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.

 

[dullard]

Official spec says P is 20-64 W. An OEM could just use H instead and cut the PL1 and PL2 to their needs I guess if they also wanted the 8 lanes.

I was mostly laughing at your mention of the U chips. Had you limited your comment to just P chips, then I wouldn't have responded.

 

[Abwx]

First slide uses SPECrate n-copy, second slide uses Blender, different workloads differnt scaling hence the "upto".

OK, so the curves essentially display the RAM bandwith in respect of CPU power, hence the 5800U curve being at about constant value, well , that s quite a desperate "benchmark"...

We can see in Blender that the scores are about equal and likely that the 6800U will be somewhat better than the 1280P.

Edit : There s no 8C TGL in those comparisons, ADL would be better in the Spec curves but wouldnt be far from the 5800U in Blender and quite an embarassement as a generational comparison.

 

[Hulk]

ADL ~15% better than Tiger at 15W. About 30% better at 28W.
Intel's graphs are wonderfully cryptic though, so it probably has some strange nonlinear scaling!

 

[Dayman1225]

hence the 5800U curve being at about constant value, well , that s quite a desperate "benchmark"...

Intels reasoning for not having a curve for 5800u and M1 was in the Tomshardware article

Intel shows Alder Lake and Tiger Lake performance across a wide range of power envelopes but says it couldn't alter the power range of competing chips. As such, it only provides one measurement for the AMD and Apple models. Additionally, Intel did not have any Ryzen 6000 models for comparison as they aren't available yet.

 

[itsmydamnation]

Intels reasoning for not having a curve for 5800u and M1 was in the Tomshardware article

thats a load of crap......

 

[IntelUser2000]

ADL ~15% better than Tiger at 15W. About 30% better at 28W.
Intel's graphs are wonderfully cryptic though, so it probably has some strange nonlinear scaling!

At 15W TGL is at 0.75 and ADL is at 0.9, so that's 20%.

There's nothing for the U at 28W, so at 25W it's 0.95 vs ~1.23 which is slightly under 30%.

The P is 70% faster at 28W.

Compared to 5800U, it simply doesn't have enough cores for the U, as 2+8 is roughly ~6.

Looks like the P can reach 2x, but only if the power is ramped higher to 35W. Sigh marketing.

The gains are a bit disappointing. Somebody was talking about the System Agent in -H reaching 6W so I checked Ultrabook's reviews out how Tigerlake-H fares.

Something is messed up with Alderlake's System Agent. You can see the Tigerlake-H's SA can easily use only 1W even when under load.

But for Alderlake, it's staying at 4-6W most of the time. 5W at 30W and 4W at 18W. That's terrible.

If this largely carried onto the U and P, it would explain the lower than expected gains.

 

[Hulk]

At 15W TGL is at 0.75 and ADL is at 0.9, so that's 20%.

There's nothing for the U at 28W, so at 25W it's 0.95 vs ~1.23 which is slightly under 30%.

The P is 70% faster at 28W.

Compared to 5800U, it simply doesn't have enough cores for the U, as 2+8 is roughly ~6.

Looks like the P can reach 2x, but only if the power is ramped higher to 35W. Sigh marketing.

The gains are a bit disappointing. Somebody was talking about the System Agent in -H reaching 6W so I checked Ultrabook's reviews out how Tigerlake-H fares.

Something is messed up with Alderlake's System Agent. You can see the Tigerlake-H's SA can easily use only 1W even when under load.

But for Alderlake, it's staying at 4-6W most of the time. 5W at 30W and 4W at 18W. That's terrible.

If this largely carried onto the U and P, it would explain the lower than expected gains.

I stand corrected. After making the graph overlay I was too exhausted to correct the percent values.

 

[IntelUser2000]

I stand corrected. After making the graph overlay I was too exhausted to correct the percent values.

That's ok, it happens.

Look at the System Agent power for Alderlake-H. It's 3-4x the one in Tigerlake-H.

So there's something big going on there. That kind of power increase in the SA would seriously rob the power going into the CPU at say, 15W. 5W at 30W. That means if we had a 1.5W SA like in Tigerlake, the chip power would have been 26W. That's substantial!

