Intel "Arrow Lake-S" Desktop Processor Projected 6%-21% Faster than "Raptor Lake-S"

[btarunr]

Intel's future-generation "Arrow Lake-S" desktop processor is already being sampled internally, and to some of the company's closest industry partners, and some of the first performance projections of the processor, comparing it with the current "Raptor Lake-S" (Core i9-13900K), have surfaced, and upcoming "Raptor Lake Refresh" desktop processor (probably the i9-14900K), have surfaced. First, while the "Raptor Lake Refresh" family sees core-count increases across the board for Core i3, Core i5, and Core i7 brand extensions, the 14th Gen Core i9 series is widely expected to be a damp squib compared to the current i9-13900 series, and it shows in the performance projection graphs, where the supposed-i9-14900K is barely 0% to 3% faster, probably on account of slightly higher clock speeds (100-300 MHz).

The "Arrow Lake-S" processor in these graphs has a core-configuration of 8P+16E. Since this is a projection, it does not reflect the final core-configuration of "Arrow Lake-S," but is a guideline on what performance increase to expect versus "Raptor Lake," assuming the same core-configuration and power limits. All said and done, "Arrow Lake-S" is projected to offer a performance increase ranging between 6% in the worst case, to 21% in the best-case benchmark, compared to the current i9-13900K, assuming an identical core-config and power-limits. The CPU benchmarks in the projection span the SPECrate2017 suite, CrossMark, SYSmark 25, WebXPRT 4, Chrome Speedometer 2.1, and Geekbench 5.4.5 ST and MT.



One area where "Arrow Lake-S" is expected to offer a performance leap is with its integrated graphics. Based on the Xe-LPG graphics architecture (DirectX 12 Ultimate capable), and armed with 8 Xe cores (128 EU), the iGPU is projected to offer a massive 240% graphics performance uplift over the current Xe-LP based iGPU of the "Raptor Lake-S" that has 32 EU.

With Intel expected to call 2023 a wrap with the "Raptor Lake Refresh" series planned for Q4-2023 on the existing LGA1700 platform, all eyes are on what Intel does in 2024. The company's subsequent desktop platform will introduce the new Socket LGA1851, and require a new motherboard. It's unclear if the platform will debut with a "Meteor Lake-S" as the microarchitecture's compute tile tops out at a 6P+16E core-count. "Arrow Lake-S" surfaced on leaked roadmap slides with a mid-2024 mass-production commencement timeline, which should put product launches some time in the second half of 2024.



View at TechPowerUp Main Site | Source

 

[john_]

The iGPU performance gains might force AMD to start integrating Infinity Cache in it's iGPUs.

As for the CPU part, until Intel fixes it's manufacturing, we shouldn't expect much. When they do, we might see a huge jump in performance with a significantly lower power consumption.

 

[JustBenching]

The iGPU performance gains might force AMD to start integrating Infinity Cache in it's iGPUs.

As for the CPU part, until Intel fixes it's manufacturing, we shouldn't expect much. When they do, we might see a huge jump in performance with a significantly lower power consumption.

Intel is already very efficient. Just don't run them at 4096 watts.

It's amd that in fact needs to do something about efficiency, in the lower mid range segment. The new and almighty r7 7700x and 7800x 3d is less efficient than the 12700f which is like 2 years old already, as tested by hwunboxed.

 

[Daven]

Doesn’t look like anything compelling coming from Intel in the foreseeable future. Good news is that you can build that new PC now and be set for years to come. No need to wait for that massive leap in performance because its not coming.

 

[N/A]

What do you mean, 12900K is 13700K is 14600K 8+8, But
But if you want to get the most out of your motherboard that you pay for dearly, s1700 stops at 14th a dead end. 1851 should be 15-17th 20 18 16A cpu tile. 6 Ghz at lower power.

 

[Wirko]

Doesn’t look like anything compelling coming from Intel in the foreseeable future. Good news is that you can build that new PC now and be set for years to come. No need to wait for that massive leap in performance because its not coming.

