Review Intel Core i9-13900K and Core i5-13600K Review: Raptor Lake Beats Ryzen 7000

[Admin]

We put all three of Intel's new 13th-Generation Raptor Lake chips, the Core i9-13900K, i7-13700K, and i5-13600K, up against AMD's competing Ryzen 7000 processors.

Intel Core i9-13900K and Core i5-13600K Review: Raptor Lake Beats Ryzen 7000 : Read more

 

[BoredErica]

AMD seems to perform well in pcmark10's application startup test. Interesting

 

[Maebius]

Great review peeps.

Curious, will you test Raptor Lake with DDR4?

Also, aren't new-gen motherboard prices (for both AMD and Intel), kinda uhm, on the expensive side, even for entry level?

 

[johnnyboy5520]

The Raptor Lake 13900K and AMD 7950X are pretty much equivalent. As for gaming, the 7000 X3d chips will bury Raptor Lake. AMDs mistake was keeping the same core count on the lower end SKUs which allowed Intel an easy win in productivity. Gaming is pretty much game dependent. I don't think AMD will make that same mistake going forward. They've got the superior architecture.

 

[Why_Me]

Great review peeps.

Curious, will you test Raptor Lake with DDR4?

Also, aren't new-gen motherboard prices (for both AMD and Intel), kinda uhm, on the expensive side, even for entry level?

The B760 boards and locked Intel cpu's are due for release this January.

 

[TerryLaze]

The Raptor Lake 13900K and AMD 7950X are pretty much equivalent. As for gaming, the 7000 X3d chips will bury Raptor Lake. AMDs mistake was keeping the same core count on the lower end SKUs which allowed Intel an easy win in productivity. Gaming is pretty much game dependent. I don't think AMD will make that same mistake going forward. They've got the superior architecture.

It wasn't a mistake from AMD, it was all they could do.
There are only so many combinations of their CCX they can do.
To increase their core counts on the lower CPUs they would have to add an CCX to them and come up with a way to add a third CCX to their 7950x.
That would mean that they would lose a big chunk of their margins or increase their prices by that amount, both of these options would be very bad for AMD.
If TSMC comes up with a good node shrink then AMD can increase the amount of cores per CCX for the next round, but then again the amount of cores is already ridiculous and only appeals to a very small amount of people.

 

[PCWarrior]

The Raptor Lake 13900K and AMD 7950X are pretty much equivalent. As for gaming, the 7000 X3d chips will bury Raptor Lake. AMDs mistake was keeping the same core count on the lower end SKUs which allowed Intel an easy win in productivity. Gaming is pretty much game dependent. I don't think AMD will make that same mistake going forward. They've got the superior architecture.

Not sure how you can claim that AMD has the superior architecture. All the IPC tests show that Raptor Cove P-Cores are ahead of Zen 4 in IPC. Also Intel’s hybrid approach is proving highly effective when it comes to heavily multithreaded workloads. The only thing that AMD has an advantage over Intel is in efficiency but that is only thanks to TSMC N5 (which is a full node ahead of Intel 7) and not due to some AMD microarchitectural design advantage. If anything, the very fact that Intel offers the same or superior performance despite being a full node behind is due to having a better architecture. Much like Nvidia’s RTX 3060 (on Samsung 8nm, an equivalent to a TSMC '10nm') versus Intel’s A770 (on TSMC 6nm).

 

[The Historical Fidelity]

Intel artificially hiding L2 cache on the alderlake chips is pretty sus to me. Another reason I am done with Intel. Anti-consumer behavior all day.

 

[The Historical Fidelity]

Not sure how you can claim that AMD has the superior architecture. All the IPC tests show that Raptor Cove P-Cores are ahead of Zen 4 in IPC. Also Intel’s hybrid approach is proving highly effective when it comes to heavily multithreaded workloads. The only thing that AMD has an advantage over Intel is in efficiency but that is only thanks to TSMC N5 (which is a full node ahead of Intel 7) and not due to some AMD microarchitectural design advantage. If anything, the very fact that Intel offers the same or superior performance despite being a full node behind is due to having a better architecture. Much like Nvidia’s RTX 3060 (on Samsung 8nm, an equivalent to a TSMC '10nm') versus Intel’s A770 (on TSMC 6nm).

