Intel Core i7-13700K

[W1zzard]

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

View attachment 294326

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!

Oh wow .. my bad .. this test setup table is copy and paste from the 13900K review where I've included DDR5-7400 as _additional_ data point

Intel Core i9-13900K Review - Power-Hungry Beast

With the Core i9-13900K, Intel delivers impressive performance. Our in-depth review confirms: Raptor Lake is the world's fastest CPU for gaming. Even in applications the processor is able to match AMD's Zen 4 Ryzen 9 7950X flagship. If only power consumption wasn't so high...

www.techpowerup.com

http://imgur.com/YpUxeMM

(1080p, note the brown bar)

All testing in the 13700K review was done at DDR5-6000 CL36

http://imgur.com/LH74Wb5

I've just updated the test system table in this review (and 13600K) to only show the tested memory configuration

 

[SundayOverclocker]

this test setup table is copy and paste from the 13900K review where I've included DDR5-7400 as _additional_ data point

Oh, thank you very much! In that case it's my bad, I was confused by the two rows for a bit, but at some point turned a blind eye to the first row entirely. Appreciate the clarification.

And thanks for your great tests! TechPowerUp is now my go-to resource for hardware reviews (and I follow quiet a few of them).

 

[Der Lokator]

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

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

I actually run my 13700k 5 Ghz 2 cores and 4,7 Ghz at all cores undervolted with 95w PL2 28s and 65w PL1 now and it runs mostly around 35-45w at gaming.

 

[SundayOverclocker]

I actually run my 13700k 5 Ghz 2 cores and 4,7 Ghz at all cores undervolted with 95w PL2 28s and 65w PL1 now and it runs mostly around 35-45w at gaming.

Wow, that's awesome and it absolutely changes how I view Raptor Lake vs Zen 4/5.
Is Z-series mobo required for undervolting and power-limiting?

 

[AusWolf]

Wow, that's awesome and it absolutely changes how I view Raptor Lake vs Zen 4/5.
Is Z-series mobo required for undervolting and power-limiting?

I'm not sure about undervolting, but for power limiting, no. I think it's only overclocking that you need a Z-series motherboard and a K-series CPU for.

 

[Berfs1]

Is Z-series mobo required for undervolting and power-limiting?

Nope, voltage controls and power controls are available on any chipset! CPU overclocking however, requires a Z chipset (unless you go through some BCLK workaround), and an unlocked multiplier, so that you can increase the multipliers past stock.

 

[chrcoluk]

I am supposed to be just stable testing my 13700k, but I have got caught up in the voltages, and power.

Struggling to find published data for this stuff specifically, but here is my observations.

First my board defaulted to 115C tjmax (its max setting), its now manually set to 100, but it meant I had hit 109C at one point with no throttling.

At first sign it seems like the auto voltage of these chips is hugely conservative (overvolted). Given that in the review an o/c was achieved with a voltage 0.14v below my stock voltage, and others reporting running these chips at 1.2v, my stock voltage is 1.4v under load.

I dropped to 1.3v which knocked off 30w from peak power (230 down to 200) and about 9C temp drop. After this point though I lowered my pl2 down to 175w to fit the cooler I am testing with.

cinebench r15, a test that runs longer than cpuz, clockspeeds throttled by a whopping 1ghz, which seems harsh for a 175w limit. But struggling to find comparable data for this so no reference point. 4.3 p cores and 3.2 e cores.

I then lowered vcore further to 1.25v and recovered 150mhz clock speed on both sets of cores, score only went up by 3% though. Although that was in line with a further temp drop. Next is 1.2v, but I probably wont go below that though as I am not expecting 1.2v to be stable for daily use.

the chip is crazy fast though, the e core only a little slower than a core on my 9900k, p core about 50% faster, even with the throttling mullti core test on cinebench r15 is almost twice as fast as 9900k. (175w pl2).

I think low down on the v/f curve (which years ago would be stock) these are actually efficient, the efficiency rep is from how agressive the factory o/c is. Also my first time using speedshift (hardware p-states) and it ramps up to turbo on the tiniest of cpu utilisation, but with utilisation so low it doesnt seem to harm idle power much.

Another observation clock drift is pretty high, if I reboot it drifts half a second, and its about 40ms every 15 minutes on ntp, my 9900k and laptop is under 10ms every 15 mins.

 

[dgianstefani]

I am supposed to be just stable testing my 13700k, but I have got caught up in the voltages, and power.

Struggling to find published data for this stuff specifically, but here is my observations.

First my board defaulted to 115C tjmax (its max setting), its now manually set to 100, but it meant I had hit 109C at one point with no throttling.

At first sign it seems like the auto voltage of these chips is hugely conservative (overvolted). Given that in the review an o/c was achieved with a voltage 0.14v below my stock voltage, and others reporting running these chips at 1.2v, my stock voltage is 1.4v under load.

