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.

 

[mikk]

Looks like all the K chips will be based on the 8+8 die. So will the 6 core die be used only for locked, low frequency chips? Will they also bin these chips or will they waste the golden samples? Why?

Yes it's for lower clocked non OC models, the fastest 6+0 seems to be the i5-12600. However this model comes with reasonable turbo speeds of 4.8 Ghz ST and 4.4 Ghz MT. On wccftech there is a SKU overview.

 

[jpiniero]

Yes it's for lower clocked non OC models, the fastest 6+0 seems to be the i5-12600. However this model comes with reasonable turbo speeds of 4.8 Ghz ST and 4.4 Ghz MT. On wccftech there is a SKU overview.

Given that it's WTFTech, it's probably not accurate. But not too far off. Not having the small cores is going to make the 12600 a bad deal compared to the 12600K unless the price gap is much bigger than in the past.

 

[Timorous]

Something interesting that came out of the reddit AMA is that in testing Intel found that some games do like the DDR5 bandwidth and others as we know prefer DDR4 due to lower latency.

In their 11900K vs 12900K slide there were some really big gains for some games but looking at them a lot have more going on CPU wise. It seems that it may be possible that some of these gains can be attributed to DDR5 bandwidth (anybody with a fast ram in a comet or rocket lake system want to compare improvements in the likes of Troy, hitman 3 etc at 3200 C14 vs 4400 C36 to get an idea how much bandwidth impacts some of those games that saw huge gains?).

At the other end of the scale there are a fair number of games with minimal improvements (and 1 regression). I think it is quite possible that in a DDR4 system the games on the left hand side end up with a better showing than Intel showed but some of the games with the big 20+% gains do worse.

It will be very interesting to find out.

 

[Asterox]

So one same board can't take DDR4 and DDR5 in the same slots? That's what I kind of figured.

No it cant, DDR4 and DDR5 memory/stick is not physically compatible.

 

[Arkaign]

Something interesting that came out of the reddit AMA is that in testing Intel found that some games do like the DDR5 bandwidth and others as we know prefer DDR4 due to lower latency.

In their 11900K vs 12900K slide there were some really big gains for some games but looking at them a lot have more going on CPU wise. It seems that it may be possible that some of these gains can be attributed to DDR5 bandwidth (anybody with a fast ram in a comet or rocket lake system want to compare improvements in the likes of Troy, hitman 3 etc at 3200 C14 vs 4400 C36 to get an idea how much bandwidth impacts some of those games that saw huge gains?).

At the other end of the scale there are a fair number of games with minimal improvements (and 1 regression). I think it is quite possible that in a DDR4 system the games on the left hand side end up with a better showing than Intel showed but some of the games with the big 20+% gains do worse.

It will be very interesting to find out.

Well, I mean also like DDR4 4000/C17 level stuff or better would probably be pretty great with it, but we'll have to wait to see how it plays out in GN/etc

 

[NTMBK]

... the "Lakes" thread having turned into a nightmare to navigate...

What, a 563 page thread about 6 different generations of product, all being argued about concurrently, isn't easy to follow?

 

[igor_kavinski]

Intel announces 12th-gen Alder Lake CPUs: Our long 14 nm nightmare is over

New desktop CPUs include DDR5 RAM support, PCI Express 5.0, and more cores.

arstechnica.com

Welcome to the 10nm nightmare!

it will be what? 2 years till we see Intel 4? I HOPE for my sake that I'm wrong.

 

[blckgrffn]

What, a 563 page thread about 6 different generations of product, all being argued about concurrently, isn't easy to follow?

I started at the beginning and reread through it and it was a slog. The price you pay for stepping away from the forum for a couple years but wanting to participate in one of the top CPU sub forum threads.

A little off topic, but they moved the notch on DDR5 only slightly? And towards the middle? Ugh. I find it much easier when the key is A) more towards one side to make it obvious and B) different than before enough if you grab a stick from the wrong generation it's stupidly obvious. Yikes.

In any case, it seems like the early reviews should really make it obvious whether DDR5 with it's higher capacities and internal error correction is immediately viable. Fast DDR4 is not hard to find right now, and prices are plummeting.

I can imagine that DDR5 support was/is a big marketing factor - higher numbers being better. I can only imagine talking with someone in line to checkout in Microcenter and how excited they might genuinely be about PCIe 5 and DDR5 support. I've had awkward conversations there in the past. I am sure Dell and HP and the rest like touting some new higher numbers too.

 

[dullard]

Well, newegg has quite a few DDR5, none are ECC

ECC is built into the module itself.

DDR5 has halfway implemented ECC: all DDR5 has ECC for everything that occurs on the memory module. So if a memory chip has a bit flip, it will be caught and corrected. But DDR5 does not have automatic full-system ECC. Meaning if a bit flip occurs during data transmission, that it will not be caught without extra resources.

 

[dullard]

Looks like all the K chips will be based on the 8+8 die. So will the 6 core die be used only for locked, low frequency chips? Will they also bin these chips or will they waste the golden samples? Why?

Do you have a source for that? It seems reasonable that they only are running the 8+8 die right now and will be using the best 6-core chips for high-end mobile. But, they theoretically could have used the 6+8 die for the 12600K and 12600KF. So, is there evidence that they are not doing so?

 

[jpiniero]

Do you have a source for that? It seems reasonable that they only are running the 8+8 die right now. But, they theoretically could have used the 6+8 die for the 12600K and 12600KF. So, is there evidence that they are not doing so?

There is no 6+8 die. Only 8+8 or 6+0.

 

[dullard]

There is no 6+8 die. Only 8+8 or 6+0.