2W at 5.6W chip power WTF! 3.6W at 16W! For Tigerlake-H, it generally stays at 1-2W even with 45W package power. And lot of systems it can go to 0.7-1.2W

I can see Raptorlake being a decent jump just by Intel engineers fixing the issue to get the power numbers for the SA back in line with the Tigerlake generation.

 

[Hulk]

That's ok, it happens.

Look at the System Agent power for Alderlake-H. It's 3-4x the one in Tigerlake-H.

So there's something big going on there. That kind of power increase in the SA would seriously rob the power going into the CPU at say, 15W. 5W at 30W. That means if we had a 1.5W SA like in Tigerlake, the chip power would have been 26W. That's substantial!

2W at 5.6W chip power WTF! 3.6W at 16W! For Tigerlake-H, it generally stays at 1-2W even with 45W package power. And lot of systems it can go to 0.7-1.2W

I can see Raptorlake being a decent jump just by Intel engineers fixing the issue to get the power numbers for the SA back in line with the Tigerlake generation.

Could the SA issue have anything to do with the P's and E's being on the same voltage?

 

[IntelUser2000]

Could the SA issue have anything to do with the P's and E's being on the same voltage?

Obviously still lots of work to do on the hybrid implementation. Can you please check what your System Agent power is?

It sounds like the System Agent has lot harder time powering down than on the previous generation, plus the overall power has gone way, way up.

You see the implications of this right? The graph they showed, if the SA power is on the level of Tigerlake it would have been 10-15% faster. The System Agent power use is so high that I was getting confused between it and rest of system power. From the tests it looks like it affects it even on relatively light workloads.

If it cannot power down as low as Tigerlake(which was already not too good compared to Cometlake) and the SA problems carryover to the U, potentially it could be a minor disaster for battery life.

Maybe it'll be fixed on the U, but I am not sure. Seems like an Alderlake problem not -H problem.

Some like Davidbepo hate the hybrid concept very much. I understand his point of view but I still think this is the way to go and it'll get lot better as time goes on.

 

[igor_kavinski]

Does the System Agent consume the same power with E-cores disabled?

 

[coercitiv]

Seems like an Alderlake problem not -H problem.

I'm not so sure about that: on desktop I don't have a System Agent reading in HWInfo, but the "CPU Package Power" and "IA Cores Power" difference is much lower something around ~2.5W versus ~8.5W on 12700H in the Notebookcheck review.

This may also explain the difference in performance in my previous tests.

 

[igor_kavinski]

PRO Z690-A DDR4 (msi.com)



What is Intel Tile technology?

EDIT: Intel and Tile Bluetooth Technology Infographic

What a stupid name for this technology. I thought they added support for CPU tiles!

 

[IntelUser2000]

I'm not so sure about that: on desktop I don't have a System Agent reading in HWInfo, but the "CPU Package Power" and "IA Cores Power" difference is much lower something around ~2.5W versus ~8.5W on 12700H in the Notebookcheck review.

This may also explain the difference in performance in my previous tests.

Interesting to think how this might unfold.

 

[igor_kavinski]

遠坂小町 on Twitter: "@VideoCardz https://t.co/mJoOEgYvCV" / Twitter

Funny how Intel is being the copycat, using AMD's HX naming scheme to indicate Ultra performance level. How things have changed.

 

[mikk]

遠坂小町 on Twitter: "@VideoCardz https://t.co/mJoOEgYvCV" / Twitter

Funny how Intel is being the copycat, using AMD's HX naming scheme to indicate Ultra performance level. How things have changed.

H and U naming scheme came from Intel years ago, AMD copy pasted all of Intels naming in the past.

 

[igor_kavinski]

H and U naming scheme came from Intel years ago, AMD copy pasted all of Intels naming in the past.

I know that but the 800 pound gorilla needing to copypaste is when it shows desperation.

