded to use different cores.

As we've seen with MS FS 2020 even distribution of load across all cores is a challenge.  You spend all this money on big CPUs with many cores and they are not being utilized correctly.

Does all this inefficiency go away with Alder?  :-)

Perhaps it is a bit early to pin hopes on emerging tech but this has my interest as MS FS could potentially see a significant boost in performance.  PCIe 5 and DDR 5 would also be a big help.

I am not a SW dev so my hope in posting this article is to solicit some feedback on the subject.  What do folks think?

Kind Regards  

I don't see how a hardware scheduler, that doesn't have knowledge of the details of all of the threads or what the operating system is doing, could do a better job then the OS scheduler.

Intel hasn't been able to get 10nm working, they are way behind in core count, in fact their upcoming Rocket Lake desktop CPUs are going to top out at 8 cores while the prior generation had 10, because they back-ported 10nm cores to a 14nm process and they are huge and take up too much die space/put out too much heat to fit 10 cores.  That 8 core CPU will have a PL2 power draw under turbo conditions of 225W, while AMD's 16-core 5950X has a max package thermal target under boost of 142W.

Doing a heterogeneous design with 'big' and 'little' cores will allow Intel to claim higher core counts while reigning in their huge power draw and still claiming they have high single threaded performance on the 'big' cores.  It's nothing but marketing BS to get around the fact that they have failed at process development and need a way to claim more cores.  (Ironic considering how they raked AMD over the coals for Bulldozer where some cores shared resources)

I'd assume whatever they are saying about on-CPU scheduling is also a load of marketing BS until proven otherwise.

You'll never have an "even core distribution" because there are only as many threads as the game spawns for the scheduler to move around.  Unless the game is using the same number of threads as there are cores, there will always be cores that aren't being used.  The problem with designing a game that spawns 16 threads, is what do you do on a CPU with only 4 cores? 

Edited December 30, 20205 yr by marsman2020

A single big fat thread cannot execute bits of itself concurrently across multiple cores.  If it did so, it wouldn't be a single big fat thread anyways.  That's what process threads allow a program to do: execute bits of itself concurrently across multiple cores if so available.  It's all in the software design, and no hardware architecture will ever be able to get around that. Unfortunately the only solutions for big fat single thread are either faster single core performance or a better software design.

Edited December 30, 20205 yr by TheFamilyMan

It is questionable how much difference it will make especially on hyperthreaded CPUs but I tend to keep an open mind on these things.

As for the AMD versus Intel nonsense,  it is not really a brand thing per se it is more about monolithic versus chiplet design philosophies.

The TLDR on AMD versus Intel is that Intel with its obsession with monolithic chip design has CPUs with inherently better performance "per core" but AMD with it's chiplet approach has been able to scale up to more cores on smaller dies with the huge advantage that chiplet based CPUs are inherently cheaper as you are binning a single chiplet not the entire CPU.

 

3 hours ago, gunif said:

As we've seen with MS FS 2020 even distribution of load across all cores is a challenge.  You spend all this money on big CPUs with many cores and they are not being utilized correctly.

I beg to differ... why is "even distribution" important, or a "challenge"??

Just go fly the sim 🙂

1 hour ago, Glenn Fitzpatrick said:

Intel with its obsession with monolithic chip design has CPUs with inherently better performance "per core" 

The AMD 5000 series is faster in single threaded applications than the Intel 10x00 series. Intel no longer has "inherently better performance "per core""

2 hours ago, Bert Pieke said:

I beg to differ... why is "even distribution" important, or a "challenge"??

Just go fly the sim 🙂

It depends on software / OS demands.  You could map a pattern but it is like nailing jelly to a tree.

bs

I'm not skeptical of the design as this is how specialized FPGA setups already work (but they do this in a purified form), and is already specifically used in graphics (video processing) in many devices, some you probably own at home!