You are absolutely wrong about efficiency, Zen is a much more power and area efficient architecture. Except for when they decide to go for the highest clock speeds possible on the process node they manufacture on. TSMC’s node designs are notoriously voltage hungry on the top of their operating speeds. Bring down the top speed by 2-300 MHz and the zen 4 7950x sips power at a 105 watt TDP (AMD ECO mode) similar to zen 3 power.

 

[TerryLaze]

You are absolutely wrong about efficiency, Zen is a much more power and area efficient architecture. Except for when they decide to go for the highest clock speeds possible on the process node they manufacture on. TSMC’s node designs are notoriously voltage hungry on the top of their operating speeds. Bring down the top speed by 2-300 MHz and the zen 4 7950x sips power at a 105 watt TDP (AMD ECO mode) similar to zen 3 power.

On average the 7950x is about 3% faster in multi, (loses hard on single) and uses 20W less power.
It is more efficient, but much more is hugely hyperbolic, it's so little most people won't even bother to call it a difference.
And you can reduce power draw on the 13900k as well and you only lose performance in a very few applications.
You know which ones because all the "trusted reviewers" use exclusively those apps and nothing more.
https://www.pcwelt.de/article/1357334/core-i9-13900k-core-i5-13600k-im-test.html

 

[The Historical Fidelity]

On average the 7950x is about 3% faster in multi, (loses hard on single) and uses 20W less power.
It is more efficient, but much more is hugely hyperbolic, it's so little most people won't even bother to call it a difference.
And you can reduce power draw on the 13900k as well and you only lose performance in a very few applications.
You know which ones because all the "trusted reviewers" use exclusively those apps and nothing more.
https://www.pcwelt.de/article/1357334/core-i9-13900k-core-i5-13600k-im-test.html

That is actually still very high power usage even with MCE off and PL1/2 adherence.

In a thorough review (reference below) of power efficiency we see:

the 7950x at 142 watts beating the 13900k at 142 watts by 13.4% on the combined productivity benching suite graph,

the 7950x at 88 watts beating the 13900k at 88 watts by 14.3% on the combined productivity benching suite graph,

and the 7950x at 65 watts beating the 13900k at 65 watts by 11.7% on the combined productivity benching suite graph.

That’s an average 13.13% better performance efficiency at the same power levels for the 7950x compared to 13900k. That’s a big difference to say the least.

Energy efficiency: Ryzen 9 7950X >40% over Core i9-13900K even after TDP limitation

Comparing the energy efficiency of different products has several fundamental pitfalls. The first is more of a philosophical or principled nature and is a

tekdeeps.com

 

[TerryLaze]

That’s an average 13.13% better performance efficiency at the same power levels for the 7950x compared to 13900k. That’s a big difference to say the least.

It is...if you run your PC at 100% load 100% of the time 24/7, 365 days a year.

The site you show uses computerbase graphs so look at the single and idle loads.
The 7950x uses 11% more power in single core and 18% more in idle.
That's an average of 14.5% better performance efficiency at the same power levels for the 13900k compared to 7950x.
If you are not running a server but instead have a normal PC and do normal usage then that's a big difference to say the least.
https://www.computerbase.de/2022-10...nitt_leistungsaufnahme_in_anwendungen_ab_werk

 

[The Historical Fidelity]

It is...if you run your PC at 100% load 100% of the time 24/7, 365 days a year.

The site you show uses computerbase graphs so look at the single and idle loads.
The 7950x uses 11% more power in single core and 18% more in idle.
That's an average of 14.5% better performance efficiency at the same power levels for the 13900k compared to 7950x.
If you are not running a server but instead have a normal PC and do normal usage then that's a big difference to say the least.
https://www.computerbase.de/2022-10...nitt_leistungsaufnahme_in_anwendungen_ab_werk

You make valid points but I’d like to point out that I, in the previous post I made before the one you replied to, stated that TSMC process nodes are notorious for high power consumption at their frequency limit which AMD definitely hit the upper threshold of single core 5.7 ghz (notice how 7700x at 5.5 ghz used 23 less watts). I also think that they were using the AMD balanced windows power plan (AMD chipset drivers automatically make that default) which makes minimum processor state 80%. If I change to normal windows balanced power plan, my 5950x idles between 5.1-5.84 watts which is a 2x reduction from my processor on AMD balanced. Granted, since AMD chipset drivers default that profile you are technically correct, the best kind of correct lol!