What's more, the interfaces already in use today seem very much future-proof: PCIe 5.0, Thunderbolt 4, USB 40 Gbit/s, DDR5-6400+. Speeds will continue to go up of course but not on the cheap; runing multi-gigahertz signals over cheap cables and PCB tracks is becoming an impossibility.

 

[Daven]

What's more, the interfaces already in use today seem very much future-proof: PCIe 5.0, Thunderbolt 4, USB 40 Gbit/s, DDR5-6400+. Speeds will continue to go up of course but not on the cheap; runing multi-gigahertz signals over cheap cables and PCB tracks is becoming an impossibility.

Good point. Core counts and I/O are pretty much set for the next few years.

 

[Hyderz]

if rumours are correct there are alot to be liked for the 14th gen,
those wanting a budget cpu i3 has 6 cores with ht ...
i5 14600k now bumped up to 8performance cores along with 8 ecores ...
the i7 and i9 might not be so stellar but regardless lets have a look at benchmarks

the cpu coming out after 14th gen should be exciting

 

[Daven]

if rumours are correct there are alot to be liked for the 14th gen,
those wanting a budget cpu i3 has 6 cores with ht ...
i5 14600k now bumped up to 8performance cores along with 8 ecores ...
the i7 and i9 might not be so stellar but regardless lets have a look at benchmarks

the cpu coming out after 14th gen should be exciting

No refreshes are ever exciting. Their very existence is defined by manufacturing delays of what a company is really trying to build.

The 14th series as well as any 12th and 13th series are good upgrades for customers who buy Intel only and still use Skylake architecture. For anyone already on socket 1700 or AM5 there is no reason to upgrade to 14th gen. Probably the same goes for upcoming Zen 5 as well.

Edit: For example, a gamer on a budget buys a Core i5 12400F six P-core for $167 MSRP two years ago. Upgrade choices are Core i5 13400F six P-core plus four E-core for $196 MSRP or Core i3 14300F six P-core for $150ish. No IPC increases, a few hundred MHz clock and loss of E-cores over 13th gen if the gamer goes 14th gen. Wow exciting. /sarcasm

 

[Dristun]

Edit: For example, a gamer on a budget buys a Core i5 12400F six P-core for $167 MSRP two years ago. Upgrade choices are Core i5 13400F six P-core plus four E-core for $196 MSRP or Core i3 14300F six P-core for $150ish. No IPC increases, a few hundred MHz clock and loss of E-cores over 13th gen if the gamer goes 14th gen. Wow exciting. /sarcasm

This is a point a lot of people keep making that I don't understand at all. People on budget are precisely the kind of people who don't upgrade every two years.
I'd argue this goes exactly the other way around: AMD is good at attracting DIY enthusiast class who want to stick every new gen in the same motherboard for 5 years for measly +10% upticks, average joe on a budget IMHO couldn't care less.

 

[Daven]

This is a point a lot of people keep making that I don't understand at all. People on budget are precisely the kind of people who don't upgrade every two years.
I'd argue this goes exactly the other way around: AMD is good at attracting DIY enthusiast class who want to stick every new gen in the same motherboard for 5 years for measly +10% upticks, average joe on a budget IMHO couldn't care less.

Many here are trying to get excited by Intel having three “generations” on one socket. This makes them feel they can upgrade the CPU without upgrading the platform…a rarity for Intel users. In reality, Meteor Lake on the desktop was cancelled and Arrow Lake will not come out until H2 2024. Intel likes yearly generation updates; henceforth the refresh.

All three “generations” use the same chip architecture with minor changes to the core config, clock speeds and cache with no IPC increases. Most understand that there is no reason to upgrade within the socket since this is just a cash grab by Intel. What we endusers really want are CPU generations on a two year cadence with MAJOR architecture changes each generation.