I dropped to 1.3v which knocked off 30w from peak power (230 down to 200) and about 9C temp drop. After this point though I lowered my pl2 down to 175w to fit the cooler I am testing with.

cinebench r15, a test that runs longer than cpuz, clockspeeds throttled by a whopping 1ghz, which seems harsh for a 175w limit. But struggling to find comparable data for this so no reference point. 4.3 p cores and 3.2 e cores.

I then lowered vcore further to 1.25v and recovered 150mhz clock speed on both sets of cores, score only went up by 3% though. Although that was in line with a further temp drop. Next is 1.2v, but I probably wont go below that though as I am not expecting 1.2v to be stable for daily use.

the chip is crazy fast though, the e core only a little slower than a core on my 9900k, p core about 50% faster, even with the throttling mullti core test on cinebench r15 is almost twice as fast as 9900k. (175w pl2).

I think low down on the v/f curve (which years ago would be stock) these are actually efficient, the efficiency rep is from how agressive the factory o/c is. Also my first time using speedshift (hardware p-states) and it ramps up to turbo on the tiniest of cpu utilisation, but with utilisation so low it doesnt seem to harm idle power much.

Another observation clock drift is pretty high, if I reboot it drifts half a second, and its about 40ms every 15 minutes on ntp, my 9900k and laptop is under 10ms every 15 mins.

Use Intel software to do a per core tune, then lower voltages until its unstable in stress testing, you'll get a better result than just modifying values that the chip adapts to based off its default VFCurve.

It's not strictly necessary to harshly limit TDP if you do those steps, and it's useful to have a higher limit in case you are in a situation that stresses all cores. Lower thread count work won't use the full stock TDP if you undervolt.

Easier to do it that way too since everything is in software and you can find the best tune much faster.

 

[chrcoluk]

Use Intel software to do a per core tune, then lower voltages until its unstable in stress testing, you'll get a better result than just modifying values that the chip adapts to based off its default VFCurve.

It's not strictly necessary to harshly limit TDP if you do those steps, and it's useful to have a higher limit in case you are in a situation that stresses all cores. Lower thread count work won't use the full stock TDP if you undervolt.

Easier to do it that way too since everything is in software and you can find the best tune much faster.

Yeah thats what I am doing, although my figures seem much worse than a few people who shared specific data.

I am at a very low vcore now 1.21v, others with similar vcore were hitting max clocks at similiar power under full load. So I am not seeing the correct correlation between voltage, power and performance.

Will need to call it a night here now though as its almost 1am.

Ok done some testing, fixed vcore does seem to such based on these results. Cache offset seems also important.

Chip seems poor on silicon lottery, dont know, as its hard to find specific vcore information (offsets by themselves arent useful as stock voltage varies). This data does include recorded actual vcore readings when using offsets. Peak r23 score nowhere near the magic 30k, I will come back to that when using a proper cooler.

Paste below

---

My 13700k. (seems a silicon loser? not sure got better with cache offset)

Note this testing is done on temporary cheap cooler, so this isnt a thermals test but analysis of vcore vs max TDP vs clock speeds.  Am aware bench scores will improve with better cooler.

Cinebench 23 multi core.

Almost stock bios settings pl1 125 pl2 253 tjmax 100, ram JEDEC ddr4 2400mhz
Only change in bios settings drop tjmax from 115 to 100 as asrock defaulting to 115 O_o
  With stock vcore hits 1.43v under load. O_o
  Max core TDP draw hits 253w cap.
  Clock speeds down to 4.6ghz due to pl2 limit then to 3.7-3.8 ghz due to thermal limit. O_o


Offset voltage (done at request of community).
  Offset -200mv unstable hit vcore 1.28 before app crash.
  Offset -150mv vcore peaks at 1.349v (quite high).
    Max core TDP hits 253w cap.
    Clock speeds down to 5.1ghz due to pl2 limit then to 4.7-4.8ghz due to thermal limit.
  Offset voltage -150mv and Cache offset -50mv vcore peaked at 1.297v
    Max core TDP hits 232w
    Clock speeds initially 5.3ghz no throttle then gradually down to 4.8ghz end of test
    Notably took about 4x as long to thermal throttle and thermal throttle was less agressive, first test it throttled down to 4.8 almost instant.
    Cache offset appearing to have bigger impact than core offset.
  Offset voltage -150mv and Cache offset -75mv, and cache offset -75mv for e-cores vcore peaked at 1.290v
    Max core TDP hits 222w
    Clock speeds 5.3ghz thermal throttle gradually down to 4.85ghz at TDP 180w
    Further improvement in taking longer to hit tjmax
    R20 score 9823
    R23 score 25634

