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.

 

[poke01]

My undervolted 13900K reaches 2360ish points at 38 W package power when running CB23 at 6.0 GHz

Is that for single core?

 

[Hulk]

But we don't learn anything about power/efficiency in this table, but just how different single-core frequencies unsurprisingly affect the results. And there are no "running at same power (253 W)" results in this table, as it is a single-core test.

My undervolted 13900K reaches 2360ish points at 38 W package power when running CB23 at 6.0 GHz, that's less than 1.35 V for a (real) single (not dual) core. And my CPU is just a "good enough" bin, not a stellar one.

That's a nice CPU. When you are running CB R23 ST at 6Ghz what is the average Vcore during the run as reported by HWinfo? Also what is the temp and your cooling solution?

 

[igor_kavinski]

My undervolted 13900K reaches 2360ish points at 38 W package power when running CB23 at 6.0 GHz, that's less than 1.35 V for a (real) single (not dual) core.

I vaguely recall that Intel clamped down on undervolting in some microcode update? Are you sticking with some older BIOS to undervolt your CPU or using Linux?

 

[coercitiv]

I vaguely recall that Intel clamped down on undervolting in some microcode update? Are you sticking with some older BIOS to undervolt your CPU or using Linux?

It's optional, mobo makers wouldn't have any other way.

 

[igor_kavinski]

It's optional, mobo makers wouldn't have any other way.

You are right. This guy got his 14700K to 67C without that much perf reduction:

 

[Timur Born]

That's a nice CPU. When you are running CB R23 ST at 6Ghz what is the average Vcore during the run as reported by HWinfo? Also what is the temp and your cooling solution?

It's "good enough" but not stellar, maybe the equivalent of a SP 105 CPU according to another user, so even somewhat "average". Biscuit rating started at 89.9 and dropped to 86.7 after a BIOS update when the KS hit the market. This is a no/low noise build and my priority goal is undervolting without *realworld* performance loss. The per-core OCs are only put on top of that and usually in a manner that does not decrease efficiency too much (VID/power increase vs. performance increase). These CPUs basically come overclocked out of the box with the point of diminishing returns long overstepped.

With my undervoltage settings it can run 6.0 GHz for non AVX offset invoking load (like CB23) only on its core no. 2. For dual-core load on cores 2/3 it uses 5.9 GHz instead, so a 14900K is 1.7% faster for dual-core load. On its own the CPU would ask for about 1.43 V for single-core CB23 at 6.0 GHz on core 2 (with AC/DC LL 1/1), with my undervoltage it uses slightly less than 1.35 V. At LLC 0.620 mOhm + AC LL 0.180 mOhm my VF curve is set to -55 mV below 57x and -75 mV starting at 57x and above (aka 1-4 core non AVX offset load, 1-2 core AVX offset load). At such low power usage (38 W) and temperature (about 62°C) my Arctic LF2 420 mm runs its fans at 500 rpm and the pump at 40% PWM (which corresponds to 75-80% pump speed or so). My pump only jumps to 100% above 65°C while the fans step up much slower.

My CPU currently uses the following ratios for non AVX offset invoking load (Turbo ratios based on number of concurrently active cores): 60/59/57/57/56/56/55/55 + 44/44/43/43
AVX offset ratios are more complex on my CPU (see earlier post), but for single-core load it's 59x and all-core load 55x except for cores 0+1 at 54x. I test single/low core stability with concurrent E-cores load to increase the stress (E-cores do not increased Vcore for P-cores, but P-cores increase Vcore for E-cores).

CB23 runs 40k+ scores right at the 253 W power limit on my CPU, fluctuating slightly based on current CPU temps. It could be dropped further towards 240 W, but my stability demands include being Folding@home (FAH) stable - aka as stable as stock - even when P95/OCCT/Yc and the like stay stable at the power-limit. And more aggressive undervolting can cause very seldom WHEAs with some FAH work units well below the power limit.

 

[Timur Born]

I vaguely recall that Intel clamped down on undervolting in some microcode update? Are you sticking with some older BIOS to undervolt your CPU or using Linux?

I use a combination of LLC + AC LL + negative offset (via VF curve).

