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.

 

[uzzi38]

I didn't check the video.
700mV is very low, true. Then let's say this score is for 45W, that's 20% better than 5900HX 45W.

Yep. Now you're on the right lines of thinking here.

45W+ Alder Lake will have a clear advantage.

35-45W is likely going to be a mixed bag depending on the workload.

<35W is Rembrandt's space.

 

[insertcarehere]

It's going to depend on sustained power limit.

Even against 6+8 RMB will be able to compete favourably at specific power targets.

The very fastest Cezannes get like ~14k on Cinebench R23, and that's with extremely aggressive settings that allow the 5900HX to draw up to 95w, thermals permitting. https://www.computerbase.de/2021-02/amd-ryzen-5000-mobile-test/2/#diagramm-package-power-blender-bmw

Compared to the (allegedly) 18.5k Cinebench R23 score out of ADL-P. I seriously doubt RMB closes a 30+% gap between ADL-P and an aggressively tuned Cezanne. https://www.notebookcheck.net/First...-3-and-Apple-M1-Max-in-the-dust.579828.0.html

I didn't watch that video until now.
700mV is very low and CPU package was 44.18W.
So the score will be worse than that.

The original post was not a leak, it was taking a 12900K, disabling two P-cores and UVing.

I'd imagine ADL-P would have more efficient IOD tuning than ADL-S.

 

[TESKATLIPOKA]

Compared to the (allegedly) 18.5k Cinebench R23 score out of ADL-P. I seriously doubt RMB closes a 30+% gap between ADL-P and an aggressively tuned Cezanne. https://www.notebookcheck.net/First...-3-and-Apple-M1-Max-in-the-dust.579828.0.html

Performance looks great.
The question is how much power did It need for this score. Officially this 12700H ADL should be 45W, but this sample could(should) be way above It.
12600K with 4 less E-cores manages 17699 in CB23 at techpowerup.

 

[insertcarehere]

Performance looks great.
The question is how much power did It need for this score. Officially this 12700H ADL should be 45W, but this sample could(should) be way above It.

I am pretty sure this was tested with a power limit of >45w, but when even Cezanne SKUs that can draw 65++w score 13-14k that's still a marked improvement.

 

[Carfax83]

Do any of you think Intel will eventually add AVX-512 to a future E core? Wouldn't that screw up the power efficiency?

Small, densely packed cores with wide vectors would be reminiscent of Knights Landing.

 

[Hulk]

Clock speeds were normalized to show IPC improvements. Allowing either chip to boost to whatever it wants would fail to show this.


Man, I have been here a long time. Longer than even my user account suggests, yet your post blew my mind. Because someone reached a different (and correct) conclusion than you, suddenly they are biased? ADL-S is an impressive part, but there is nothing Intel can possibly do that will make it consistently beat a 5950x in multicore workloads on a perf/watt basis. I don’t care about power consumption. Cap the 5950X at 142W and dump all the power into the 12900k. Even if you pushed the chip to the limits, It would still lose several benchmarks to the 5950X. Also yes, we get to compare the two chips because while Intel has to send power consumption through the roof to add more cores, AMD sticks with the 142W power consumption, regardless of the number of cores or the frequencies.

Mark and others (including myself) will have something positive to say about Intel when Intel delivers.


I don’t mean to nitpick here, but that 99% figure is flat out wrong. We don’t know the exact number, but I bet it is under 90%. Possibly even under 80%. Shoot, I could even argue for 70%. Almost anyone in a science that relies on tech would benefit. Almost anyone involved in 3D rendering, video production, etc. would benefit. Shoot, outside of college students, gamers, and folks who only use Microsoft Office, is there anyone who wouldn’t benefit from more cores and higher frequencies?

In an age where climate is changing rapidly, should we really accept a CPU that needs 40% more power to come close to or meet last years flagship? Intel has work to do on the high end. They are getting there, but they aren’t there yet.

Shoot, I may even replace my Cezanne based laptop with Alder Lake because that appears to be where Intel is going to lap AMD. We will see.

Moving from a 4770K to a 12700K made me realize some things. First, I was slowly "boiling" with the 4770K and didn't realize it. Sure I could do everything I needed to but many tasks were frustratingly slow. The thing is as your processor "ages out" it's a slow process. Compared to the curve it starts out at the top and then slowly moves to the bottom. But the user never experiences "the curve" so what you are using is your normal... until you upgrade.

I did some photoshop work last night with the new rig and it made photoshop fun again. Everything was instantaneous. Previewing video projects is now stutter free.

So did I NEED the 12700K? Would the 5800X or 12600K have been a sufficient upgrade? Of course. Despite all the arguing here you really can't go wrong with Zen 3 or Alder Lake. Both are great.

