Discussion Intel current and future Lakes & Rapids thread

[BorisTheBlade82]

@mikk
I fully agree. As a sidenote: There will be no getting around BIG.little for AMD as well. And I bet that in the long run the big Cores will top out at around 8 because that simply is enough for most partially-parallelizable workloads while you never can have enough small cores for the truly parallel ones.

 

[DrMrLordX]

Why is it so bad that it might perform like a 5950X? Even if it requires 250W to do so, that's a huge jump in one generation.

. . . sort of? It isn't one generation though. Intel's bizarre development timeline has thrown everything off-kilter. We should have had 10nm Ice Lake on desktop, followed by I guess 10nm TigerLake on desktop, and then either 10nm or 7nm Golden Cove something-or-other on desktop. Instead we got, well, you know. What we got.

In any case, comparing Alder Lake-S to Rocket Lake-S and saying it's a single-generation jump really isn't accurate, since Rocket Lake-S is kind of a gimped Ice Lake on the wrong process (14nm+++). It's really two generations (Sunny Cove/Cypress Cove -> Golden Cove).

It doesn't even look like Alderlake was aiming at desktops at all.

Alder Lake-S will be first to market, though. Alder Lake-P comes later.

 

[leoneazzurro]

Alder Lake-S is aimed at desktops, the M version (with only two P cores) comes later and yes, there are two P cores for a reason.
Edit: corrected P with M

 

[jpiniero]

Alder Lake-S is aimed at desktops, the P version (with only two P cores) comes later and yes, there are two P cores for a reason.

P has 6 big cores. M has 2.

 

[BorisTheBlade82]

P has 6 big cores. M has 2.

Yes, but that is nitpicking. The point still stands. There is some upper border for BIG Cores in the way that at some point "typical" workloads don't benefit enough of them.

 

[leoneazzurro]

P has 6 big cores. M has 2.

True, my point is that for low wattage P cores are a no go.

 

[IntelUser2000]

Alder Lake-S will be first to market, though. Alder Lake-P comes later.

Sure.

But Alderlake-S with 8+8 clearly didn't start as a desktop chip. That's why most were expecting 5900X performance, because even with Gracemont performing better than expected, it doesn't have enough cores.

-S is only coming first because desktops are where they are doing badly. But that -S is only in name.

I am not saying being equal to last year's chip is an excellent result. But that they are making great progress. If the hybrid setup can provide consistency then I bet you mobiles will do awesome. 2+8 is enough to beat 4 core Tigerlake-U by a significant amount and maybe they can do it at 15W rather than 20W+ it requires now. 4+8 might double the MT performance and 6+8 will be an icing on the cake.

@mikk Mentions of GT3 means they were considering it at one point, even for Icelake-U. At 45mm2 size for 96EU GT2, they just need to add about 25mm2 to double it to 192EUs.

 

[Accord99]

AMD could have set the maximum power to 250W, but chose not to because they achieved total victory on perf/W. But please, crank that 5950x to 250W and post benchmark results. It is always fun to see where things really stand, again. We already know that would be a massacre even against ADL.

Also because it starts to hit the really exponential part of the power consumption curve. Kitguru using the built-in overclocker got 13% more performance for 133W more at the wall (with a Titanium rated PSU). A 5950X with a package power of 250W doesn't look all that much better than a 142W package power 5950X.

That's well within the rumored range of the 250W 12900K with a properly mounted cooler.

 

[Ajay]

Also because it starts to hit the really exponential part of the power consumption curve. Kitguru using the built-in overclocker got 13% more performance for 133W more at the wall (with a Titanium rated PSU). A 5950X with a package power of 250W doesn't look all that much better than a 142W package power 5950X.

That's well within the rumored range of the 250W 12900K with a properly mounted cooler.

Wall power is useful for looking at the efficiency of Systems, not CPUs.

 

[Accord99]

Wall power is useful for looking at the efficiency of Systems, not CPUs.

They are useful for comparing the same system where the only difference is overclocked settings for the CPU.

 

[Det0x]

AMD could have set the maximum power to 250W, but chose not to because they achieved total victory on perf/W. But please, crank that 5950x to 250W and post benchmark results. It is always fun to see where things really stand, again. We already know that would be a massacre even against ADL.