But at the end of the day, AMD’s better performance efficiency in multi core is more important to enthusiasts, especially if you use windows balanced setting for 2x better idle power usage, and these days there is not a whole lot of single thread programs (I rarely see my 5950x at 5 ghz which is the frequency it hits with only one thread 100% active so AMD’s better performance efficiency at greater than 1 but less than max threads is a more common condition state that these modern processors will encounter.

 

[johnnyboy5520]

Not sure how you can claim that AMD has the superior architecture. All the IPC tests show that Raptor Cove P-Cores are ahead of Zen 4 in IPC. Also Intel’s hybrid approach is proving highly effective when it comes to heavily multithreaded workloads. The only thing that AMD has an advantage over Intel is in efficiency but that is only thanks to TSMC N5 (which is a full node ahead of Intel 7) and not due to some AMD microarchitectural design advantage. If anything, the very fact that Intel offers the same or superior performance despite being a full node behind is due to having a better architecture. Much like Nvidia’s RTX 3060 (on Samsung 8nm, an equivalent to a TSMC '10nm') versus Intel’s A770 (on TSMC 6nm).

Without those efficiency cores, Raptor Lake would lose in multithreaded workloads (16 core vs 16 core), so where is that supposed IPC gain helping Intel?

 

[AlistairAB]

This review is terrible. You used DDR5 6800 for Intel. Seriously. Who is going to spend that kind of money $500+ for DDR5 for a $350 CPU. You didn't get valid results at all. You're gonna make buyers think they are getting more speed than they actually are, especially all the gaming results are silly. The 5800X3D works fine with $50 DDR4-3600.

Do everything with DDR5 5600C36 and get back to me. Useless. I've never seen a release where so many people used different ram for the AMD and Intel reviews. I have no idea what my 13600K result would be with normal DDR5.

I actually had to find a youtuber, Jarrod's Tech, to find SOMEONE that used the same ram. Even HUB did this nonsense. Intel seeding people with free ram, clearly.

 

[TerryLaze]

This review is terrible. You used DDR5 6800 for Intel. Seriously. Who is going to spend that kind of money $500+ for DDR5 for a $350 CPU. You didn't get valid results at all. You're gonna make buyers think they are getting more speed than they actually are, especially all the gaming results are silly. The 5800X3D works fine with $50 DDR4-3600.

Do everything with DDR5 5600C36 and get back to me. Useless. I've never seen a release where so many people used different ram for the AMD and Intel reviews. I have no idea what my 13600K result would be with normal DDR5.

I actually had to find a youtuber, Jarrod's Tech, to find SOMEONE that used the same ram. Even HUB did this nonsense. Intel seeding people with free ram, clearly.

That ram speed is only for the overclock results and they also overclocked the ryzen ram to 6000, so they pushed both systems as high as either system can go.
Why would an OVERCLOCK test limit the OVERCLOCK of the faster system to the clocks the slower system is limited to?

And even on non overclock reviews why the heck would anybody test both systems at the same ram speed if the systems support different ram speeds?

How many people that buy a system are going to say "well the systems of the other company can only do that ramspeed so I'm going to limit my ram speed to that" , that's completely insane.

If you are not going to spend a lot on ram then just look at the FPS of the stock system that runs at 5600 ram.

 

[The Historical Fidelity]

That ram speed is only for the overclock results and they also overclocked the ryzen ram to 6000, so they pushed both systems as high as either system can go.
Why would an OVERCLOCK test limit the OVERCLOCK of the faster system to the clocks the slower system is limited to?

And even on non overclock reviews why the heck would anybody test both systems at the same ram speed if the systems support different ram speeds?

How many people that buy a system are going to say "well the systems of the other company can only do that ramspeed so I'm going to limit my ram speed to that" , that's completely insane.

If you are not going to spend a lot on ram then just look at the FPS of the stock system that runs at 5600 ram.

Because when you are testing the performance of different brands of CPU’s the scientific method requires controlling all other variables. I want to know how good the cpu is, not how well the particular cpu-memory combo is. CPU-memory combo reviews should be a completely different article or at least taken out of the main results graphs and given their own graphs specifying the reason why the separated results are not comparable to the main results graphs.