- Most current non beta Gigabyte BIOS (F9). I need to use XTU to workaround a AVX offset bug (also hits at offset 0) that would force me to boot twice every morning, but since GB support does not understand the issue to begin with this will never reach the BIOS developer team (along multiple other "needs two boots" bugs in their BIOS).

- Windows 11.

 

[Panino Manino]

Incredible gains.
I wonder if AMD will have the same problems to fix.

 

[eek2121]

6.2ghz is absurd.

Don't get me wrong, I'd choose the 7800X3D any day over Raptor Lake. I don't want a space heater near me for only an insignificant performance advantage, especially if I have to rely on a lucky RAM overclock. In fact, I'm in the process of building a small form factor PC with the 7800X3D. I've gathered some parts, and I'm waiting for Black Friday to get the rest.

In general, I agree with you. In practice? My office is getting to be pretty cold in the morning and I often wonder if using an Intel chip instead of my AMD one would have helped with that. 🤣

Are you sure that it's only CPU limited and not I/O bound? Easy way to test it is to create a pretty large RAM drive, put the files you work on there and benchmark a typical session. If the program settings have the option to specify a temp drive that it can use for temporary storage, you can point it to a folder on the RAM drive for that too. Even the fastest NVMe SSD is no match for a RAM drive.

All you are doing here is creating a memory bandwidth bottleneck.

They COULD have made some improvements, like more cache or better tuning of the IMC or heck, a 40 thread monster powered by more E-cores. But they chose not to, probably coz the decision was made at the last minute. Now they gonna try to survive the onslaught of Zen 5 with unattractive, boring desktop silicon, depending primarily on the stupidity of the common buyer, especially the corporate hardware acquisition "experts". Intel is probably the luckiest corporation in the history of the US. They should have been hanging by a thin thread by now but it seems they have too many thin threads keeping them hanging.

I do wish Intel and AMD would work on improving the IMC…and IO in general. Give me 40 PCIE lanes!

 

[coercitiv]

In general, I agree with you. In practice? My office is getting to be pretty cold in the morning and I often wonder if using an Intel chip instead of my AMD one would have helped with that. 🤣

Depends on what you're doing in the morning. If you start your day by drinking coffee and going through emails/documentation/forums and everything else that is lightly threaded, then the Intel system would actually lower your average room temp by a fraction of degree. You need to go straight into compilation/simulation/rendering to make your room cosier with Intel.

That being said, average energy and/or temperature is only one side of the coin, the other being peak values during gaming (such as what @Racan was probably hinting at). When both CPU and GPU are blasting during a gaming session, every 50W matter for short term thermal comfort and system temps.

 

[Hulk]

It's "good enough" but not stellar, maybe the equivalent of a SP 105 CPU according to another user, so even somewhat "average". Biscuit rating started at 89.9 and dropped to 86.7 after a BIOS update when the KS hit the market. This is a no/low noise build and my priority goal is undervolting without *realworld* performance loss. The per-core OCs are only put on top of that and usually in a manner that does not decrease efficiency too much (VID/power increase vs. performance increase). These CPUs basically come overclocked out of the box with the point of diminishing returns long overstepped.

With my undervoltage settings it can run 6.0 GHz for non AVX offset invoking load (like CB23) only on its core no. 2. For dual-core load on cores 2/3 it uses 5.9 GHz instead, so a 14900K is 1.7% faster for dual-core load. On its own the CPU would ask for about 1.43 V for single-core CB23 at 6.0 GHz on core 2 (with AC/DC LL 1/1), with my undervoltage it uses slightly less than 1.35 V. At LLC 0.620 mOhm + AC LL 0.180 mOhm my VF curve is set to -55 mV below 57x and -75 mV starting at 57x and above (aka 1-4 core non AVX offset load, 1-2 core AVX offset load). At such low power usage (38 W) and temperature (about 62°C) my Arctic LF2 420 mm runs its fans at 500 rpm and the pump at 40% PWM (which corresponds to 75-80% pump speed or so). My pump only jumps to 100% above 65°C while the fans step up much slower.