When it comes to highly threaded tasks more cores are going to get it done better assuming architectures and processes are relatively close and I think we can make that assumption with Zen 3 and ADL. The cost of more cores is more silicon and that is more expensive. I think Intel was smart in how they targeted the market with ADL. I'm having a hard time describing Zen 3 as either better or worse than ADL. They are different and it takes a little bit of user nuance to understand which one would better suit your needs and budget.

 

[Hulk]

BTW the one application that is still dropping the hammer, even on my new rig is DxO PureRaw. It's an amazing little application that converts RAW images but is absolutely crazy when it comes to compute!

With my 4770K it was actually faster using the iGPU over the CPU. I think with the 12700K it's faster using the CPU but I haven't timed it yet. I do know that when processing an image it floors all 20 logical cores!

Anyone using DxO PureRaw? If so what is your experience compute-wise? I have a feeling the only way to tame this compute hungry beast is with a powerful OpenCL graphics card.

 

[DrMrLordX]

The 5950X actually has a lot more headroom

Not really. You can do some tuning with curve optimizer but in terms of voltage/power headroom, N7 isn't very forgiving.

Besides, that wasn't the original proposition anyway. It was stated that a stock 5950X was going to win some benchmarks no matter how much power you throw at Alder Lake-S. That simply isn't true.

 

[uzzi38]

The very fastest Cezannes get like ~14k on Cinebench R23, and that's with extremely aggressive settings that allow the 5900HX to draw up to 95w, thermals permitting. https://www.computerbase.de/2021-02/amd-ryzen-5000-mobile-test/2/#diagramm-package-power-blender-bmw

Compared to the (allegedly) 18.5k Cinebench R23 score out of ADL-P. I seriously doubt RMB closes a 30+% gap between ADL-P and an aggressively tuned Cezanne. https://www.notebookcheck.net/First...-3-and-Apple-M1-Max-in-the-dust.579828.0.html

Wow, you didn't read what I wrote at all, did you?

I never claimed that Rembrandt would top the charts at high power levels, did I?

 

[igor_kavinski]

Anyone using DxO PureRaw? If so what is your experience compute-wise? I have a feeling the only way to tame this compute hungry beast is with a powerful OpenCL graphics card.

DeepPRIME* hardware acceleration – Customer Support (dxo.com)

Windows recommended requirements
NVIDIA GTX™ 1060, AMD Radeon™ RX 580 or better

On such GPUs, you should expect a processing time of about 2 Mpx per second (or faster with better GPUs).

PassMark - Intel UHD 770 - Price performance comparison (videocardbenchmark.net)
PassMark - GeForce GTX 1060 - Price performance comparison (videocardbenchmark.net)

Compute performance of UHD 770 as measured by Passmark doesn't seem to be that bad compared to the Geforce 1060. However, DX12 performance is four times slower.

 

[lobz]

I think the 12600K best exemplifies Intel Big-Little working.

Intel Core i5-12600K Review: 5600X Defeated

The Core i5-12600K is Intel's latest mainstream CPU and a direct competitor to AMD's popular Ryzen 5 5600X. The Alder Lake chip packs 6 P-cores and 4...

www.techspot.com

It's a 6+8 design. Which is similar area to an 8+0 design, and it wins overall performance against the 8+0 Ryzen 5800x.

And it does it while using a tiny bit less power, despite Intel reportedly being on a worse process than AMD:

The node ADL is on is not a worse process anymore than the N7, which AMD uses since forever. Intel has arrived, yes, but I wouldn't really say they arrived in time 🤣

 

[lobz]

Good point, though it's a complacent strategy that led them to the desperation of releasing the power hogging 12900K, just to win some benchmarks. I hope Zen3D is not just Zen 3 with V-cache. I like Intel in desperate mode. Forces them to move their performance unlocking secrets from the research phase to production phase much quicker.

Zen3D *is* Zen 3 with V-cache.

 

[igor_kavinski]

Zen3D *is* Zen 3 with V-cache.

Yeah on the surface it seems so, but what if AMD engineers tweak it, like making the L1 and L2 cache latencies lower? They seem to have had quite a bit of time on their hands since they announced it in May so there is a possibility of some last minute tweaks to make it more competitive with Alder Lake. Or they might be able to boost it up to 300-500MHz higher with the new stepping.

 

[DrMrLordX]

Not really the right thread for that.

 

[TheELF]

Killing AMD = MOAR MONEY!

Dead AMD = ( intel = at&t )
= far less money, if any at all

Also look at the money intel made before ryzen and after,
healthy amd = moar money

 

[igor_kavinski]

healthy amd = moar money

I hope you or someone else will help me understand how that works. How does Intel sell more CPUs if their competitor is thriving and taking away market share from them?