 

[TerryLaze]

Because when you are testing the performance of different brands of CPU’s the scientific method requires controlling all other variables. I want to know how good the cpu is, not how well the particular cpu-memory combo is.

They call that IPC, where they make everything the same including clocks, and there are reviews out there that include it.
The average end-user doesn't care the least about that, they care about what performance you can get out of the whole of the system you buy without voiding warranty and enough of them care also about the overclocked performance so they include that as well.

Also how well the memory controller works, how fast of ram it can use within its warranty, is part of how good the CPU is...it's integrated.

CPU-memory combo reviews should be a completely different article or at least taken out of the main results graphs and given their own graphs specifying the reason why the separated results are not comparable to the main results graphs.

They are clearly marked within the graph, so you can tell the difference easily.

 

[The Historical Fidelity]

They call that IPC, where they make everything the same including clocks, and there are reviews out there that include it.
The average end-user doesn't care the least about that, they care about what performance you can get out of the whole of the system you buy without voiding warranty and enough of them care also about the overclocked performance so they include that as well.

Also how well the memory controller works, how fast of ram it can use within its warranty, is part of how good the CPU is...it's integrated.


They are clearly marked within the graph, so you can tell the difference easily.

All due respect, because you do make fair points, but I don’t share your philosophy. By not normalizing all other factors, you cannot judge the true performance of the cpu. Yes you are correct that most people will only purchase what memory speed the cpu is rated for but you are forgetting that memory speed is only part of the spec. A ryzen 9 7950x with 5200 mhz speed cl30 dual rank memory beats the core i9 13900k in most games when it is bottlenecked by a 5600mhz cl42 single rank memory configuration. Since cpu manufacturers do not specify ddr timings or rank, and there are many different timing configurations and 2 rank types at each specified speed, the only proper way to test CPU’s against each other is to eliminate all variables outside of the CPU’s (internal factors like cpu microcode, normal boost algorithms, and stock power usage are okay since it is built in to the cpu itself and are stock) . In fact, I would surmise that the only true way to test CPU’s is to use the same memory modules and have them run at its JEDEC speed for both processors so that secondary and tertiary timings are equal as well. Unfortunately, until “Omni-motherboards” are invented so that each CPU can share the same socket and thus same vrm and bios tunings, it will never be 100% fair. But that does not mean we can’t try to level the playing field as much as we can.

 

[TerryLaze]

All due respect, because you do make fair points, but I don’t share your philosophy. By not normalizing all other factors, you cannot judge the true performance of the cpu.

Nobody, and by that I mean all of the people that read this article to make decisions on buying one or the other, care about "true" performance, they care about what their money will get them.

You can't get "true" performance comparison anyway unless you benchmark every single piece of software with every possible combination of workloads you can run in them...

A ryzen 9 7950x with 5200 mhz speed cl30 dual rank memory beats the core i9 13900k in most games when it is bottlenecked by a 5600mhz cl42 single rank memory configuration.

Just curious about this part, but where do you base this opinion on?! Any link?

Since cpu manufacturers do not specify ddr timings or rank, and there are many different timing configurations and 2 rank types at each specified speed,

They don't specify timings, unless that is part of jedec, I'm not that into ram, but they do have max speeds for number of ranks and channels used.
And things like expo have overclocking right in their name.
AMD EXPO™ Memory Overclocking Technology
https://www.amd.com/en/product/12151

Max Memory Speed
2x1R
DDR5-5200
2x2R
DDR5-5200
4x1R
DDR5-3600
4x2R
DDR5-3600

 

[The Historical Fidelity]

Nobody, and by that I mean all of the people that read this article to make decisions on buying one or the other, care about "true" performance, they care about what their money will get them.

You can't get "true" performance comparison anyway unless you benchmark every single piece of software with every possible combination of workloads you can run in them...

Just curious about this part, but where do you base this opinion on?! Any link?

