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]

you did not mean this comment seriously...

Psst: one of those two chips drops well below turbo speed during gaming sessions unless you have great cooling.

https://www.reddit.com/r/Amd/comments/c3vsvc

 

[epsilon84]

I agree with your IPC improvement cannot overcome a 50% core deficit statement.

But, I think we should move past generic statements such as "6 core CPU minimum". It should depend on use cases. If you truly have single-threaded use cases, the 2 core Alder Lake Celeron G6900 can barely beat the 10 core Comet Lake i9 10900K. https://wccftech.com/intels-most-en...ormance-on-par-with-a-5-3-ghz-core-i9-10900k/ If I were to choose a chip today, it would probably be 6 cores (the 12500T seems ideal to replace my HTPC), but I just don't think we should make general statements.

If you are building a PC for grandma, then sure. I was thinking from an enthusiasts perspective (this being Anandtech) upgrading to an ADL chip, unless CPU performance is of almost no importance (or you strictly need ST performance) I don't see the point in getting a quad (or dual!) core in 2022.

Quads were the mainstream choice over a decade ago, since Sandy Bridge in 2011 and the i5 2500K. I don't think raising the mainstream bar from 4 to 6 cores after 11 years is that unreasonable.

 

[epsilon84]

How is this even possible?



View attachment 55820

Intel's Most Entry-Level Alder Lake CPU, The Dual-Core Celeron G6900, Has Single Core Performance On Par With a 5.3 GHz Core i9-10900K

The Intel Celeron G6900, Entry-Level Alder Lake, dual-core CPU has been tested and offers better single-threaded performance than i9-10900K.

wccftech.com

It seems that it was running at 4.4GHz, not 3.4GHz. Which then makes it entirely possible to beat a Skylake based CPU, even if its running at 5.1GHz single core.

 

[igor_kavinski]

Intel Core i5 12400 "Alder Lake": A Great ~$200 CPU For Linux Users Review - Phoronix

www.phoronix.com

Compared to the 5700G, the i5-12400:

Champ in Web browsing
Champ in code compilation
Almost as good in compression

Almost as good in Creator workloads involving OSPray, Blender, Appleseed, V-RAY, IndigoBench, SVT-VP9, SVT-HEVC, SVT-AV1, AOM-AV1, Rav1e, AVIFenc, Stargate digital audio workstation, libraw, WebP encode, JPEGXL encode, RawTherapee, libjpeg-turbo tjbench, GIMP, Hugin, Darktable, Embree, ASTCenc, and Etcpak

Almost as good in cryptography
Almost as good in machine learning
Champ in Python workloads
Pretty competitive in rendering workloads despite the core count deficit

Gets totally Bulldozered(pun intended) in iGPU tests but then it also posts pretty low power consumption during these tests.



So unless a gamer on a budget absolutely has to have better gaming FPS without a discrete 3D card, Intel has a killer CPU here for $100 less.

 

[repoman27]

I noticed as I was looking over the press decks and product briefs from Intel's CES announcements that Alder Lake-P doesn't support PCIe Gen 5. H45 and U15 are both Gen4 2x8 or 1x8 + 2x4, and if the IoT SKUs are anything to go by, P28 only supports up to 8 lanes total from the CPU.

Also, Andreas Schilling posted some high-res package and wafer shots of ADL-P:

https://twitter.com/x/status/1478429772556521478

 

[IntelUser2000]

The 12100F and 11400 has the same L3 cache. So the performance difference must be due to the latter having 50% more cores.

Though it doesn't explain poorer performance in Rainbow Six Siege. That's pretty old.

The power efficiency argument can become silly. Cause if you have a 200W CPU doing that's twice as fast as a 100W one, you'd do twice as more work rather than doing the same but at half the time. And lots of time there are applications that needs the full power all the time. Absolute power use matters as much as task power use.

The power efficiency part matters in millisecond levels where the CPU can ramp up and down before you even notice it.

 

[Mopetar]

The 12100F and 11400 has the same L3 cache. So the performance difference must be due to the latter having 50% more cores.

Though it doesn't explain poorer performance in Rainbow Six Siege. That's pretty old.

