Intel Core i7-13700K

[Ferrum Master]

Wait for the reviews soon of the upcoming 7800X3D and its versions before pulling the trigger as they might be something special.

I'm normally an Intel man, but these are looking really good and you don't have to put up with e cores.

I haven't poked around more than virtualbox when I need need to run shady large manufacturer service tools that are infested with spying crap code, otherwise they don't run.

But does 3D cache improve virtualization? Not sure. IMHO only brute rule applies there... how many spare cores you donate.

 

[qubit]

I haven't poked around more than virtualbox when I need need to run shady large manufacturer service tools that are infested with spying crap code, otherwise they don't run.

But does 3D cache improve virtualization? Not sure. IMHO only brute rule applies there... how many spare cores you donate.

I have no idea if it does. However, these CPUs are supposed to deliver performance uplifts in all areas, so I'd look out for this in reviews.

Are you thinking of upgrading your system anytime soon?

 

[dgianstefani]

I haven't poked around more than virtualbox when I need need to run shady large manufacturer service tools that are infested with spying crap code, otherwise they don't run.

But does 3D cache improve virtualization? Not sure. IMHO only brute rule applies there... how many spare cores you donate.

Cache won't do anything for virtualization.

It will be the same situation as 5800X3D, lower clocks, with higher barrier to heat transfer, but with 3x cache for workloads and games that can benefit.

I'd say the 13900K, 13600K, 5800X3D and 7950X are the only CPUs worth buying at the moment for serious users.

Intel wins both value and single core performance. AMD wins multicore/efficiency.

The Zen 4 X3D have a gimmick on the 7900/7950X3D where one die is frequency with no additional cache and one die has 3D cache. For those that like to whine about P/E cores, this is much worse. The 7800X3D is the only worthwhile CPU in that stack IMO.

AMD has better support for "future proofing" as much as that applies to most people (not many people), and their platform is more ambitious, but a lot more buggy and early access. Everyone on Zen 4 is basically a beta tester. Intel has faster cores with tighter latency, and the E cores will only get better, but it can be a little hard to cool if you don't know what you're doing.

The other advantage to Intel Raptor Lake 13th gen is the memory controller is significantly better than Zen 4. You can reach 7000+, probably 8000 MHz on some motherboards with a 13900K. The architecture also scales very well with memory frequency.

 

[AusWolf]

Cache won't do anything for virtualization.

It will be the same situation as 5800X3D, lower clocks, with higher barrier to heat transfer, but with 3x cache for workloads and games that can benefit.

I'd say the 13900K, 13600K, 5800X3D and 7950X are the only CPUs worth buying at the moment for serious users.

Intel wins both value and single core performance. AMD wins multicore/efficiency.

The Zen 4 X3D have a gimmick on the 7900/7950X3D where one die is frequency with no additional cache and one die has 3D cache. For those that like to whine about P/E cores, this is much worse. The 7800X3D is the only worthwhile CPU in that stack IMO.

AMD has better support for "future proofing" as much as that applies to most people (not many people), and their platform is more ambitious, but a lot more buggy and early access. Everyone on Zen 4 is basically a beta tester. Intel has faster cores with tighter latency, and the E cores will only get better, but it can be a little hard to cool if you don't know what you're doing.

The other advantage to Intel Raptor Lake 13th gen is the memory controller is significantly better than Zen 4. You can reach 7000+, probably 8000 MHz on some motherboards with a 13900K. The architecture also scales very well with memory frequency.

Except that Zen 4 doesn't even need a better memory controller because the architecture isn't sensitive to memory speed. Also, the most annoying bugs have been ironed out already with BIOS updates still rolling in from all vendors. I'd say the 13600K or the 7700 non-X are equally good options for a gaming or lightweight work PC. Other than these, I agree.

 

[qubit]

The Zen 4 X3D have a gimmick on the 7900/7950X3D where one die is frequency with no additional cache and one die has 3D cache. For those that like to whine about P/E cores, this is much worse. The 7800X3D is the only worthwhile CPU in that stack IMO.