 

[Space Lynx]

Raptor Lake-S is the refreshed Raptor Lake coming in October, which probably will get 5% gains, so 6% gains over that... 11% total over current Raptor Lake is what we can expect in Fall 2024? If that is all Intel can bring to the table than 3nm 8800x3d cache is going to rip them to pieces as far as gamers are concerned, but good luck to them.

 

[chrcoluk]

No refreshes are ever exciting. Their very existence is defined by manufacturing delays of what a company is really trying to build.

The 14th series as well as any 12th and 13th series are good upgrades for customers who buy Intel only and still use Skylake architecture. For anyone already on socket 1700 or AM5 there is no reason to upgrade to 14th gen. Probably the same goes for upcoming Zen 5 as well.

Edit: For example, a gamer on a budget buys a Core i5 12400F six P-core for $167 MSRP two years ago. Upgrade choices are Core i5 13400F six P-core plus four E-core for $196 MSRP or Core i3 14300F six P-core for $150ish. No IPC increases, a few hundred MHz clock and loss of E-cores over 13th gen if the gamer goes 14th gen. Wow exciting. /sarcasm

Upgrading to a new gen when you already have current or previous gen generally isnt sensible for most people anyway, most people will be keeping their existing platform for 3 years minimal with some closer to a decade (note those who are only just upgrading now from sandy bridge).

These CPU's I expect to be of interest to people on coffee lake or older on intel or maybe people on older AM4 who for whatever reason dont want to upgrade to Zen 3 or AM5.

Thats one area where reviewers can be lacking as they dont always show older gens (which are the likely upgrade path) in their tables, instead just comparing most recent 2-3 gens. I assume because they dont want to retest on older systems and consider old data not valid (due to diff os/drivers etc.).

However there is still a sane upgrade path on LGA 1700 which would be of course maybe from something like an i3 to i5/7/9 (upwards on product stack) if you want a P core/clock bump, or perhaps want the e-cores for encoding/compiling purposes.

 

[TheinsanegamerN]

Upgrading to a new gen when you already have current or previous gen generally isnt sensible for most people anyway, most people will be keeping their existing platform for 3 years minimal with some closer to a decade (note those who are only just upgrading now from sandy bridge).

These CPU's I expect to be of interest to people on coffee lake or older on intel or maybe people on older AM4 who for whatever reason dont want to upgrade to Zen 3 or AM5.

The biggest attraction will be new budget builds, likely from people still rocking i3/i5 builds from haswell or skylake. This gen brings 6 p core i3s and 8 p core i5s to the table, thats a major move so long as intel doesnt bumble the price.

Thats one area where reviewers can be lacking as they dont always show older gens (which are the likely upgrade path) in their tables, instead just comparing most recent 2-3 gens. I assume because they dont want to retest on older systems and consider old data not valid (due to diff os/drivers etc.).

It's time. Rebenching takes hours, and every additional CPU multiplies that time. Doing just two-three gens from both AMD and intel already takes days.

 

[Darmok N Jalad]

The "Arrow Lake-S" processor in these graphs has a core-configuration of 8P+16E. Since this is a projection, it does not reflect the final core-configuration of "Arrow Lake-S," but is a guideline on what performance increase to expect versus "Raptor Lake," assuming the same core-configuration and power limits. All said and done, "Arrow Lake-S" is projected to offer a performance increase ranging between 6% in the worst case, to 21% in the best-case benchmark, compared to the current i9-13900K, assuming an identical core-config and power-limits. The CPU benchmarks in the projection span the SPECrate2017 suite, CrossMark, SYSmark 25, WebXPRT 4, Chrome Speedometer 2.1, and Geekbench 5.4.5 ST and MT.

