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.

 

[Timur Born]

As some people seem to think that AC LL is a "fixed" offset:

 

[Henry swagger]

Intel Raptor Lake-S Refresh for Desktop Coming in Fall 2023, Emerald Rapids-X Workstation by December | Hardware Times

www.hardwaretimes.com

Raptor refresh with higher clocks.. improved mode ? 🤔

 

[Markfw]

Intel Raptor Lake-S Refresh for Desktop Coming in Fall 2023, Emerald Rapids-X Workstation by December | Hardware Times

www.hardwaretimes.com

Raptor refresh with higher clocks.. improved mode ? 🤔

Looks like skylake ++++ to me, nothing here, move along....

 

[Geddagod]

Should be fine, imo, because it would still only be competing with Zen 4 X3D. Unless Zen 5 comes to desktop early 2024, Intel should be fine since they have ARL out by the end of 2024.

 

[DrMrLordX]

Intel should be fine since they have ARL out by the end of 2024.

God help them if they don't. They haven't even gotten Intel 4 out yet, and they're expected to iterate through Intel 4 and Intel 3 to launch Intel 20a Arrow Lake in late 2024? More-likely they'll abandon 20a and move completely to TSMC N3. And then it's over for their fab side.

 

[Det0x]

*mixed up the threads*

 

[Hulk]

So as you know my 13900K has been restarting over 200W and Intel is reimbursing me for the cost of the CPU.

I thought about it long and hard and just got back from my local Microcenter with a 13600K. At $250 the price to performance ratio was just too good. Especially knowing that I'm generally not getting that "deep" into the E's anyway. I almost went for the 13700K but $55 each for those additional P's I couldn't justify. It's just a "hold me over" CPU until I see what Raptor refresh brings to the party.

 

[Hitman928]

So as you know my 13900K has been restarting over 200W and Intel is reimbursing me for the cost of the CPU.

I thought about it long and hard and just got back from my local Microcenter with a 13600K. At $250 the price to performance ratio was just too good. Especially knowing that I'm generally not getting that "deep" into the E's anyway. I almost went for the 13700K but $55 each for those additional P's I couldn't justify. It's just a "hold me over" CPU until I see what Raptor refresh brings to the party.

Let us know how the new CPU works out for you. Hopefully this one is fully functional

 

[IEC]

So as you know my 13900K has been restarting over 200W and Intel is reimbursing me for the cost of the CPU.

I thought about it long and hard and just got back from my local Microcenter with a 13600K. At $250 the price to performance ratio was just too good. Especially knowing that I'm generally not getting that "deep" into the E's anyway. I almost went for the 13700K but $55 each for those additional P's I couldn't justify. It's just a "hold me over" CPU until I see what Raptor refresh brings to the party.

13600K is a solid chip and at $250 is a great value.

 

[Hulk]

Let us know how the new CPU works out for you. Hopefully this one is fully functional

Going to install it this weekend and starting playing around with it. Of course I'll report back with details.

Funny the last time by 13900K popped a restart. It was yesterday. I had PL1 set at 225W, power at 130%, XMPI, all else stock.

I go to compress some Winrar files and BAM! Restart. Really? WinRar?

Nice thing about the 13600K is that I can just run it XMP I all else stock and I should get all core 5.1GHz with sane temps on my 280mm AIO.

I'll probably just try to push the P's to 5.2, E's to 4.0 and bump the GPU to 2GHz. I need the extra GPU compute for DxOmark PureRaw.

 

[Kocicak]

Going to install it this weekend and starting playing around with it. Of course I'll report back with details.

Funny the last time by 13900K popped a restart. It was yesterday. I had PL1 set at 225W, power at 130%, XMPI, all else stock.

I go to compress some Winrar files and BAM! Restart. Really? WinRar?

It does not matter what you run, higher power draw causes higher voltage drop and if your chip for some reason consumed its voltage safety margin, it simply needed a little bit more voltage.

I asked you TWICE if I remember correctly to try to give the CPU a positive voltage offset to see if it stabilizes the chip and report the outcome to my degradation thread, to no avail.

 

[igor_kavinski]

I asked you TWICE if I remember correctly to try to give the CPU a positive voltage offset to see if it stabilizes the chip and report the outcome to my degradation thread, to no avail.

Better if you buy the chip from him and torture it yourself

 

[Hulk]

It does not matter what you run, higher power draw causes higher voltage drop and if your chip for some reason consumed its voltage safety margin, it simply needed a little bit more voltage.