Well, I'm one of those "whiners" lol. Good info there and the rest of the post, I'll keep it in mind when reading the reviews.

It's that stability which concerns me the most. My current ancient 2700K system just works and that's how I expect the new one to work too.

 

[dgianstefani]

Except that Zen 4 doesn't even need a better memory controller because the architecture isn't sensitive to memory speed. Also, the most annoying bugs have been ironed out already with BIOS updates still rolling in from all vendors. I'd say the 13600K or the 7700 non-X are equally good options for a gaming or lightweight work PC. Other than these, I agree.

Zen architecture is literally tied to memory speed. It's just the infinity fabric can't clock higher than 2000 MHz, and the latency penalty from going past 3:2 isn't worth it.

Saying it isn't sensitive to memory speed is just... wrong. What's more accurate is that it's unable to take advantage of faster memory because it doesn't have the capability past 6200 MHz or so.

 

[AusWolf]

Zen architecture is literally tied to memory speed. It's just the infinity fabric can't clock higher than 2000 MHz, and the latency penalty from going past 3:2 isn't worth it.

Zen 4 is different (link). It doesn't scale much (if at all) beyond 5200-5600 MHz, even with UCLK at 1:1.

 

[dgianstefani]

What's more, since the entire CPU is tied to memory speed, and high FPS gaming is tied to memory latency, Zen architectures are basically obligated to run at the highest memory speed they can. Even so, despite the IPC advantage Zen 3 for example had over Comet lake, they were slower for high FPS gaming until the 5800X3D. This is due to the fact even the best tuned Zen 3 chip won't go below 50-55ns memory latency, whereas a good Coffee Lake with high frequency B die can go into the low 40s.

 

[AusWolf]

What's more, since the entire CPU is tied to memory speed, and high FPS gaming is tied to memory latency, Zen architectures are basically obligated to run at the highest memory speed they can. Even so, despite the IPC advantage Zen 3 for example had over Comet lake, they were slower for high FPS gaming until the 5800X3D. This is due to the fact even the best tuned Zen 3 chip won't go below 50-55ns memory latency, whereas a good Coffee Lake with high frequency B die can go into the low 40s.

I don't disagree on Zen 3, but Zen 4 is different, trust me. The real world application benefit beyond 5200-5600 MHz is zero. Latencies, transfer speeds and artificial benchmarks don't interest me.

 

[dgianstefani]

I don't disagree on Zen 3, but Zen 4 is different, trust me. The real world application benefit beyond 5200-5600 MHz is zero. Latencies, transfer speeds and artificial benchmarks don't interest me.

Silicon Power XPOWER Zenith RGB DDR5-5600 CL40 2x 16 GB Review

Silicon Power brings the XPOWER Zenith line to DDR5! Equipped with a 5600 MT/s XMP profile, using modest CL40-40-40-76 timings, Silicon Power has a memory kit that is great starting point for both Intel and AMD platforms. Follow along as we test this memory and see how it stacks up the competition!

www.techpowerup.com

Corsair Vengeance RGB DDR5-6000 CL30 (AMD Expo) 2x 16 GB Review

Corsair's Vengeance RGB takes the AMD DDR5 platform to a sweet spot. Using the new EXPO 6000 MT/s profile that's optimized for Zen 4 and tight timings of 30-36-36-76, this kit is designed to fit right into the Corsair ecosystem to provide a great end-to-end user experience.

www.techpowerup.com

It scales just fine.

If you don't do high FPS gaming or actually leverage memory bandwidth and latency in applications then of course you won't be interested, but that is irrelevant to the facts.