So that still leaves uncertainty around the rumor that ArL might top out at 6P cores? I suspect that extra IPC is going to come at a P-core size increase, and Intel might not be too crazy about packing 8 of them on a die, especially if they aren't using the desired node to get there. Queue additional E Cores for multithreaded wins. The other catch would be to keep the same power limits. If we do see a drop to 6 P cores, but power consumption remains the same, it could get interesting to see the net result. Curious if Intel really boosted L3 sizes as a way to counter 3DV cache?

The iGPU performance gains might force AMD to start integrating Infinity Cache in it's iGPUs.

As for the CPU part, until Intel fixes it's manufacturing, we shouldn't expect much. When they do, we might see a huge jump in performance with a significantly lower power consumption.

Perhaps. Eventually you get to a place where iGPU performance just needs to be supplemented by an actual GPU though. I do like the idea of infinity cache to give it a boost, but it might not be enough to overcome the limits of packing that all into one place. Even then the 7840U and Z1 look like a really nice offering already, and drivers are still going to be a differentiator.

 

[Daven]

The biggest attraction will be new budget builds, likely from people still rocking i3/i5 builds from haswell or skylake. This gen brings 6 p core i3s and 8 p core i5s to the table, thats a major move so long as intel doesnt bumble the price.

It's time. Rebenching takes hours, and every additional CPU multiplies that time. Doing just two-three gens from both AMD and intel already takes days.

That’s not the budget argument being made by fevgatos et al. because upgrading the whole platform is much more expensive. They are saying budget users who are already on socket 1700 with 12th gen have a GREAT upgrade path to 14th gen because the platform is the same. This argument is laughable at best since the performance difference is negligible for a budget user.

Of course if you are going from Core i3 to Core i7 or higher within or across any generation on the same platform, you get a big performance uplift but this is also true for AMD users going from Ryzen 3 to Ryzen 7 or higher. However, this is not a budget argument but maybe applies to you if you started a higher paying job sometime in between. Again this has always been true for any platform and not unique to the current socket 1700 situation.

 

[chrcoluk]

The biggest attraction will be new budget builds, likely from people still rocking i3/i5 builds from haswell or skylake. This gen brings 6 p core i3s and 8 p core i5s to the table, thats a major move so long as intel doesnt bumble the price.

It's time. Rebenching takes hours, and every additional CPU multiplies that time. Doing just two-three gens from both AMD and intel already takes days.

Yeah, I think I would add old test results to table with an asterisk, the asterick disclaimer being not matching software so results wont be fully accurate. I would consider that better than no data.

 

[Wye]

Projected by who? Do you have a source or was this just shower thoughts?

 

[Daven]

Projected by who? Do you have a source or was this just shower thoughts?

To whom are you replying?

Edit: oh you were replying to the article. The source link is Igor’s labs. And they just state that they were given the projections by an unnamed third party. I guess they want to stay anonymous to prevent getting sued/fired for leaking pre-release info. This is typical for tech news and media in general.

Edit 2: Upon further reading, the source link (Igor Labs) implies that the projections came from Intel directly.

 

[JustBenching]

That’s not the budget argument being made by fevgatos et al. because upgrading the whole platform is much more expensive. They are saying budget users who are already on socket 1700 with 12th gen have a GREAT upgrade path to 14th gen because the platform is the same. This argument is laughable at best since the performance difference is negligible for a budget user.

Of course if you are going from Core i3 to Core i7 or higher within or across any generation on the same platform, you get a big performance uplift but this is also true for AMD users going from Ryzen 3 to Ryzen 7 or higher. However, this is not a budget argument but maybe applies to you if you started a higher paying job sometime in between. Again this has always been true for any platform and not unique to the current socket 1700 situation.

The performance difference is negligible on amd as well though. We had what, a 15% increase in 3 generations? From r7 3700x (300$) to r5 7600x (300$) the mt performance difference is tiny.

 

[TumbleGeorge]

The performance difference is negligible on amd as well though. We had what, a 15% increase in 3 generations? From r7 3700x (300$) to r5 7600x (300$) the mt performance difference is tiny.

