Question Raptor Lake - Official Thread

[Hulk]

Since we already have the first Raptor Lake leak I'm thinking it should have it's own thread.
What do we know so far?
From Anandtech's Intel Process Roadmap articles from July:

Built on Intel 7 with upgraded FinFET
10-15% PPW (performance-per-watt)
Last non-tiled consumer CPU as Meteor Lake will be tiled

I'm guessing this will be a minor update to ADL with just a few microarchitecture changes to the cores. The larger change will be the new process refinement allowing 8+16 at the top of the stack.

Will it work with current z690 motherboards? If yes then that could be a major selling point for people to move to ADL rather than wait.

 

[pakotlar]

They did not say it, but we saw something similar on chilled water, and they did say "high end water" which wreaks of chilled water. Not impressed until I have more details AND see what Zen 4 does.

Would somebody who remembers what CPU was infamous for the "chilled water incident" remind us ?

Found it : https://www.tomshardware.com/news/intel-28-core-cpu-5ghz,37244.html

Quote from it:
"Intel's recent demonstration of a 28-core processor running at 5GHz has certainly stirred the pot here at Computex, particularly because the presentation appeared to imply this would be a shipping chip with a 5.0GHz stock speed. Unfortunately, it turns out that Intel overclocked the 28-core processor to such an extreme that it required a one-horsepower industrial water chiller. That means it took an incredibly expensive (not to mention extreme) setup to pull off the demo. You definitely won't find this type of setup on a normal desktop PC. "

Sound like a repeat ??????

I read high end water as custom loop or 360+mm aio. If it’s 6ghz on chilled water I would consider that exotic. Hope thats what this is, but if it’s chilled, agreed not so impressive.

 

[pakotlar]

Looks like we have some more data...

https://twitter.com/x/status/1553910359630970880

Mark, it looks like my interpretation is probably the right one; high end water here is 360mm aio and hits 5.9.

 

[pakotlar]

What? No we haven't. Raptor Lake isn't going to get +50% MT out of just adding 8 Gracemont cores. Even in a best-case scenario.

5.7GHz is pretty good but that power draw, woof. Hard to expect anything else though.

You sure about that?

 

[DrMrLordX]

You sure about that?

Yes, I'd have to see clocks but if the PL2 stays about the same for the 13900k then 8 e-cores won't do the trick. Unless Alder had a bug that got fixed.

 

[pakotlar]

Yes, I'd have to see clocks but if the PL2 stays about the same for the 13900k then 8 e-cores won't do the trick. Unless Alder had a bug that got fixed.

I just showsd you the leaked bench and circled where performance uplift was 1.51x 12900KF LOL.

 

[DrMrLordX]

I just showsd you the leaked bench and circled where performance uplift was 1.51x 12900KF LOL.

That's a leaked bench. It's basically meaningless. No proper reviews etc. You ought to know that the likelihood of Raptor Lake getting that much extra MT punch is incredibly small to non-existent given what's changed since Alder Lake. If it had 4P + 4e cores to the good I might believe it, but it doesn't.

If Alder Lake was artificially crippled in that one specific benchmark, maybe.

 

[Markfw]

Mark, it looks like my interpretation is probably the right one; high end water here is 360mm aio and hits 5.9.

5.9 is not 6.0. But I guess it could be.

Bottom line is these are not verified, as DrMrLordx says. And its not comparing to its real competitor, Zen4, so its all speculation, and lets wait until we see the two compared for real.

 

[deasd]

7z compress/decompress are heavily depends on memory performance.

The list already show R20 / geekbench to be 35%/23% uplift, in fact, so many situation would be around that, and it won't be like 7z which is 1.5x.

(well, I don't even care about how many E-cores being added anymore..... Intel has to immediately fix the compatibility issue of hybrid design)

 

[coercitiv]

You’ll see up to 50% better MT, in fact we’ve already seen better than that in some of the leaks.

So you really do swing whichever way the wind blows.

Can we just wait for reviews?

 

[DrMrLordX]

7z compress/decompress are heavily depends on memory performance.

The list already show R20 / geekbench to be 35%/23% uplift, in fact, so many situation would be around that, and it won't be like 7z which is 1.5x.

(well, I don't even care about how many E-cores being added anymore..... Intel has to immediately fix the compatibility issue of hybrid design)

You make a good point. It could be that Alder Lake's cache/memory was holding back 7-zip.

 

[coercitiv]

5.7GHz is pretty good but that power draw, woof. Hard to expect anything else though.

The really good part is the voltage, 1.38V for 5.7Ghz is quite decent relative to Alder Lake.

One other thing to keep in mind is that 103C max temp under the 360 AIO. In the video from the leaker we can see a very high temperature delta between the hottest core (103C) and the coldest core (85C). Assuming the leaker did a decent mount, it looks like future enthusiast Raptor owners will need to carefully plan their build with curved surface coolers or custom mounting brackets.

I'm cautiously optimistic for 13600K, it will come with conservative stock clocks and probably less affected by pricing change, might be a good part to pick up and overclock/tune. Alder Lake has good overclocking options, in the sense that one can setup a dynamic OC with only a few cores hitting max clocks and MT loads scaling back (both in clocks and voltage). It even uses linear voltage scaling for frequency steps with undefined VIDs, so efficiency loss is minimal.