 

[AusWolf]

Silicon Power XPOWER Zenith RGB DDR5-5600 CL40 2x 16 GB Review

Silicon Power brings the XPOWER Zenith line to DDR5! Equipped with a 5600 MT/s XMP profile, using modest CL40-40-40-76 timings, Silicon Power has a memory kit that is great starting point for both Intel and AMD platforms. Follow along as we test this memory and see how it stacks up the competition!

www.techpowerup.com

Corsair Vengeance RGB DDR5-6000 CL30 (AMD Expo) 2x 16 GB Review

Corsair's Vengeance RGB takes the AMD DDR5 platform to a sweet spot. Using the new EXPO 6000 MT/s profile that's optimized for Zen 4 and tight timings of 30-36-36-76, this kit is designed to fit right into the Corsair ecosystem to provide a great end-to-end user experience.

www.techpowerup.com

It scales just fine.

If you don't do high FPS gaming or actually leverage memory bandwidth and latency in applications then of course you won't be interested, but that is irrelevant to the facts.

We're talking about a couple of FPS which is within margin of error (1-5% in most cases), not to mention totally undetectable for the human eye. The benefit in applications (not benchmarks) is even smaller.

 

[dgianstefani]

We're talking about a couple of FPS which is within margin of error (1-5% in most cases), not to mention totally undetectable for the human eye. The benefit in applications (not benchmarks) is even smaller.

Try telling that to someone who works with memory sensitive workloads or people seeking locked 236 FPS for aim consistency.

And the difference is more like 25 FPS from 5600 to 6000, and that's with a 3080ti. Faster GPU would put more stress on the CPU and memory differences would become more pronounced.

Remember, just because you can't notice something or don't use relevant software, doesn't mean that's the case for others.

You use a 24" 60 Hz monitor so discussing the perceivable differences in high frequency memory/FPS is kind of pointless since you don't have the hardware to actually display any of those frames.

 

[AusWolf]

Try telling that to someone who works with memory sensitive workloads or people seeking locked 236 FPS for aim consistency.

And the difference is more like 25 FPS from 5600 to 6000, and that's with a 3080ti. Faster GPU would put more stress on the CPU and memory differences would become more pronounced.
View attachment 279061

Remember, just because you can't notice something or don't use relevant software, doesn't mean that's the case for others.

What do you mean "memory intensive workloads"? The only work test I see a difference in is file compression.

As for gaming, I highly doubt anyone can feel 25 FPS difference above 300. Unless you're the CS:GO world champion, it's kind of a first world problem.

You use a 24" 60 Hz monitor so discussing the perceivable differences in high frequency memory/FPS is kind of pointless since you don't have the hardware to actually display any of those frames.

But I have a phone with a 90 Hz display that can be turned down to 60. Even there I have to focus very hard to see any difference.

 

[dgianstefani]

What do you mean "memory intensive workloads"? The only work test I see a difference in is file compression.

As for gaming, I highly doubt anyone can feel 25 FPS difference above 300. Unless you're the CS:GO world champion, it's kind of a first world problem.


But I have a phone with a 90 Hz display that can be turned down to 60. Even there I have to focus very hard to see any difference.

That's the point. You have no experience of the types of work where memory bandwidth and latency matters, or high FPS gaming, or even have the hardware to display those FPS (GPU and screen). Doesn't lend much credence to your opinion, or doubts thereof. Competitive gamers want their FPS to be consistent, which means having a 1/0.1% low FPS that is higher than the maximum refresh rate of their screen, since framerates are linked to the muscle memory of aiming. It's literally better to run at locked 144 FPS on a 144 Hz monitor than it is to have a 240 Hz monitor and have framerates that fluctuate between 100-300.

For people who do dataset based work, e.g. scientific computing, Bloomberg trading (currently recommend 32 GB of RAM and a Intel 11th gen/Zen 3 or better CPU with a passmark above 2500), IBM SPSS, the types of huge Excel files some corporations work with, certain Adobe software for content creation, etc. memory performance makes a big difference. Typically more channels is better than higher frequency for the simple reason that high frequency is hard to maintain with high capacity sticks, but with DDR5 you can get 32/64 even 128 GB dual DIMM kits that run at high speeds 5000 MHz plus (technically 4x32 in two 2x32 bit DIMMS, this isn't conventional quad channel which would be 4x64 bit). Lots of workstation owners are put off by the cost of threadripper pro/xeon workstations so they buy the higher core count consumer grade CPUs and use high bandwidth and capacity memory sticks.