That said, it won't be ground-breaking, as it is just a tweak overall, because as good as an FPGA parallel HW scheduler is, a translated CPU HW scheduler for an OS will never be as perfect because regular higher level coding principles aren't as easily mapped to a HW scheduler. However, it is the "next step" when it comes to graphics intensive processing, if Intel and NVIDIA can work together more closely, the problem is IP conflicts and that NVIDA and Intel and AMD all have between each other and market pull influence. 

This is not an "all or nothing" type thing, as they will slowly move only certain calculations to a HW scheduler at first, so you cannot think of this in terms of it completely removing OS thread scheduling, but only supplementing, but one day we may see full-blown FPGA-like schedulers in consumer PC's.

CPU's and GPU algorithms in how they work together are actually archaic and way behind because of market forces, because it is possible to take certain intensive calculations and remove them from being CPU/GPU bound and move them to a direct hardware scheduling layer that works like a specialized FPGA just for certain draw calls, but to maximize the functionality, we need to see a change at 4 different levels (base HW, SW, GPU, and OS+compiler). Without that, can only count on partial implementations. 

Also, when you enable Virtualization in your BIOS, some similar things are happening as well with HW scheduling, but it's sort of in the reverse in that it gives the OS more control of the cores and takes away some of the HW schedulers abilities.

FPGA's are truly parallel and do not require software threads in the same way that you are used to thinking about them in code, but they have the advantage that the programming is done directly into the chips using a sort of 'virtual piping / wiring' at a lower level.

This is how they are able to make certain high-end powerful real-time video processing devices in commercial filming equipment and also used in some TV's and projectors, there is FPGA hardware instead of or in addition to a CPU/GPU that has a nearly perfect RTS buffer, so the chip can run much fewer total cycles and produce better response times. I am not sure if smaller stereos and Roku like devices use an FPGA or GPU like designs (never paid attention), but some devices definitely use FPGA's.

 

Edited December 31, 20205 yr by Alpine Scenery

Just to add one point, unfortunately in the end it all comes down to market control.

There are already ways to improve our graphics that take away certain functionality from the GPU's that they shouldn't even be doing anymore. All modern PC's should already have a post-processing FPGA chip in them prior to the signal reaching the final video layer. However, you have the companies that make monitors adding processing to sell more monitors as well as NVIDIA wants to make sure they fully control the GPU stack.

Also if they moved to programmable logic gates in consumer level PC processing (a few exist in bus controllers already and servers already use them), then all the above companies take a sales hit. Because even just putting FPGA's in certain places can allow much more sophisticated HW changes with firmware updating than is currently possibly, but NO HW company wants this to happen, because it's a market war.

Then the problem is you have the video processing companies that make these algorithms (like HDR processing or tone mapping) also including them in TV's as a selling point.

So you essentially have 4 different industries all competing against each other to add post-processing tricks that should actually have been moved to an FPGA like 5-10 years ago. The CPU makers have the advantage in the market control though, because Intel could actually make changes that the HW MFR's are forced to follow which essentially removes some of the silly overload that shouldn't even exist at the GPU level, like certain post-processing controls. 

It's a war though, and Intel knows if they try it, it will start a full blown HW war at all levels like we've never seen before. 

So sadly, like everything else, it comes down to money, and the reason MSFS stutters is because companies don't want to lose market control.

Edited December 31, 20205 yr by Alpine Scenery

7 hours ago, marsman2020 said:

The AMD 5000 series is faster in single threaded applications than the Intel 10x00 series. Intel no longer has "inherently better performance "per core""

The chiplet design is inherently inefficient, but of course they can get faster single threaded performance DESPITE the inherent issues with chiplets if they get them small enough and get the clocks up and optimise the design.

In the end though the majority of CPU sales are enterprise and basically enterprise sales are not based on personal choice or supporting one brand over another they are based on what gets the job done and AMD are making some inroads in enterprise.

