Question Alder Lake - Official Thread

[Hulk]

With the release of Alder Lake less than a week away and the "Lakes" thread having turned into a nightmare to navigate I thought it might be a good time to start a discussion thread solely for Alder Lake.

 

[Ed1]

Thread Director and Windows 11 observations with my new rig.

Last night I'm editing some video with Vegas Pro and then rendering using Render+, which is the "Happy Otter" add on for Vegas. It basically allows high quality x264/265 rendering. I notice that when Vegas is "on top" all 20 logical processors are at it but when I move to another apps the Gracemont cores take over. The thing Ian was concerned about.

I think I actually like this behavior. I was watching the UFC fight night on ESPN+ on the computer and also doing some Photoshop work so I had all of the P's at my disposal for Photoshop. The video was rendering just fine in the background and I didn't need it done immediately. Plus it can always be set to batch render.

If I was to fine tune this behavior I would have 4 P's "stay with me" on my current app while the rest stay with background processes.

But this is going to be a subjective thing, which will vary from person-to-person so I think the Thread Director is making the right decision by keeping the P's with the user. Perhaps in the future there could be a setting in Control Panel that allows you to assign P's and E's to foreground/background applications. So if you have 8 P's and 4E's like the 12700K you could select 4 P's as foreground and the rest to background as the OS deems necessary. Or if you really needed the background apps to crank you could select 2 E's for foreground for a little web browing or something?

It sounds like that Vegas Pro and Render+ are using dynamic cpu priority, next time see if priority changes when it goes to the background.

Either that or do you use Processlasso, that can have the same effect, but you can configure it however you want.

Some programs run below normal priority at startup which can sometimes give low cpu % usage, setting priority on these processes to above normal can force high usage.

 

[Accord99]

View attachment 53862

What do you suppose would happen if you select Background Services on your system? Would that make all except one P-core tend to background stuff?

A long time ago, I recall reading that it mainly equalizes the priority of background services so that they have a better chance of being scheduled next to run and increasing the time that a thread can run before being thread switched away and therefore increasing throughput at the cost of responsiveness. But this was in the era of a single core systems, you probably wouldn't even notice anything when systems today have so many cores and SMT.

 

[igor_kavinski]

Question is, would this setting make background rendering faster by assigning more time slices from all cores, or would only E-cores do the background work while the P-cores sit mostly idle waiting to give time slices to foreground tasks as quickly as possible?

 

[JoeRambo]

But this was in the era of a single core systems, you probably wouldn't even notice anything when systems today have so many cores and SMT.

Yeah, it is from era of single core. What it did, was provide the boost to the foreground process and the threads owned by that process. It did so at the cost of "fairness" to the background processes even if they were of same priority class. I don't think i have ever touched it in my career, even on those dual use workstations with Win2000 that also served as DB server for some business app, it was best kept alone.

 

[Leeea]

It seems for gaming only, the thing to do is turn off the e-cores:

Spoiler: hardware unboxed video

 

[coercitiv]

It seems for gaming only, the thing to do is turn off the e-cores:

These results are better than what I feared based on the STR initial testing, as long as the small cores are still being used we're still open for further software improvements and also complete chip utilization if some games end up scaling past 8 cores. (think turn based games where AI turn time matters). Turning off the E-cores is a simple solution for now (just like staying on Win10), but in the future I'd like to use the full potential of Intel's flagship chips.

If Intel is doubling down on E-cores with Raptor Lake then they need to keep improving thread allocation and inter-core latency, otherwise turning off 30% of your flagship compute silicon will be quite hard to swallow. As I said before, I expect to see further tuning from them early 2022, when mobile ADL gets launched. Those 2+8 chips better run like proven racetrack engines, not some early funding demo from a "revolutionary" startup.

 

[Heartbreaker]

It seems for gaming only, the thing to do is turn off the e-cores:

Spoiler: hardware unboxed video

I posted that 2 pages back.

Look again at the "I got it wrong" title in the thumbnail. All you need to do, is don't disable your P cores.

There is only a problem here if you disable P cores, and only in Shadows of the Tomb Raider.

So basically ZERO issue in the real world. Zero need to disable e-cores.

 

[nicalandia]

I agree that's the logical expectation. Are we seeing it in practice?

Sure, Intel performance boost in SMT before it was 21% at best on Skylake based CPUs, with Golden Cove that boost in performance goes up to 30%

 

[Heartbreaker]

Sure, Intel performance boost in SMT before it was 21% at best on Skylake based CPUs, with Golden Cove that boost in performance goes up to 30%

Pretty sure it was "up to" 30% on older Intel CPU like 8700K as well.

 

[Markfw]

Pretty sure it was "up to" 30% on older Intel CPU like 8700K as well.

I seem to remember the ~20% number for most Intel systems for years(preceding alder lake), and AMD was 30%. So its good that the new Intel's as 30% also.

 

[dullard]

I seem to remember the ~20% number for most Intel systems for years(preceding alder lake), and AMD was 30%. So its good that the new Intel's as 30% also.

SMT gains are an indirect measurement of how much of the CPU is sitting idle. It is great to be able to recover that performance for applications that can benefit. But it is also a measure of how overengineered the chips are and not all applications benefit from it. In an ideal world, both Intel and AMD would both be simply 30% faster and not need SMT to recover that lost performance.

When Alder Lake went wider and deeper on many of its components, that meant that there is more excess capacity. More SMT performance is to be expected.

 

[Heartbreaker]

I seem to remember the ~20% number for most Intel systems for years(preceding alder lake), and AMD was 30%. So its good that the new Intel's as 30% also.