My CPU currently uses the following ratios for non AVX offset invoking load (Turbo ratios based on number of concurrently active cores): 60/59/57/57/56/56/55/55 + 44/44/43/43
AVX offset ratios are more complex on my CPU (see earlier post), but for single-core load it's 59x and all-core load 55x except for cores 0+1 at 54x. I test single/low core stability with concurrent E-cores load to increase the stress (E-cores do not increased Vcore for P-cores, but P-cores increase Vcore for E-cores).

CB23 runs 40k+ scores right at the 253 W power limit on my CPU, fluctuating slightly based on current CPU temps. It could be dropped further towards 240 W, but my stability demands include being Folding@home (FAH) stable - aka as stable as stock - even when P95/OCCT/Yc and the like stay stable at the power-limit. And more aggressive undervolting can cause very seldom WHEAs with some FAH work units well below the power limit.

Nice work. You really have that CPU set up well. Thanks for the information. I'm currently running a slight undervolt on my 13600K with cores running between 5.4 and 5.0Ghz. That variation is due to temps, I tweak clocks to get core temps the same.

 

[DAPUNISHER]

HUB tests APO

TLDW -
It works.
In game can have 20%+ increase. That's outstanding. Faster than any other gaming approach e.g. disabling e-cores and or HT.
There is a nice efficiency gain.
Only 2 games to date with support; both of which are snoozers from a relevancy standpoint.
Hypothesis is ecores being utilized better.
Intel provided generic, craptastic responses as to why it isn't included with 12&13. And what future game support to expect.
Owners need to raise hell with Intel to have any chance of Intel caving in, and including your Alder and Raptor CPUs.

 

[igor_kavinski]

Owners need to raise hell with Intel to have any chance of Intel caving in, and including your Alder and Raptor CPUs.

And also to include more AAA game support, preferably day one support for future games.

 

[DAPUNISHER]

And also to include more AAA game support, preferably day one support for future games.

Steve was speculating how many games they went through to cherry pick? these 2. The Total War games are PC exclusives that always run great on Intel. You would think those are perfect candidates to highlight a feature like this. Must not do much for them?

If it is indeed artificially limited to 14th, that's right out of the Nvidia playbook.

 

[Timur Born]

Nice work. You really have that CPU set up well. Thanks for the information. I'm currently running a slight undervolt on my 13600K with cores running between 5.4 and 5.0Ghz. That variation is due to temps, I tweak clocks to get core temps the same.

If you regularly run into power/temp/current limits then I suggest to use Ring downbin. Contrary to popular belief it only kicks in when a limit is hit and then downbins the Ring "as needed" instead of the full 300 MHz that the tooltip suggests. In return it allows much higher core clocks at the limit, because the Ring needs voltage for high ratios and directly increase core VIDs (instead of "Uncore").

 

[Timur Born]

HUB tests APO

TLDW -
It works.
In game can have 20%+ increase. That's outstanding. Faster than any other gaming approach e.g. disabling e-cores and or HT.
There is a nice efficiency gain.
Only 2 games to date with support; both of which are snoozers from a relevancy standpoint.
Hypothesis is ecores being utilized better.
Intel provided generic, craptastic responses as to why it isn't included with 12&13. And what future game support to expect.
Owners need to raise hell with Intel to have any chance of Intel caving in, and including your Alder and Raptor CPUs.

I consider any game not getting along with HT or E cores to be bugged in that regard.

 

[Heartbreaker]

Right now APO is pretty much an experiment. If they actually get this working with a pile of games. Intel really have no excuse to not at least support 13th gen, which is essentially the same exact part. There is absolutely no excuse for not supporting 13th gen, and probably 12th gen as well.

 

[Timur Born]

Steve was speculating how many games they went through to cherry pick? these 2. The Total War games are PC exclusives that always run great on Intel. You would think those are perfect candidates to highlight a feature like this. Must not do much for them?

If it is indeed artificially limited to 14th, that's right out of the Nvidia playbook.

Total War is bottlenecked by its Aeon old engine, likely further hampered by its Lua integration and general lack of GPU optimization. It does not create enough multi-threaded CPU load to benefit from using E cores on top of P cores, because it cannot even saturate P cores enough.