Slow progress though. My IT manager is happy to buy AMD for home use where a weekend spent on a driver issue is no big loss but at work where a system down for the day costs tens of thousands he does not even consider it.  There is also issues with software, I just spent $50,000 on a new server for our statistical work (2 Xeon gold, 1,5TB ram, 4 x 1KVA PSU to allow up to 4 A100 tesla cards) and could have got the same performance for $10K cheaper in AMD but the people writing the genetic analysis package we run,  told us it simply will not run on AMD CPUs. They actually recommended Xeon Platinum but the prices on those things are insane.  Of course Centos suddenly pulling the plug has put a real spanner in the works, nothing is ever simple.

 

 

 

Edited December 31, 20205 yr by Glenn Fitzpatrick

1 hour ago, Glenn Fitzpatrick said:

In the end though the majority of CPU sales are enterprise and basically enterprise sales are not based on personal choice or supporting one brand over another they are based on what gets the job done and AMD are making some inroads in enterprise.

In the future (Say 10-20+ years), both Intel and AMD are going to have potential direct competition from Samsung, as they will be planning a mobile like architecture to run a desktop OS that becomes a Windows knock-off, trying to fill in the gaps. 

Samsung has experimented in the past, but desktops still need dedicated hardware for now, but eventually you can power everything with either just your phone or your phone + supplemental box that becomes a PC.

Whether or not they can succeed is another story, but I'd say Samsung is the only with a chance to do it. The reason I don't see any Chinese company doing it, is I think the govt wont allow it to happen. But, Samsung is a S-Korean company, and our govt is friendly with Korea, so it's unlikely they'd try to stop them.

 

Edited December 31, 20205 yr by Alpine Scenery

Hi everyone

I've read every post top to bottom.   You guys know your stuff and each of you have great points.   Stuff could run a lot faster for us if we were not tied to politics and share price.   Who knows what Intel or AMD have up their sleeve.  We at least seem to be the beneficiaries of this perpetual leap frog.

I thank you all for sharing thoughts and ideas.

As stated I am not pinning hopes to anything but wow do I need to upgrade as my i2600k just doesn't keep up any more.  LoL

I will look forward to posting again in future to get your ideas for a dream build.

Regards to all

 

I actually think Intel might have gotten it right with the new 11 series; at least in terms of what we need for MSFS.  (Which seems like 6 to 8 cores, but just as importantly, a significantly faster single core processing speed).

We need to see more evidence, but some released (not leaked, why do they always say leaked?) test results of the 11700 show excellent single core performance, with a gain over AMD Ryzen7 single core performance, and also Intel's previous 10900 series chips.

Things went a bit mad with multiple cores.  Unless you sit there all day running CineBench scores (just to do some whizzer wangling - if you get my point 😁), then all of these extra cores (beyond 6 or 8) are of no real benefit to a typical gamer / flight simmer. 

Some people say they are seeing MSFS use up to six cores; so with one reserved for Windows, and maybe one for i.e. TrackIR (if you know how to set up sharing these cores out), then that would seem to be as much as you need for MSFS at the moment.  So if Intel have sacrificed lots of extra cores to help get single core speed up, then great!

At that stage, increased performance on the main thread seems to be important, and that is what these new 11 series chips should deliver by the look of it, as I think it has been proven that single core speed is still just as important as multiple cores for MSFS, if not more so, due to the 'limited by main thread' issue.

The 11700K seems to show excellent performance at 5.2 GHz (without crazy cooling), with good instruction per core (IPC) gains as well (double figure, so being sceptical I will expect 10-11% 🙂), so the performance of the top end 11900K should be even better. 

So an 11900K, with decent m.2 drive performance for loading, and any of the recent higher end graphics cards (AMD or NVidia), and we should be rocking along nicely!
That is of course if the supply is there, and you are lucky enough to afford the upgrade cost. 😄

Edited January 4, 20215 yr by bobcat999

Hi Bobcat

Great point.   Maybe rocket Lake is worth a look vs alder Lake.    I just read they are throwing atom chips on alder.   Not sure this is going to be pain free.

And they will probably be offered at a discount t once alder is released. 

 

Interesting 

 

Thanks for the perspective. 

I wish cpu manufacturers would make progress in single core speed. We are stuck with 5ghz wall for 10years almost. I want my 10ghz cpu now