I asked you TWICE if I remember correctly to try to give the CPU a positive voltage offset to see if it stabilizes the chip and report the outcome to my degradation thread, to no avail.

I'm sorry I didn't reply. I told Intel I never overclocked the chip or added voltage and I want them to receive the chip for investigation with that being a true statement.

 

[Hulk]

Okay finally got some time to install the new CPU.

First, in order to take the edge off of 13900K withdraw I set my current CPU to 6+8 last week. While it was still ramping up to frequencies beyond stock 13600K at least I'd get used to the reduction in cores right?! Ha ha

Anyway. Here we go. Popped in the 13600K set XMP I and set power to 250W, which is beyond what a 13600K will pull I believe, under air anyway.

Cinebench R23 MT
13900K stock but set to 6+8
25,500, 196W, 5.3/4.3 frequencies, hitting my temp throttle limit set at 95C.
I don't remember the VID/Vcore but I know this chip always "asked for and received" over 1.3 volts.

13600K stock
24,000 CB score
144W, 5.1/3.9 frequencies (of course)
VID 1.15, Vcore 1.18
Max temp 71C

This chip is looking good out of the box.
Let's see it's using 74% of the power and getting 94% of the performance. Or looking at it another way 26% less power and 6% less performance.
Plus temps and Vcore are way down.

Anandtech Forums Handbrake Test
13900K set 6+8
160.27 seconds, 11.27fps running 5.3/4.3 and again bumping the 95C max throttling temp. Pulling 200W.

13600K
169.89 seconds, 10.63fps (faster than my old 12700K on the same board)
144W and 71C max temp, same as CB result.
Again about 6% reduction in performance and 28% less power.

I am really happy about the power draw and temps. I'm definitely okay with the performance of this CPU even at stock. It will definitely tide me over until I see what the Raptor Refresh brings to the table. I'm pretty sure I got a very low SP rated 13900K and it degraded a bit from there. I have to say Intel was very easy to deal with and even told me I can keep the old box, which is cool because the 13900K packaging looks nice on the shelf.

Okay so I know what you are all thinking. Set the 13600K to 5.3/4.3 like the 13900K and retest. Yes, yes I'll get there. Let me just enjoy not running the nuke plant under my desk for a bit.

First thing I'm going to do is try and run the E's up a bit. Since we all know there is one voltage plane for P's and E's I'm thinking the current voltage for 5.1P's might be good enough for 4.2E's, which should give a little performance boost without breaking the power/temp bank.

Forgot to mention anyone know what the following on my CPU means? I remember you used to be able to tell the week/batch of production years ago.
13900K
SRMBH
X236F565

13600K
SRMBD
X2471760

 

[Hulk]

After messing around with the 13600K a bit it, while unreasonable for many reasons a 6+24 Raptor Lake Refresh would be quite a part. If the new process could simply run such a part at 5.5/4.3 with reasonable power draw it would do 44,000 in Cinebench.

 

[Hulk]

I don't see how Anandtech got their 13600K to pull 238 watts. They got the same Cinebench score as me so obviously they didn't overclock but I'm seeing 144 watts in CB and even max heat/stress Prime 95 is 174 watts. Anyone know what app they use for max power?

 

[Hulk]

The performance gains for setting my 13600K at 5.3/4.2 aren't worth it for me. CB from 24,000 to 25,000, .4fps in the Handbrake test for another 35 Watts and higher temps. Max Vcore at 5.3 is 1.225 vs 1.18 so not a big difference there. But for that tiny increment of performance it's not worth it. I like 5.1/3.9 with 144 Watts max and mid 70's temps under long term loads.

Intel got it right with the stock frequencies on this chip in my opinion.

Holy shotgun Hulk posts! Just noticed how many I put in a row.

 

[igor_kavinski]

Holy shotgun Hulk posts! Just noticed how many I put in a row.

All of them useful and informative. Thank you

BoomerD, you made the right choice going with 13600K

 

[Timur Born]

I had to do more homework after correcting DC LL to match LLC. This increased "measured" package power by about 10 W, so I had to increase my effort to lower it further down again.





Testing for Folding@home stability isn't fun at such low voltages. Everything else is much easier to stabilize, because the load is more predictable.

 

[Hulk]

The performance of this 13600K got me thinking about the binning articles that were done at Igor's Lab. I can really appreciate the time and effort it took to log all of that data but I don't feel as though they took enough time and thought to present the data effectively. Here's one chart I made that I think sums up their efforts pretty well. I only included P cores because the rest is really quite irrelevant.