Here's a HT/No HT test on the 8700K:

No hyperthreading VS hyperthreading. Testing with V-Ray Next

What is the speed difference when rendering with V-Ray Next on the same CPU with and without hyperthreading. How does hyperthreading work and more inside!

www.aeco.space

the HT provides 29% more performance compared to no HT

More tests: (30% higher for 8700K with HT, 35% higher 7700K):

How Screwed is Intel without Hyper-Threading?

Today we're exploring the impact disabling Hyper-Threading has on Intel processors. We've done this in the past and it's an interesting test, however there are new incentives...

www.techspot.com

 

[dullard]

Here's a HT/No HT test on the 8700K:

One thing to note is that benchmark is about the ideal situation to show off SMT. It will be near the top end of gains. A lot of other software show much more moderate gains. Some rare software has performance losses with SMT.

 

[nicalandia]

It's not a secret that AMD SMT from OG Zen has been extracting more performance from it's wide cores if compared to Intel Skylake based CPUs, this has changed since their new Sony Cove/Cypress Cove wider cores, they are about even now about 30%

 

[Heartbreaker]

One thing to note is that benchmark is about the ideal situation to show off SMT. It will be near the top end of gains. A lot of other software show much more moderate gains. Some rare software has performance losses with SMT.

Well obviously something has to be highly MT to show bigger gains from HT. Techspot tested multiple MT applications, and they got 25%-35%. Which is basically around 30%.

If someone has HT/No HT test for Alder Lake, we can see what it gets and if there is any overlap in the tests.

 

[IntelUser2000]

It's not a secret that AMD SMT from OG Zen has been extracting more performance from it's wide cores if compared to Intel Skylake based CPUs, this has changed since their new Sony Cove/Cypress Cove wider cores, they are about even now about 30%

The two designs don't use exactly the same SMT. The way they split resources are different.

So I suspect Intel wanted to prioritize for single thread performance by going for a more unified approach(So resources will be split into two with HT) while AMD went for a distributed approach because that'll result in better SMT gains. The "unified vs distributed" design philosophy between the Core team and AMD's team is pervasive. AMD went for a more distributed scheduler design and Intel a unified Int/FP scheduler for example.

Sunny Cove gains weren't as good as AMD's. The consensus is that AMD got greater gains and only with Golden Cove Intel changed that.

 

[nicalandia]

Sunny Cove gains weren't as good as AMD's. The consensus is that AMD got greater gains and only with Golden Cove Intel changed that.

To be honest, I was speculating on the Sunny Cove/Cypress Cove part as they are a new uArch as opposed to an evolution of the Skylake uArch

 

[nicalandia]

I suspect Intel wanted to prioritize for single thread performance by going for a more unified approach(So resources will be split into two with HT) while AMD went for a distributed approach because that'll result in better SMT gains. The "unified vs distributed" design philosophy between the Core team and AMD's team is pervasive. AMD went for a more distributed scheduler design and Intel a unified Int/FP scheduler for example.

Can you Elaborate on this Fascinating Subject? As far as I am aware way back on the old Pentium 4 days, Intel spent 5% die area for a maxium 20% performance uplift, but I am not sure if this has been remain the same all the way to Alder lake.





Has AMD disclose how much of the Die Size they are using for SMT?

 

[TheELF]

One thing to note is that benchmark is about the ideal situation to show off SMT. It will be near the top end of gains. A lot of other software show much more moderate gains. Some rare software has performance losses with SMT.

No it's not, just as you yourself mentioned "SMT gains are an indirect measurement of how much of the CPU is sitting idle. " software that uses the least amount of CPU resources would be the most ideal situation, things like games or general windows usage where threads are just made to do what they need instead of software that does all the compute it can get it's hands on.
In games HTT can give as much performance as a real core, but of course if you already have enough real cores to max out a game then HTT won't do anything.

 

[IntelUser2000]

Has AMD disclose how much of the Die Size they are using for SMT?

You can see Zen presentations. The amount of die Intel/AMD is using for SMT is still trivial. SMT is not the easiest to implement or validate but it sure is pretty power efficient and uses next to no transistors.

 

[igor_kavinski]

but it sure is pretty power efficient and uses next to no transistors.

Not sure about that, at least when it comes to i7-1165G7. It was 65 degrees without HT but with HT, it was hitting 95 degrees in the TPU ThrottleStop stress test.

 

[jpiniero]

Intel i7-12650H 10-core "Alder Lake-P" CPU to feature fewer cores than i5-12500H (12-core) - VideoCardz.com

Core i7-12650H with fewer cores than i5-12500, but at least they are bigger Intel 12th Gen Core “Alder Lake-P” lineup is slowly becoming confusing. The hybrid architecture mixing Golden Cove (Performance/Big) cores with Gracemont (Atom/Small) is supposed to increase the power efficiency of...

videocardz.com

Looks like Intel is going to release a i7 H with 6 big and 4 small, even though the top i5 might have 4 big and 8 small. Also the EU count is cut to 64.

 

[Asterox]

All sorts of problems with socket 1700.

Cooling issues with Intel’s Alder Lake - Problems with the LGA-1700 socket and a possible workaround | Page 3 | igor´sLAB

Badly fitting coolers, too low performance with actually good waterblocks and bent CPUs – meanwhile not only the feedback of the readers is piling up, but also very similar reports from the board…

www.igorslab.de

 

[R81Z3N1]

Wow, just wonder what cost this will add to folks build. Now that you have to consider a good socket, as well as a good cooler. Not to mention the Win 11 issues with threads and NVME drives.

 

[DrMrLordX]

Has anyone here experienced bent sockets on their Z690 boards yet?