I think a lot of performance oddities can just be attributed to Alder Lake being so new. It may also come down to how some code is compiled and I wouldn't be surprise if some of the design changes that Intel made can result in performance regression with binaries targeting older Intel architectures in some specific instances.

 

[Asterox]

2/2 is ok, SMT or Hyperthreading is to expensive even in 2022.

Strange, Far Cry 6 only 2fps with discrete GPU.

Athlon 3000G it can run Far Cry 6 with Vega3 iGPU.

Geekbench Search - Geekbench

browser.geekbench.com

 

[IntelUser2000]

I think a lot of performance oddities can just be attributed to Alder Lake being so new. It may also come down to how some code is compiled and I wouldn't be surprise if some of the design changes that Intel made can result in performance regression with binaries targeting older Intel architectures in some specific instances.

Unless Rainbow Six Siege is branchy and it doesn't like the extra pipeline stage. But that's a huge impact for one stage if that's the case.

 

[JoeRambo]

Interesting rumour about Raptor Lake cache.

Intel 13th Gen Raptor Lake CPUs Could Feature Up To 55% More Cache Over Alder Lake For Enhanced Gaming Performance

Intel may be introducing a major cache redesign in its upcoming 13th Generation Raptor Lake Desktop CPUs to tackle AMD's V-Cache solution.

wccftech.com

L3 slices stay the same @ 3MB, for total of 36MB of L3 ( versus current gen 30MB of L3 ).
But Intel seems to be doubling down on increasing L2 caches. IF they can reign in latency, larger L2 means less "need" of L3 size and bandwidth.

 

[igor_kavinski]

Wow. More cache and lower power usage are pretty good reasons for Alder Lake owners to chuck out their current babies for some shiny new Raptor Lake love.

 

[Mopetar]

Wow. More cache and lower power usage are pretty good reasons for Alder Lake owners to chuck out their current babies for some shiny new Raptor Lake love.

I think the smart move is to get a solid board and something like a 12600K that's still going to deliver excellent gaming performance without breaking the bank. That gives the option of upgrading to Raptor Lake or something further down the line later on in 3 or so years.

Unless they're the type of person that's always upgrading to the latest and greatest how many people are going to jump on the next generation immediately?

 

[Exist50]

Wow. More cache and lower power usage are pretty good reasons for Alder Lake owners to chuck out their current babies for some shiny new Raptor Lake love.

That would only be a couple percent extra performance. Minor refresh.

 

[igor_kavinski]

Still, it's good to know this is going to be an excit

That would only be a couple percent extra performance. Minor refresh.

The increased cache could make that double digit in gaming. Also, lower power usage could allow it to boost higher.

 

[Exist50]

Still, it's good to know this is going to be an excit

The increased cache could make that double digit in gaming. Also, lower power usage could allow it to boost higher.

I'd love to be wrong, but even in gaming, I think a double digit improvement is very optimistic. And the increased efficiency rumor was tied to DLVR, which is mobile only.

 

[Abwx]

Did find this slide at NBC :

AMD Zen 3 Cezanne vs. Intel Tiger Lake-H 45 W power efficiency comparison: Ryzen 9 5900HS and Ryzen 5 5600H trounce the entire Tiger Lake-H lineup - NotebookCheck.net

We compare the power efficiencies between AMD Zen 3 Cezanne and Intel Tiger Lake-H 45 W processors during a Cinebench R15 multi-core run when connected to an external monitor.

www.notebookcheck.net

Thanks to some powerfull point Intel start at parity with TGL vs Cezanne.

That being said if at least the improvement from TGL to ADL is accurate then AMD can rest on their laurels in the waiting of their next gen.

 

[mikk]

Cinebench is more of a floating point intensive benchmark, this is not comparable at all to Integer of SPEC CPU. Prior to Golden Cove Intel wasn't great in Cinebench versus Zen 3.

 

[IntelUser2000]

Cinebench is more of a floating point intensive benchmark, this is not comparable at all to Integer of SPEC CPU. Prior to Golden Cove Intel wasn't great in Cinebench versus Zen 3.

Cinebench doesn't really benefit from AVX, meaning at some point it became a decent benchmark for showing architectural gains, and thus similar to Integer gains.