Even the gaming tests make the 13600K the bigger winner over 13700K and probably 13900K, obviously in relative terms.

 

[biostud]

It’s similar to Ampere: hungry and fast. We’re crying over an extra 50 watts for up to 50% better MT, but are getting ready to buy 450-600W GPUs.

Well, AMD has said 50% increase in performance/watt for RDNA3 over RDNA2 and Intel is nowhere near that number with RKL, neither in ST or MT.

No one is forcing you to buy the most power demanding video card, if you don't want to.

 

[Zucker2k]

H

Dont think these new gaming benchmarks are something to write home about..

13900k = ~12900KS
13700k = ~12900k
13600k = ~12700K

With a ~20% price increase on the gen13 SKU's above we should be right back at todays status que with the same price/performance..
View attachment 65193

Did you consider the video card performance in your analysis? I think it could be a factor, unless current high-end CPUs are a bottleneck in games. I also think the real clue is in the minimums, where RPL is about 10% better than ADL on average.

 

[Abwx]

You make a good point. It could be that Alder Lake's cache/memory was holding back 7-zip.

Only in decompression, in compression the improvement is 13.47%...

 

[TESKATLIPOKA]

Looks like we have some more data...

https://twitter.com/x/status/1553910359630970880

Enabling SMT provides 51% higher score, WTF? The score without SMT has to be wrong If nothing else.

 

[mikk]

H

Did you consider the video card performance in your analysis? I think it could be a factor, unless current high-end CPUs are a bottleneck in games. I also think the real clue is in the minimums, where RPL is about 10% better than ADL on average.

I think he didn't, some games are clearly GPU limitied in this test. In some games the gains are much higher than the clock speed increase of 4%/8% suggests, like in Far Cry 6 with avg gains over 20%. This must be the L2 cache increase. And this isn't even tested with DDR5 according to spec, DDR5 on ADL-S runs OC.

 

[eek2121]

You sure about that?

View attachment 65199

You do realize that the ES chip was likely running faster memory, right?

The 8 additional cores add around 20% to multicore performance for applications that take advantage of it. Higher frequencies will add a bit more, and faster memory will also help.

 

[mikk]

You do realize that the ES chip was likely running faster memory, right?

The 8 additional cores add around 20% to multicore performance for applications that take advantage of it. Higher frequencies will add a bit more, and faster memory will also help.

In this test he did use DDR5-6400 on both CPUs. Alder Lake-S is oddly slow in 7zip decompressing, here is a test: https://www.guru3d.com/articles-pages/core-i9-12900k-review,13.html

I think they unlocked something on uarch level, maybe something was bugged and they couldn't fully enable it in time. Remember what Raichu and some other said:

https://twitter.com/x/status/1460567857947807748

 

[nicalandia]

Looks like we have some more data...

https://twitter.com/x/status/1553910359630970880

The e cores had to be disabled due to not working properly at that voltage. So 5.7 is not that impressive on a 13900K because it becomes a lower tier chip and the 7700X will beat it at ST And MT

 

[lobz]

In this test he did use DDR5-6400 on both CPUs. Alder Lake-S is oddly slow in 7zip decompressing, here is a test: https://www.guru3d.com/articles-pages/core-i9-12900k-review,13.html

I think they unlocked something on uarch level, maybe something was bugged and they couldn't fully enable it in time. Remember what Raichu and some other said:

https://twitter.com/x/status/1460567857947807748

While the story seems to be totally plausible, it remains an undeniable fact that ever since Zen, AMD is the undisputed decompression king. So whatever they may have found, they should be talking about it as finally catching up, instead of using the word 'counter'... Right now, generally speaking, they have absolutely nothing to counter the competing products with, regardless of what the opposing company's called.

 

[coercitiv]

The e cores had to be disabled due to not working properly at that voltage.

What isn't working properly at that voltage? Based on Alder Lake experience, the E cores don't have a problem at 1.38V, the real problem is thermals on P-core overclocks (and indirectly stability of the OC).

 

[moinmoin]

Right now, generally speaking, they have absolutely nothing to counter the competing products with, regardless of what the opposing company's called.

Ain't that the merciless truth.

(Well, they still do have the quantity. Of outdated products. That will still sell.)

 

[nicalandia]

What isn't working properly at that voltage? Based on Alder Lake experience, the E cores don't have a problem at 1.38V, the real problem is thermals on P-core overclocks (and indirectly stability of the OC).

8 vs 16. I cant find the Twitter post about the 16 core Gracemont not posting.

 

[Mopetar]

Don’t be absurd. What CPU have you seen hit 8-core 6ghz clocks without exotic cooling? This is a first.

The top end Bulldozer parts could do 6 GHz on water, and that was over a decade ago. The IPC was horrible though and they chewed through power just as bad as these newest Raptor Lake CPUs are rumored to, so no one really remembers them as it was basically NetBurst 2: Electric Boogaloo.

I will hand it to Intel for being able to squeeze that much more out of their node though. Once the do get one up and running they do a good job of refining it.

 

[coercitiv]

Once the do get one up and running they do a good job of refining it.

Last time Intel did a masterful job refining their node... they lost leadership in all x86 market segments. Not all kung fu is worth pursuing.

Then again, node refinement is now a form of risk mitigation for Intel, and from that perspective it has merits.