I already posted my comparison back in post #13198

Bone stock memory + 5950x load-optimized-defaults in bios:
125w = 26182 points


165w @ 1.1vcore get -> 4550/4400mhz = 30008 points


181w @ 1.138 vcore get -> 4600/4500mhz = 30538 points


If i throw power/performance out the window and push anything to the max = 322279 points at 300w

 

[pakotlar]

Cant really say i'm impressed with either Alderlake or Intel "7nm" IF it took PL2 255w to achieve those 30.5k points in Cinebench r23..

This is my everyday 1 year old 5950x

165w @ 1.1vcore get -> 4550/4400mhz = 30008 points
View attachment 50539

181w @ 1.138 vcore get -> 4600/4500mhz = 30538 points
View attachment 50540

If i let my 5950x use above 250w freely i can score up to and above 32k in cinebench r23

*edit*
Bone stock memory + 5950x load-optimized-defaults in bios:

125w = 26182 points
View attachment 50542

I wonder why my 5950x, with a 200W cap and PBO at +0, -15 on 2 best cores, -20 on next best 2, -30 on others scores only 28,700 on air. EDC at 145, TDC at around 160.
Clocks according to HWInfo64 are between 4425 or 4450

 

[majord]

Good point but on one hand you are writing that desktop power is a disaster and then in the next sentence you're implying desktop power is irrelevant (e cores a waste of time on desktop)? Perhaps these ADL systems might run at very low power on the e-cores much of the time and only ramp up the watts for shorts bursts for many user scenarios. Make actual real-world power usage not so high. Performance leaks suggest ADL is going to be competitive with the 5950X in some areas and may cost less. I think ADL is going to be much better than RKL. Where RKL was basically behind Zen 3 everywhere there may be usage cases where ADL might be a better solution for Zen 3. Don't get me wrong Zen 3 was and still is a masterpiece, but we should give Intel some credit for having the guts to bring something completely different to the arena.

e cores are a waste or time if they are being operated out of their ideal efficiency range to help achieve competitive peak performance.

Secondly , the sort of power savings a big.little arrangement brings regard to spinning off non perf critical tasks to the e cores is completely irrelevant in the context of typical desktop power use. Any power savings will get lost in amongst the typical desktop platform base power draw .

 

[jpiniero]

e cores are a waste or time if they are being operated out of their ideal efficiency range to help achieve competitive peak performance.

Can't look at it that way. You have to compare to having 10 big cores. Also remember there will be mobile parts using Alder Lake-S.

 

[DrMrLordX]

But Alderlake-S with 8+8 clearly didn't start as a desktop chip.

I would have to see a 12900k's performance when power limited to 45W or 65W before I would agree with that statement.

 

[Hulk]

e cores are a waste or time if they are being operated out of their ideal efficiency range to help achieve competitive peak performance.

Secondly , the sort of power savings a big.little arrangement brings regard to spinning off non perf critical tasks to the e cores is completely irrelevant in the context of typical desktop power use. Any power savings will get lost in amongst the typical desktop platform base power draw .

You most likely know more than I do but it is possible that...?

Point 1 - The e-cores might not be a waste of time if they assist in completing tasks faster or allow better, smoother multitasking even if they are outside of peak efficiency range? In addition how about the times when the user is simply web browsing or e-mailing? Maybe shutting down the coves and running monts only will show good system efficiency?

Point 2 - While the Coves might draw 10 times the power of the Monts, say 200W vs 20W, the Coves are likely to be used 10 times less in regards to total operating time. So this difference in usage between Big.Little could make a difference to total power draw over a long period of time. Meaning if we don't simply examine high compute situations.

Typical Mont only conditions could be watching streaming video while browsing the web, e-mail, and doing a little MS office. Fire up Photoshop for some editing and here come the Coves, but only for short bursts during sharpening or other intensive tasks. Or when video editing when high compute is needed for preview or final render. I can imagine lots of situations where the Monts, if they are nearly Skylake compute will be able to run the show.

 

[lobz]

It's a desktop K product, what do you expect. They are definitely pushing the power draw but that was to be expected.

That was nowhere near the question at hand.

 

[lobz]

@Carfax83
While your argument holds true this basically just means that TigerLake is a bit less inefficient than Zen3 when both are way beyond their Perf/W sweet spot. Looking at it from this way that is not too overwhelming.

BTW
Just have a look at my PES Thread. There are samples of a 8C TGL as well as a 5900HS

Well yeah, if you demolish both an iphone and a galaxy s with a waffle iron, they both become trash in every metric.......