Phoronix does good testing of workstation and server software under Linux, which has fewer limitations than Windows when it comes to scaling in professional software. Here's an example of Code compilation, AI, rendering, database software, deep learning etc. https://www.phoronix.com/review/ddr5-epyc-9004-genoa. Channels are bandwidth, so channel scaling and bandwidth scaling is the same thing - you'd see similar results from moving from 3000-6000 MHz memory all else considered.

Actually hardcore overclocking (YT channel) goes into how most reviewers and consumers don't really understand the memory subsystem. Subtimings can be tuned to offer significant performance differences at the same frequency, because it's not actually frequency that is the only limiter, it's absolute latency, which is a formula of the data rate (MT/s) and CAS latency (CL). Remember when the CPU is waiting for data from the cache/memory, it's stalled, absolute latency is what is important to get as low as possible, so your CPU can actually stretch it's legs. This is precisely why the X3D chips removed much of the limitations of Zen 3 which was memory starved in high FPS gaming (huge fast cache).


DDR5 channel (bandwidth) scaling for composite workload tests

 

[AusWolf]

That's the point. You have no experience of the types of work where memory bandwidth and latency matters, or high FPS gaming, or even have the hardware to display those FPS (GPU and screen).

Doesn't lend much credence to your opinion or doubts thereof.

For people who do dataset based work, e.g. scientific computing, Bloomberg trading (currently recommend 32 GB of RAM and a Intel 11th gen/Zen 3 or better CPU with a passmark above 2500), IBM SPSS, the types of huge Excel files some corporations work with, certain Adobe software for content creation, etc. memory performance makes a big difference. Typically more channels is better than higher frequency for the simple reason that high frequency is hard to maintain with high capacity sticks, but with DDR5 you can get 32/64 even 128 GB dual DIMM kits that run at high speeds 5000 MHz plus (technically 4x32 in two 2x32 bit DIMMS, this isn't conventional quad channel which would be 4x64 bit). Lots of workstation owners are put off by the cost of threadripper pro/xeon workstations so they buy the higher core count consumer grade CPUs and use high bandwidth and capacity memory sticks.

Phoronix does good testing of workstation and server software under Linux, which has fewer limitations than Windows when it comes to scaling in professional software. Here's an example of Code compilation, AI, rendering, database software, deep learning etc. https://www.phoronix.com/review/ddr5-epyc-9004-genoa. Channels are bandwidth, so channel scaling and bandwidth scaling is the same thing - you'd see similar results from moving from 3000-6000 MHz memory all else considered.

Actually hardcore overclocking (YT channel) goes into how most reviewers and consumers don't really understand the memory subsystem. Subtimings can be tuned to offer significant performance differences at the same frequency, because it's not actually frequency that is the only limiter, it's absolute latency, which is a formula of the data rate (MT/s) and CAS latency (CL). Remember when the CPU is waiting for data from the cache/memory, it's stalled, absolute latency is what is important to get as low as possible, so your CPU can actually stretch it's legs. This is precisely why the X3D chips removed much of the limitations of Zen 3 which was memory starved in high FPS gaming (huge fast cache).


DDR5 channel (bandwidth) scaling for composite workload tests

View attachment 279064
View attachment 279063

I'm not saying that you're wrong, or that I have any experience with such software. All I'm saying is that I don't see a difference in performance in any application (except for file compression) in reviews that I could find online. Applications that show bigger difference probably exist (if you say so), but I don't see them.

Also, just because I don't do high FPS gaming, I still don't think the difference between 300 and 320 FPS is massive and worth spending a lot more money on. In my system's terms, it's like 55 vs 60 FPS.

 

[dgianstefani]

I'm not saying that you're wrong, or that I have any experience with such software. All I'm saying is that I don't see a difference in performance in any application (except for file compression) in reviews that I could find online. Applications that show bigger difference probably exist (if you say so), but I don't see them.