They don't specify timings, unless that is part of jedec, I'm not that into ram, but they do have max speeds for number of ranks and channels used.
And things like expo have overclocking right in their name.
AMD EXPO™ Memory Overclocking Technology
https://www.amd.com/en/product/12151

You are right, but that doesn’t change the fact that it’s not proper. I’m a validation specialist by day so testing is kind of my thing. And rule # 1 is control all variables that affect the outcome of the system being evaluated. So yes Tom’s hardware’s testing methodology is for layman readers, but when judging parts analytically, it requires the methodology I described. Not saying Tom’s hardware is bad, I have been enjoying their content for over a decade now. Let’s just agree that we have different philosophies towards judging computer components. BTW, can’t find where I saw the disproportionate memory test results between 7950x and 13900k, but in my search I did find a review site that set the memory variable equal by using the same exact memory kit set to the same speed and timings, and the 7950x wins are equals the 13900k gaming performance while using less power.
https://www.digitaltrends.com/wp-co...tel-raptor-lake-far-cry-6.jpg?p=1&fit=400,267

https://www.digitaltrends.com/wp-content/uploads/2022/10/intel-raptor-lake-red-dead-redemption-2.jpg?p=1&fit=400%2C267

https://www.digitaltrends.com/wp-co...aptor-lake-cyberpunk-2077.jpg?p=1&fit=400,267

 

[TerryLaze]

BTW, can’t find where I saw the disproportionate memory test results between 7950x and 13900k, but in my search I did find a review site that set the memory variable equal by using the same exact memory kit set to the same speed and timings, and the 7950x wins are equals the 13900k gaming performance while using less power.
https://www.digitaltrends.com/wp-content/uploads/2022/10/intel-raptor-lake-far-cry-6.jpg?p=1&fit=400,267

https://www.digitaltrends.com/wp-content/uploads/2022/10/intel-raptor-lake-red-dead-redemption-2.jpg?p=1&fit=400%2C267

https://www.digitaltrends.com/wp-content/uploads/2022/10/intel-raptor-lake-cyberpunk-2077.jpg?p=1&fit=400,267

First of all overclocked which you don't seem to like from what you said until now for comparing CPUs.

Out of only 6 games intel wins one by any margin and ryzen wins one by any margin, and we are talking about 16 and 11 FPS here in games that run far above 100FPS, the other 4 are too close even when comparing the 13600 to them.
Also there are no power numbers associated with the FPS numbers so how do you know that ryzen uses less power for those FPS?!
The site only states the max power limit, which is 230W compared to 253W, and doesn't have a single power measurement bench.

 

[The Historical Fidelity]

First of all overclocked which you don't seem to like from what you said until now for comparing CPUs.

Out of only 6 games intel wins one by any margin and ryzen wins one by any margin, and we are talking about 16 and 11 FPS here in games that run far above 100FPS, the other 4 are too close even when comparing the 13600 to them.
Also there are no power numbers associated with the FPS numbers so how do you know that ryzen uses less power for those FPS?!
The site only states the max power limit, which is 230W compared to 253W, and doesn't have a single power measurement bench.

Okay I tried to be nice but you seem to be incapable of polite conversation. Look it up. There is this thing called multiple sources and I’m under no obligation to write a dissertation with works cited for you. Bless your heart.

 

[bit_user]

I recall that Alder Lake had a lower DDR5 frequency limit for simply using it on a motherboard with 4 DIMM slots, even if you populated only 2 of them (1 DIMM Per Channel).

I didn't see any mention of that, in this review. Also, the MSI Z790 board tested seems to have 4 DIMM slots. Does that mean Raptor Lake has no such restriction?

 

[digitalgriffin]

AMD got razzed pretty hard in the free chat about the vapor chamber cooler issue and high cost of motherboards. That was spot on. They also got razzed about the amount of execs they put on stage. A big snore fest.

The message was "we aren't interested in consumers any more." Funny because it's CES. CONSUMER electronics show.

The xilinx for mobile chips was interesting. But I'm a bit confused why it was implemented for mobile. AI workloads are specialty and heavy duty. They require a workstation class card that can handle heat/power. Mobile is power limited. So why?

The debate between which processor is better (AMD vs Intel) is a bit bogus. I'm going to be honest, Intel gives you better bang for your buck once a similar dollar platform is out together.

However the AMD all performance cores, compared to intels hybrid approach of p and e isn't an apples to apples comparison. If Intel made 100% all P cores and tried to match AMDs core count it would be a heat power disaster.

But not every process requires P cores. There are a few like video transcodes, AI workloads, some db ops, and compilers that will use all P cores to the max. But that is a very small percentage of us. For the average user Intel is the much better buy this round (even with extra heat)