 

[nicalandia]

Factory OC 12900K

Intel Core i9-12900KS 5.5 GHz CPU Benchmarks Leak: Up To 5% Faster In Single & 10% Faster In Multi-Threaded Tests

First benchmarks of Intel Core i9-12900KS 5.5 GHz Alder Lake CPU have leaked out & show a decent performance increase over the 12900K.

wccftech.com

 

[lobz]

This will explain it all:

You realize he doesn't mean the brand and the company itself with its 100000+ employees, right?

 

[dullard]

You realize he doesn't mean the brand and the company itself with its 100000+ employees, right?

You made a good joke! If you need another:

I just hate Intel, and it will be many years before I can forgive their greed in the past.

I could go on and on spamming this thread with dozens of times he called Intel or its chips jokes, crap, etc. But, I prefer just to quote himself when he says he hates Intel in multiple threads.

 

[Markfw]

You made a good joke! If you need another:

I could go on and on spamming this thread with dozens of times he called Intel or its chips jokes, crap, etc. But, I prefer just to quote himself when he says he hates Intel in multiple threads.

That quote from me was in 2002. Talk about digging up the past.

I still don't like their marketing, and the 12900k and their server chips, but the recent alder lake (this thread) 12600k and 12700k and 12400 are good gaming cpus.

 

[dullard]

That quote from me was in 2002. Talk about digging up the past.

I wouldn't dig up the past if you were honest with yourself. You posted 2 decades of this:

AMD is ahead right now, read and weep Intel fanboys.

and then keep claiming to be neutral. A search of your posts calling people "intel fanboys" runs multiple pages. You keep going into almost every thread about an Intel chip and say to buy AMD instead. You keep claiming that you'll wait for reviews for your purposes, then say Intel chips are a joke without reviews on any program you use, etc. And your second quote was from 2004, when AMD was ahead.

It is fine to be biased. AMD makes great chips. But your dishonesty about your feelings just really is annoying here.

 

[Markfw]

I wouldn't dig up the past if you were honest with yourself. You posted 2 decades of this:

and then keep claiming to be neutral. A search of your posts calling people "intel fanboys" runs multiple pages. You keep going into almost every thread about an Intel chip and say to buy AMD instead. You keep claiming that you'll wait for reviews for your purposes, then say Intel chips are a joke without reviews on any program you use, etc.

It is fine to be biased. AMD makes great chips. But your dishonesty about your feelings just really is annoying here.

I am honest. When Intel is ahead, I support them, when AMD is ahead, I support them. You just quote from 20 years ago, and only pick posts that you don't like, none of the recent ones. 20 Years ago, AMD was ahead. Until alder lake they were ahead recently. Now its a mixed bag. But you want to ignore all that. I was a conroe fan for years, dumped AMD like a hot rock. Then Ryzen happened.

Now shut up about things non Alder lake in the alder lake thread.

 

[igor_kavinski]

I personally find it amusing when people are not impressed by Alder Lake. I especially don't understand it when I see this from the TPU i9-12900K review:



That's 16 ADL threads smacking down 32 Zen 3 threads in MySQL. How is that NOT impressive? Of course the 5950X will win in highly parallel workloads. It has almost double the threads! I say almost because wimpy 8 E-cores equal 4 P-cores so that's like 20 threads in i9-12900K. And 5950X costs more too. And you pay for better multicore performance with 2nd class single core performance. Sure, ADL won't suit you if you run highly parallel workloads at 100% CPU utilization most of the time but if your workloads are mixed, ADL likely makes more sense for you to buy.

 

[Markfw]

I personally find it amusing when people are not impressed by Alder Lake. I especially don't understand it when I see this from the TPU i9-12900K review:

View attachment 57961

That's 16 ADL threads smacking down 32 Zen 3 threads in MySQL. How is that NOT impressive? Of course the 5950X will win in highly parallel workloads. It has almost double the threads! And it costs more too. And you pay for better multicore performance with 2nd class single core performance. Sure, ADL won't suit you if you run highly parallel workloads at 100% CPU utilization most of the time but if your workloads are mixed, ADL likely makes more sense for you to buy.

First, that is one benchmark. Second, you are ignoring power consumption. But even I acknowledge they have the better gaming cpu, just not the 12900k.

Oh, and MSRP might be higher, but my last 2 5950x were $600 and $580. And they go for $650 everyplace.