Right in the middle:

https://images.anandtech.com/doci/16881/69.jpg

 

[jpiniero]

Right in the middle:

https://images.anandtech.com/doci/16881/69.jpg

That's for mobile not desktop.

 

[dullard]

That's for mobile not desktop.

Yes, but what is to stop them from using it?

 

[jpiniero]

Yes, but what is to stop them from using it?

Doubt it's fully compatible with the socket. Even if it was they wouldn't do that since there's plenty of mobile products to dump defective dies.

 

[Saylick]

I wouldn't have guessed that there's a smaller 6+0 die, which I interpret as being only 6 GC cores and zero Gracemont cores. What's the point of such a die if Intel is going all in with the hybrid approach? It could have been a 4+4 or 4+8 product to offer a smaller die for the lower end desktop SKUs, but they went with 6 big cores instead? Seems like a headscratcher.

 

[jpiniero]

I wouldn't have guessed that there's a smaller 6+0 die, which I interpret as being only 6 GC cores and zero Gracemont cores. What's the point of such a die if Intel is going all in with the hybrid approach? It could have been a 4+4 or 4+8 product to offer a smaller die for the lower end desktop SKUs, but they went with 6 big cores instead? Seems like a headscratcher.

Marketing. Didn't want lower end products to have more cores than higher end products yet didn't want to regress on the number of big cores on i5.

 

[naukkis]

I wouldn't have guessed that there's a smaller 6+0 die, which I interpret as being only 6 GC cores and zero Gracemont cores. What's the point of such a die if Intel is going all in with the hybrid approach? It could have been a 4+4 or 4+8 product to offer a smaller die for the lower end desktop SKUs, but they went with 6 big cores instead? Seems like a headscratcher.

For desktop 6 big cores is way better approach than less big cores plus small cores. Intel's choice is good, that 4+8 would lose to 5600x at games that scale beyond 4 threads. This time 6 big cores is sweet spot, most of users won't need more but many workloads prefer 6 cores over 4.

 

[blckgrffn]

Marketing. Didn't want lower end products to have more cores than higher end products yet didn't want to regress on the number of big cores on i5.

Appears to be lots of marketing going on here. Plus E cores only bring 1T, so to surpass/match the old 6C/12T model you'd need to go at least 4P+4E to get to core and thread counts that are "higher".

6P+4E would get you to 16T total which seems like a forward looking sweet spot (to be redundant).

Perhaps the I3's will be even more interesting? I don't know that I've seen any leaks on that front, but it's possible we could get 4P+2E or even 2P+4E - if 4C/8T was the old spec even a 6C/10T would appear to be a solid upgrade. You know, on paper.

 

[tomatosummit]

Marketing. Didn't want lower end products to have more cores than higher end products yet didn't want to regress on the number of big cores on i5.

I think it's silicon space and performance.
Largest mobile cpu would be a simimilar die size to the desktop cpu assuming 64eu are a similar size to two Pcores.
So they want a smaller die for the mainstream part and the volumes allow them to create a unique part for the performance required, the ultra mobile really wouldn't fit in where the 12400/500/600 cpus are going to be marketed, be it by offered cores or performance (games) given, although it's odd that they'd chose a small igpu cpu for the skus that will go into alll business and budget desktop computers for the next couple of years.

 

[blckgrffn]

I think it's silicon space and performance.
Largest mobile cpu would be a simimilar die size to the desktop cpu assuming 64eu are a similar size to two Pcores.
So they want a smaller die for the mainstream part and the volumes allow them to create a unique part for the performance required, the ultra mobile really wouldn't fit in where the 12400/500/600 cpus are going to be marketed, be it by offered cores or performance (games) given, although it's odd that they'd chose a small igpu cpu for the skus that will go into alll business and budget desktop computers for the next couple of years.

Is the i5 really a different die, though? If we already ascertained that, I missed it

Wait, I see that discussion right above. In that screen shot though... it looks like a 6+ some amount of little cores though doesn't it? What are the two colors of blue supposed to be for?

 

[tomatosummit]

Is the i5 really a different die, though? If we already ascertained that, I missed it

I didn't belive it initially but it appears to be real with the msi presentation. I presume intel leaves it out of marketing materials because it doesn't show off hybrid computing which is the big new thing.

I expect it to have other cheaper choices taken in manufacturing too. Probably has no solder and is tim on a thicker die with thinner ihs, need to eek out those margins for cpus that will cost ~$100.

 

[blckgrffn]

I didn't belive it initially but it appears to be real with the msi presentation. I presume intel leaves it out of marketing materials because it doesn't show off hybrid computing which is the big new thing.

I expect it to have other cheaper choices taken in manufacturing too. Probably has no solder and is tim on a thicker die with thinner ihs, need to eek out those margins for cpus that will cost ~$100.

I mean it makes some sense to have P only dies, but with the E cores being so small I would have thought they would keep them right down to the new pentium line which could be one P core and 4 E cores for some really whack use of a small die but "moar cores".

Like the octa core A55 phones, there is absolutely value in those small cores being marketed as upgrades.

Plus it seems like if Intel was serious about it the would have the heterogenous model be full stack.

 

[jpiniero]

Perhaps the I3's will be even more interesting? I don't know that I've seen any leaks on that front, but it's possible we could get 4P+2E or even 2P+4E - if 4C/8T was the old spec even a 6C/10T would appear to be a solid upgrade. You know, on paper.

i3 is 4+0, Pentium/Celeron is 2+0.

 

[blckgrffn]

i3 is 4+0, Pentium/Celeron is 2+0.

:/

Thanks for the info, sorry I didn't know.

I guess they gotta save something for the refresh. It seems like the number of PC's using the new P & E model is going to be a lot lower than I was expecting it to be.