I would ask your input on the following forum members.

There is a significant difference in SP rating when moving from 13600/13700 to 13900. Why?

My theory is this. If a chip has defects that do not allow it to operate at 8+16, which would make it immediately binned down to a 13600 series or 13700 series part, then those defects are also exhibiting in the silicon in less severe way, which cause the part to have a lower SP rating.

If this was not the case then wouldn't we expect to see more outlier 13600/13700 dies that SP rate higher?

Or put another way, most if not all fully functional 8+16 dies by virtue of the fact that they don't have core killing defects are in general very defect free, resulting in higher SP values than the 6+8 and 8+8 parts.

It just seems strange that only 3% of the 13600/13700's are SP>105 yet 91% of the 13900 dies are SP>105.

And why didn't Igor's lab take some 13900's with SP rating of 95, 105, 115, and 127 and run them in Cinebench at stock setting and record score, Vcore, power, and temps on the same rig? That would have been really interesting. I mean they installed the CPU it would have only taken a minute to run the test. Unless of course they installed without cooling and just booted into the BIOS and back out. Even then you just need to fully install 4 processors to run the tests.

 

[Kocicak]

... If a chip has defects that do not allow it to operate at 8+16 ... then those defects are also exhibiting in the silicon in less severe way, which cause the part to have a lower SP rating.

There is no reason for this assumption. You can literally have one tiny defect in one of the 16 E cores which causes just occasional error in some specific task, while overall this chip is excellent, and this chip can never become any 13900 CPU. It will most likely become 13700K.

BTW there is an interview on Youtube somewhere with a guy from Intel responsible for sorting dies into CPUs and what I got from the interview is that they SOMEHOW make what they need to deliver, and they want to use if possible almost every piece of silicone they make. That is all.

I do not believe they have any rigorous sorting algorithms they follow like a bible script and that would explain variability of the quality of the different pieces of one CPU model.

 

[Hulk]

There is no reason for this assumption. You can literally have one tiny defect in one of the 16 E cores which causes just occasional error in some specific task, while overall this chip is excellent, and this chip can never become any 13900 CPU. It will most likely become 13700K.

BTW there is an interview on Youtube somewhere with a guy from Intel responsible for sorting dies into CPUs and what I got from the interview is that they SOMEHOW make what they need to deliver, and they want to use if possible almost every piece of silicone they make. That is all.

I do not believe they have any rigorous sorting algorithms they follow like a bible script and that would explain variability of the quality of the different pieces of one CPU model.

Okay so what's your theory at why the 13900 series SP ratings are so much higher then the 13600/13700?

If it's just a defect that kills a core or two then should we expect to see average SP rating across all K's closer than they are in these samples? Especially considering the best KS chips aren't even included in the 13900K/KF data.

I don't understand your 3rd paragraph at all about the "bible" or whatever? I do think the sorting is rigorous. Processors are tested and there is a algorithm that sends them to different bins. Mathematical algorithms by definition are "rigorous."

Do you think they just kind of hold the die up close to their eyes, take a good gander at it, hike up their pants, and exclaim, "By golly I gots me a 13900KS!"

 

[Timur Born]

59/59/58/58/57/57/56/56 below 60°C and -1 bin above that is what I am currently testing within my undervoltage limits.

 

[Kocicak]

Okay so what's your theory at why the 13900 series SP ratings are so much higher then the 13600/13700?
....
I do think the sorting is rigorous. Processors are tested and there is a algorithm that sends them to different bins.
....

SP rating is a number that one MB manufacturer puts in a few selected motherboards and is generally irrelevant.

Testing and binning may be rigorous.
How CPU models are made from different bins is not.

 

[Timur Born]

My Gigabyte board offers a "Biscuit" rating, with a (supposed) maximum of 100.

- My CPU originally was measured as 89-91 (slightly changes with every reboot).
- When I undervolted the CPU the Biscuit rating went "out of range".
- The first BIOS after the KS hit the market at first changed my own CPU's Biscuit rating to 86.x, but then it went down to 81.x without without undervolting.
- The latest (withdrawn) BIOS version lists the CPU as 86.8.

So overall the Biscuit rating is rather useless and Gigabyte seems to have dropped the rating of my K by 3-4 points to make room for KS CPUs towards the (supposed) maximum of 100.