 

[Markfw]

I reserve all comments until a recognized authority like Anandtech has FULLY reviewed the final product. And actually, I want 3-5 of these before I will comment. First looks say it may be competitive, but at higher power usage. What if its only in a couple of benchmarks that it wins ? What if it fails in most things ? What if the power usage is way too high ? What if its a combination of all of these ? And at what magnitudes ? Lets talk once its out... And maybe even a week later.

 

[coercitiv]

You most likely know more than I do but it is possible that...?

Point 1 - The e-cores might not be a waste of time if they assist in completing tasks faster or allow better, smoother multitasking even if they are outside of peak efficiency range? In addition how about the times when the user is simply web browsing or e-mailing? Maybe shutting down the coves and running monts only will show good system efficiency?

Point 2 - While the Coves might draw 10 times the power of the Monts, say 200W vs 20W, the Coves are likely to be used 10 times less in regards to total operating time. So this difference in usage between Big.Little could make a difference to total power draw over a long period of time. Meaning if we don't simply examine high compute situations.

Typical Mont only conditions could be watching streaming video while browsing the web, e-mail, and doing a little MS office. Fire up Photoshop for some editing and here come the Coves, but only for short bursts during sharpening or other intensive tasks. Or when video editing when high compute is needed for preview or final render. I can imagine lots of situations where the Monts, if they are nearly Skylake compute will be able to run the show.

I don't know how to explain this for the 10th time:

• Big cores on the desktop consume an irrelevant amount of power when idle or under light loads. These cores are also used in 10-15W CPUs where they sip power with appropriate power management enabled. The same can be achieved on the desktop, light loads can sip power on big cores today.
• When doing heavy work, expect E cores to use comparable amounts of power as P cores, obviously following a scaling ratio that I expect Intel will bake into the firmware / software to maximize throughput. There's a very high chance that the E cores will be thermally limited before the P cores, given specific MT workloads. E does not stand for Eco.
• Even when scheduling one single thread, if the workload isn't light enough then the E core will likley use more energy to complete the job.

Obviously we can agree to disagree on this, but I hope nobody will act all surprised when it turns out E cores on the desktop are used solely for area efficiency. Expect the scheduler to heavily prioritize P cores. Expect E cores to be milked only in heavy throughput situations, with power and thermals to match.

We'll talk more on review day, I'm sure Anandtech will give us good insight into power distribution among many other interesting things. I'll eat my crow if I have to, but I suspect the day will leave me with an empty stomach.

 

[Asterox]

In short, vs R5 5600X in Cinebench R20.

But what about system memory choise? "Cheep ddr5 with high latency", or classic choise cheep ddr4 3200mhz CL16 for 70$.

https://www.amazon.com/Patriot-PC4-25600-288-Pin-Desktop-PVS416G320C6K/dp/B07KXMWHBY

- 10% faster in ST Cinebench
- 10% faster in MT Cinebench

https://twitter.com/x/status/1447499861255802885

AMD Zen 3 Ryzen Deep Dive Review: 5950X, 5900X, 5800X and 5600X Tested

www.anandtech.com

 

[Zucker2k]

So, in single thread, GC vs Zen 3 4.4GHz vs 4.6GHz, the 12400 is still 10% faster, and according to a commenter in the twitter link, 21% faster than 11400. I think Zen 4 delay is beginning to make sense.

 

[mikk]

This looks really good for a moderate clock speed of 4.4/4 Ghz. 5600x can boost up to 4650Mhz in comparison. If this is a 200USD SKU AMDs 5600x is way overpriced, they have to lower it.

 

[Zucker2k]

This looks really good for a moderate clock speed of 4.4/4 Ghz. 5600x can boost up to 4650Mhz in comparison. If this is a 200USD SKU AMDs 5600x is way overpriced, they have to lower it.

And package power is 78w. I'm beginning to think Intel's 7 is probably a tad more efficient than TSMC's 7 given the perf/consumption at these clocks. Those who keep clocks in the sane zone should see some drastic power reduction.

 

[mikk]

And package power is 78w. I'm beginning to think Intel's 7 is probably a tad more efficient than TSMC's 7 given the perf/consumption at these clocks. Those who keep clocks in the sane zone should see some drastic power reduction.

Yes 12400 is more efficient than the 5600x which is also close to 80W package power in other tests. However Intel cannot go as high as Zen 3 with the clock speeds to achieve this efficiency, 4 Ghz vs ~4.5 Ghz in multithread is quite a big difference. Assuming this 12400 test is representative.