 

[Heartbreaker]

You said all, that's one case where the 6+0 would for sure be faster. Given that the E cores are more efficient in MT workloads you'd have to think that 4+8 would be better for the typical OEM desktop with low TDPs and crappy cooling. Albeit that users likely wouldn't notice.

So you trot out an edge case that has nothing to do with relative core counts, but deals only with enabling or disabling a specific instruction?

Why would 4+8 be better than 6+0 for the typical desktop? Better at embarrassingly parallel, isn't better at everything.

The typical desktop likely isn't spending much time doing embarrassingly parallel tasks, like running Cinebench.

In a hybrid system, you have to determine the point where you have enough P-Cores for most workloads, before you can start relying on E-Cores to crunch through the embarrassingly parallel loads.

Is that 1-P, 2-P, 4-P, 6-P, 8-P, 12-P, 16-P or more??

Given that Intel will have done this analysis when designing a hybrid chip, and even their mainstream Laptop part has 6-P cores, it really seems like 6-P cores, is what they determined a reasonable minimum coverage for a wide range of use cases.

 

[LightningZ71]

Given that, with existing products, there's rarely much of a difference in the user experience between six core products and eigh core products, and precious few non-specialized tasks that are so parallel that they evenly load 7+ cores, it looks like 6+4 and 6+8 are very good design targets.

 

[jpiniero]

So you trot out an edge case that has nothing to do with relative core counts, but deals only with enabling or disabling a specific instruction?

Already explained that. 4+8 would be faster or comparable at lower power levels in most benchmarks (meaning those not AVX-512) but users aren't going to notice either way. Hence why it's marketing driven.

 

[Heartbreaker]

Given that, with existing products, there's rarely much of a difference in the user experience between six core products and eigh core products, and precious few non-specialized tasks that are so parallel that they evenly load 7+ cores, it looks like 6+4 and 6+8 are very good design targets.

I suspect one of the important targets where that had Intel maintain 6 P-core minimum, was gaming. Gaming can benefit from more than 4 cores but they need to be strong cores. Gaming is multi-threaded, but nowhere near embarrassingly parallel.

 

[Heartbreaker]

Dead AMD = ( intel = at&t )
= far less money, if any at all

Also look at the money intel made before ryzen and after,
healthy amd = moar money

Monopolies are often bad for consumers, and innovation, but they are great for raking in money.

Ask Standard Oil, American Steel, and yes AT&T which was the most profitable company in the world before it's breakup.

So I'm really not getting the far less money argument, using AT&T as an example.

 

[Markfw]

There are 8 charts there, and the 5950x only came ahead in two instances, by a small margin. Yet, somehow you couldn't figure out how to apportion a clear winner even in this clearly open and shut instance. This is why you get accused of bias. If the numbers fit your bias they're good, if it doesn't they're bad.
Then there's also this:

AGAIN, you and that articles text are comparing the 5900x, not the 5950x, so I won't bother replying anymore, SINCE YOU CAN'T READ.

 

[Zucker2k]

AGAIN, you and that articles text are comparing the 5900x, not the 5950x, so I won't bother replying anymore, SINCE YOU CAN'T READ.

How many tests did the 5950x win? Honest.

 

[eek2121]

The very fastest Cezannes get like ~14k on Cinebench R23, and that's with extremely aggressive settings that allow the 5900HX to draw up to 95w, thermals permitting. https://www.computerbase.de/2021-02/amd-ryzen-5000-mobile-test/2/#diagramm-package-power-blender-bmw

Compared to the (allegedly) 18.5k Cinebench R23 score out of ADL-P. I seriously doubt RMB closes a 30+% gap between ADL-P and an aggressively tuned Cezanne. https://www.notebookcheck.net/First...-3-and-Apple-M1-Max-in-the-dust.579828.0.html



The original post was not a leak, it was taking a 12900K, disabling two P-cores and UVing.

I'd imagine ADL-P would have more efficient IOD tuning than ADL-S.

Indeed, based on recent leaks, I think Rembrandt will likely cover < 45W designs along with some 45W designs, and the leaked Raphael-H chip could cover the top 45-54W SKUs. We will see. That is the only way AMD will make up the gap.

How many tests did the 5950x win? Honest.

Nearly all the multicore ones.

 

[Hulk]

Appreciate if you can run with only P-cores and AVX-512. Curious if that scores higher than stock setting.

I was able to turn off the E cores in the BIOS but unlike Ian's motherboard mine did not then show a setting to turn on AVX512, or I couldn't find it.

I did run the test in 2+4 configuration and the result was pretty good. Have a look at the updated chart: http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=thread...k-your-system-new-benchmark-criteria.2588294/

I also ran Cinebench R23 in the 2+4 configuration and the result was 8258. Just for kicks I booted up with only 1 GC core active. Kind of reminded me of my old P4 3.06 from nearly 20 years ago!