Also, just because I don't do high FPS gaming, I still don't think the difference between 300 and 320 FPS is massive and worth spending a lot more money on. In my system's terms, it's like 55 vs 60 FPS.

You don't have to spend more money, you just need to learn how to tune.

Whether you care about FPS or not the point is there's a clear, measurable difference in CPU performance when you change memory speed and latency. This is less obvious in gaming, especially when GPU limited as with the 3080ti in the TPU test, but becomes incredibly obvious when you actually test memory sensitive workloads, of which many exist.

 

[AusWolf]

memory sensitive workloads, of which many exist.

Like I said, I believe you - I just don't see them represented in reviews, so I don't know how memory scales in those applications, and cannot base my recommendations around them.

 

[CyberCool]

Cache won't do anything for virtualization.

It will be the same situation as 5800X3D, lower clocks, with higher barrier to heat transfer, but with 3x cache for workloads and games that can benefit.

I'd say the 13900K, 13600K, 5800X3D and 7950X are the only CPUs worth buying at the moment for serious users.

Intel wins both value and single core performance. AMD wins multicore/efficiency.

The Zen 4 X3D have a gimmick on the 7900/7950X3D where one die is frequency with no additional cache and one die has 3D cache. For those that like to whine about P/E cores, this is much worse. The 7800X3D is the only worthwhile CPU in that stack IMO.

AMD has better support for "future proofing" as much as that applies to most people (not many people), and their platform is more ambitious, but a lot more buggy and early access. Everyone on Zen 4 is basically a beta tester. Intel has faster cores with tighter latency, and the E cores will only get better, but it can be a little hard to cool if you don't know what you're doing.

The other advantage to Intel Raptor Lake 13th gen is the memory controller is significantly better than Zen 4. You can reach 7000+, probably 8000 MHz on some motherboards with a 13900K. The architecture also scales very well with memory frequency.

Curious, why don't you consider the 13700k & 7900(x) worth buying at the moment for serious users?

Not sure what you meant by serious users as you included the 5800X3D which is really a gaming geared CPU.

I'm going to wait for reviews of the 7900/7950X3D for gaming / VM use, but I totally agree on the whole one die cache thing, doesn't look good.

 

[lors111]

Anyone here who switched from 5800X3D to 13700K? Thinking about going back to Intel, because of the better 1% lows compared to AMD.

 

[Taraquin]

Anyone here who switched from 5800X3D to 13700K? Thinking about going back to Intel, because of the better 1% lows compared to AMD.

12:20 if you run xmp then you will gain about 10% on lows vs 5800X3D in 1080p (12700K is about 5% faster on lows but slower on avg than 5800X3D, 13700K is about 5% faster than 12700K stock and tuned).

If you tune them both 13700K will be 15-20% faster on lows. 20:43.

 

[lors111]

If you tune them both 13700K will be 15-20% faster on lows. 20:43.

Do you have a link with kind of a walkthrough?

 

[Taraquin]

Do you have a link with kind of a walkthrough?

In the video they show the bios-settings used for the 12700K, it can be applied to the 13700K aswell if you have the right ram kit At www.overclock.net you can get more specific help

 

[Der Lokator]

Another CPU that I won't be recommending to average users due to its insane cooling requirements at stock.

What? Who cares about Cinebench temps? Gaming is between 100 and 160w and a Noctua NH-D15 cools easily 240w AT 1000 rpm.

 

[AusWolf]

What? Who cares about Cinebench temps? Gaming is between 100 and 160w and a Noctua NH-D15 cools easily 240w AT 1000 rpm.

100-160 W in gaming? That's a lot!

I like my PCs stable and under throttling temperatures in every situation, just in case.

 

[SundayOverclocker]

So I was reading this review, aiming to decide whether I want 13600K or 13700K (or 7700X, or 7900X). And then I noticed this:

I don't know about other readers interested in this review, but I hate this! I want to know how this CPU will perform in my real-world setup if I get it, not with a kit of RAM that costs as much as the CPU itself. 6000 CL40 might have been fine, but not this. Ugh!