 

[Hulk]

Right now APO is pretty much an experiment. If they actually get this working with a pile of games. Intel really have no excuse to not at least support 13th gen, which is essentially the same exact part. There is absolutely no excuse for not supporting 13th gen, and probably 12th gen as well.

Game performance is so important now we're seeing Intel tuning their rigs for specific games. I guess the eventual outcome is they will start modifying drivers for Intel CPU's! Or more likely providing "incentives" for game developers that make them run better on Intel hardware.

 

[DAPUNISHER]

That certainly tracks if Steve's hypothesis is correct. Which means this is far from the killer app needed to stimulate sales. Unless we suddenly see dozens of newer titles get boosted in a similar fashion. These products have a rather limited shelf life, they have to get cracking.

Or more likely providing "incentives" for game developers that make them run better on Intel hardware.

They have been doing this for decades. I also would not be surprised if windows 12 breathes new life into hybrid CPUs by having better scheduling and optimizations built in. Wintel isn't going to suddenly stop being a thing.

 

[Timur Born]

Nice work. You really have that CPU set up well. Thanks for the information. I'm currently running a slight undervolt on my 13600K with cores running between 5.4 and 5.0Ghz. That variation is due to temps, I tweak clocks to get core temps the same.

Try to prioritize non adjacent cores, preferably P cores 0+1, 4+5 and E core clusters 2+3 on your CPU. If this isn't possible due to P cores 2+3 being the best/most stable cores then at least try the E core clusters. You do this kind of prioritization by setting higher per core ratios (and then control final clocks by Turbo ratios). CPPC will make sure that Windows will schedule load on these cores first.

On my CPU P cores 2+3 are the best cores, so my priorities are 2+3 -> 6+7 -> 4+5 -> 0+1. And then E core clusters 3 (coolest) -> 4 -> 1/2.

This makes sure that there are idle cores in between busy cores, which makes a *huge* difference in temperature compared to running the same load on adjacent cores (hotspots).

 

[Timur Born]

Is that for single core?

Of course, all-core scores for 13900Ks are between 35-40.5K at stock settings (depending on power/temp limits). Mine does 40k+ at the 253 W power limit after being undervolted, before it was up to 290 W and other reported something like 320 W or so (likely depending on BIOS defaults).

 

[Mopetar]

There is absolutely no excuse for not supporting 13th gen, and probably 12th gen as well.

How many people would even upgrade from 12/13th gen anyways? There's just not enough gain in general to warrant it and you'd be better off saving for 15th gen which will have the support (or an improved version of it) baked in.

I don't have a good reason to believe that something changed under the hood to enable this, but if I had 12th gen Intel I'd be pretty happy that I got most of the performance that the platform was ever going to see out of the gate.

 

[Hulk]

Try to prioritize non adjacent cores, preferably P cores 0+1, 4+5 and E core clusters 2+3 on your CPU. If this isn't possible due to P cores 2+3 being the best/most stable cores then at least try the E core clusters. You do this kind of prioritization by setting higher per core ratios (and then control final clocks by Turbo ratios). CPPC will make sure that Windows will schedule load on these cores first.

On my CPU P cores 2+3 are the best cores, so my priorities are 2+3 -> 6+7 -> 4+5 -> 0+1. And then E core clusters 3 (coolest) -> 4 -> 1/2.

This makes sure that there are idle cores in between busy cores, which makes a *huge* difference in temperature compared to running the same load on adjacent cores (hotspots).

Here are the avg core temps and frequencies as I have them right now during a CB R24 MT run. Cores 0 and 1 are so much cooler than the rest. I realize there is thermal conduction among these cores. Wish I knew where they were located on the die.

Core 0 5.5Ghz 72C (This seems to be my best core)
Core 1 5.4Ghz 73C
Core 2 5.1Ghz 75C
Core 3 4.8Ghz 77C (This is the other hot one)
Core 4 5.2Ghz 75C
Core 5 4.7GHz 78C (This one I call "The Furnace Room")

 

[Timur Born]

Cores 0+1 are cooler, because they are at the edge of the cores area, they only have neighbors on one side. Look for die shots of the 13900K, that should give you an idea.