That being said if at least the improvement from TGL to ADL is accurate then AMD can rest on their laurels in the waiting of their next gen.

It may be higher power but performance wise TGL-H is not far behind. If increasing TDP further for Rembrandt-H increases performance dramatically sure, but at this point it's higher performance period.

The top end performance is the greatest importance for these large laptops anyway.

Also it seems a Gracemont cluster being equal to a Golden Cove core in MT is a good rough estimate. So 6+8 is like a 10, and a 30-40% improvement over TGL-H makes sense.

 

[Exist50]

Also it seems a Gracemont cluster being equal to a Golden Cove core in MT is a good rough estimate. So 6+8 is like a 10, and a 30-40% improvement over TGL-H makes sense.

It should be greater than 1:4, no? Think there might be some weirdness with power.

 

[Saylick]

I'd love to be wrong, but even in gaming, I think a double digit improvement is very optimistic. And the increased efficiency rumor was tied to DLVR, which is mobile only.

It took AMD +64 MB of L3 cache for +15% fps at 1080p. I doubt adding 1 MB of L2 to the big cores and 6 MB of shared L3 (which will be used by the 8 extra small cores) will generate anything other than a few % gains. Probably like 1-3% at best. Otherwise, the math just doesn't equate.

 

[Khato]

It took AMD +64 MB of L3 cache for +15% fps at 1080p. I doubt adding 1 MB of L2 to the big cores and 6 MB of shared L3 (which will be used by the 8 extra small cores) will generate anything other than a few % gains. Probably like 1-3% at best. Otherwise, the math just doesn't equate.

It's entirely dependent on what the bottleneck is in the design for a particular workload. For games which show benefits from increased L3 size we know that AMD sees what, around 20% increase going from their 16MB L3 APU to the 32MB L3 CPU? And supposedly 15-20% from tripling to 96MB. Pretty clear diminishing returns. Meanwhile there are some indications that running an i9-12900k at same settings as an i5-12600k to create a 20MB to 30MB L3 comparison shows a 10-15% increase. Hence going from 30MB to 36MB could easily be an additional 4-6% by itself. The increased L2 size by comparison is more problematic to predict. It certainly is interesting to note that Intel has gone from 256KB with the Skylake iterations to 512KB on Icelake and finally 1.25MB on Tigerlake/Alderlake - certainly indicative of L2 cache size having been identified as a vector for improving performance.

 

[Markfw]

Is it me, or are all these "issues" about E cores, kind of killing the whole P-core/E-core deal thing ? Personally, I love the equal cores that AMD has right now due to their efficiency and performance. I had the money, and almost got a 12900k, but after all this, I got my 4th 5950x for $649.

Sorry Intel.

 

[DrMrLordX]

Is it me, or are all these "issues" about E cores, kind of killing the whole P-core/E-core deal thing ?

Might be under Windows, but maybe not so much under other operating systems. MS just hasn't gotten the balance right for desktop.

 

[CakeMonster]

Windows will get b.L right (or at least decently right) sooner or later. Those extra E cores will help with the longevity of the CPU's that have them, which is arguably when it matters most since they're not doing all that much now. b.L is the direction things are going anyway, so I wouldn't discount E cores at the moment if you plan on keeping your platform for longer than say 2 years.

 

[IntelUser2000]

It should be greater than 1:4, no? Think there might be some weirdness with power.

Yea, bit of a mind fart haha.

I should have said the two clusters equal 4 cores, hence the 30-40% improvement of Alderlake-P 6+8 over Tigerlake-H. 6+8 = ~10, and add a bit for Golden Cove gains.

That's why the gains should be disproportionately large on the U and the Ps.

U: 2-4 to 2+8 which is like a 6 core
P: 4 to 4+8 and 6+8 which is like 8 and 10 cores.

Though that's mostly catching up to AMD, hence it's more worth it for them to shout about the 45W+ chips.

I would like to eventually see proper M parts(proper I mean by not Lakefield!) at 5-7W power envelopes. I'd loved to have seen more but based on Intel saying zero about Alderlake and battery life, we'd have to wait until Raptor for anything.