Unfair comparison. Because number of cores and this:

The U.S. Inflation Calculator measures the dollar's buying power over time.

Inflation Calculator​
	If in		(enter year) 2019
	I purchased an item for $300
	then in		(enter year) 2023
	that same item would cost: $358.03
	Cumulative rate of inflation:

 

[evernessince]

Intel is already very efficient. Just don't run them at 4096 watts.

It's amd that in fact needs to do something about efficiency, in the lower mid range segment. The new and almighty r7 7700x and 7800x 3d is less efficient than the 12700f which is like 2 years old already, as tested by hwunboxed.

The 7800X3D consumes less than half the power of the 12700K while putting out more performance per watt: https://www.techpowerup.com/review/amd-ryzen-7-7800x3d/23.html

And that's a generous comparison, Intel's competitor for the 7800X3D in the 13900KS consumes many times more power.

Second, HWUB stated nothing of the sort. They publish their text version on TechSpot: https://www.techspot.com/review/2657-amd-ryzen-7800x3d/

They didn't even test the 12700f or 12700K, you seem to have fabricated numbers.

Intel efficiency is good if you don't crank up the frequency but the Zen 4 architecture scales much better in that regard loosing less performance at much lower power envelopes, particularly the X3D cache models. The problem for Intel is that they need that frequency on the current architecture to remain competitive in the high end.

 

[JustBenching]

Unfair comparison. Because number of cores and this:

The number of cores is irrelevant though, maybe they shouldn't have decreased them but that's an amd problem

The inflation between 2019 and 2023 (actually the 7600x was released in 2022 but its fine) is basically 20%. If you add that 20% on the 15%, that's just 3%. So sure, after 3 generations we got a 18% performance increased for the money. That's 4.5% a year, lol

The 7800X3D consumes less than half the power of the 12700K while putting out more performance per watt: https://www.techpowerup.com/review/amd-ryzen-7-7800x3d/23.html

And that's a generous comparison, Intel's competitor for the 7800X3D in the 13900KS consumes many times more power.

Second, HWUB stated nothing of the sort. They publish their text version on TechSpot: https://www.techspot.com/review/2657-amd-ryzen-7800x3d/

They didn't even test the 12700f or 12700K, you seem to have fabricated numbers.

Intel efficiency is good if you don't crank up the frequency but the Zen 4 architecture scales much better in that regard loosing less performance at much lower power envelopes, particularly the X3D cache models. The problem for Intel is that they need that frequency on the current architecture to remain competitive in the high end.

Techspot did in fact test the 12700.


https://www.techspot.com/review/2391-intel-core-i7-12700/

At 65w with an intel stock cooler and an el cheapo b660 board it scored 16.017 points. That's 246 pts / watt, and the system power draw was at 136 watts.

The 7800x 3d at 77w scored 18395 points (as per your link), so that's 238 pts / watt, while using a 200€ ΑΙΟ and a 500-600€ motherboard. System power draw was a 207w

So, the 7800x 3d is in fact less efficient than a locked and terribly binned i7 from soon to be 3 generations ago, all the while consuming 50% more power (136 vs 207), with a much more expensive cooler and motherboard. That freaking amd efficiency, you gotta love it.

AMD is just not efficient due to the lack of cores - especially at the low end. They only seem efficient cause Intel uses higher power limits in the reviewed models. If they start reviewing the T and non k versions, oh boy, its's going to be a rude awakening

 

[AnotherReader]

The inflation between 2019 and 2023 (actually the 7600x was released in 2022 but its fine) is basically 20%. If you add that 20% on the 15%, that's just 3%. So sure, after 3 generations we got a 18% performance increased for the money. That's 4.5% a year, lol

Is there a reason that you didn't pick the 7700 instead of the 7600X? The 7700's MSRP is $330 which is the same as that of the 3700X.

 

[Dyatlov A]

So the single core performance improvment will be the 6% and that 20%+ from E Cores in Chinebench? That is not much…