Fixed voltage
  1.21v unstable at high pl2
  1.25v
    Offset minus 50mv on both cache's (dont think this does anything on fixed vcore, might just affect vid's)
      Max core TDP hits 241w
      Clock speeds 5.3ghz then thermal throttle gradually down to 4.15ghz at TDP 141w O_o (ok so fixed vcore seems to suck on raptor lakeon this board at least, although some have reported being able to be stable at fixed 1.15v and good performance, my chip seems nowhere near good enough for that though)
      R20 score 9388
      R23 score 24009 (lower than when at 175w pl2 fixed 1.21v without negative cache offset voltage)

Bios set to pl2 175w

  Offset voltage -150mv and Cache offset -75mv, and cache offset -75mv for e-cores vcore peaked at 1.208v
  Max core TDP hits 174w
  Clock speed around 4.9ghz (was around 4.4ghz on fixed 1.21v at pl2 175w), No thermal throttling
  R20 score 9550 (lower than 253w limit by about 2.5% but with about 25w lower average TDP about 12%)
  R23 score 25181

 

[dgianstefani]

Yeah thats what I am doing, although my figures seem much worse than a few people who shared specific data.

I am at a very low vcore now 1.21v, others with similar vcore were hitting max clocks at similiar power under full load. So I am not seeing the correct correlation between voltage, power and performance.

Will need to call it a night here now though as its almost 1am.

Ok done some testing, fixed vcore does seem to such based on these results. Cache offset seems also important.

Chip seems poor on silicon lottery, dont know, as its hard to find specific vcore information (offsets by themselves arent useful as stock voltage varies). This data does include recorded actual vcore readings when using offsets. Peak r23 score nowhere near the magic 30k, I will come back to that when using a proper cooler.

Paste below

---

My 13700k. (seems a silicon loser? not sure got better with cache offset)

Note this testing is done on temporary cheap cooler, so this isnt a thermals test but analysis of vcore vs max TDP vs clock speeds.  Am aware bench scores will improve with better cooler.

Cinebench 23 multi core.

Almost stock bios settings pl1 241 pl2 253 tjmax 100, ram JEDEC ddr4 2400mhz
Only change in bios settings drop tjmax from 115 to 100 as asrock defaulting to 115 O_o
  With stock vcore hits 1.43v under load. O_o
  Max core TDP draw hits 253w cap.
  Clock speeds down to 4.6ghz due to pl2 limit then to 3.7-3.8 ghz due to thermal limit. O_o


Offset voltage (done at request of community).
  Offset -200mv unstable hit vcore 1.28 before app crash.
  Offset -150mv vcore peaks at 1.349v (quite high).
    Max core TDP hits 253w cap.
    Clock speeds down to 5.1ghz due to pl2 limit then to 4.7-4.8ghz due to thermal limit.
  Offset voltage -150mv and Cache offset -50mv vcore peaked at 1.297v
    Max core TDP hits 232w
    Clock speeds initially 5.3ghz no throttle then gradually down to 4.8ghz end of test
    Notably took about 4x as long to thermal throttle and thermal throttle was less agressive, first test it throttled down to 4.8 almost instant.
    Cache offset appearing to have bigger impact than core offset.
  Offset voltage -150mv and Cache offset -75mv, and cache offset -75mv for e-cores vcore peaked at 1.290v
    Max core TDP hits 222w
    Clock speeds 5.3ghz thermal throttle gradually down to 4.85ghz at TDP 180w
    Further improvement in taking longer to hit tjmax
    R20 score 9823
    R23 score 25634

Fixed voltage
  1.21v unstable at high pl2
  1.25v
    Offset minus 50mv on both cache's
      Max core TDP hits 241w
      Clock speeds 5.3ghz then thermal throttle gradually down to 4.15ghz at TDP 141w O_o (ok so fixed vcore seems to suck on raptor lake, although some have reported being able to be stable at fixed 1.15v and good performance, my chip seems nowhere near good enough for that though)
      R20 score 9388
      R23 score 24009 (lower than when at 175w pl2 fixed 1.21v without negative cache offset voltage)

Bios set to pl2 175w

  Offset voltage -150mv and Cache offset -75mv, and cache offset -75mv for e-cores vcore peaked at 1.208v
  Max core TDP hits 174w
  Clock speed around 4.9ghz (was around 4.4ghz on fixed 1.21v at pl2 175w), No thermal throttling
  R20 score 9550 (lower than 253w limit by about 2.5% but with about 25w lower average TDP about 12%)
  R23 score 25181

Why are you using JEDEC RAM speeds?

 

[chrcoluk]

The per core tuning looks pretty cool, I probably wont go that far, but it is nice its there.

On the ram, after I realised the system the way it was configured wasnt stable when I let it use more power, I just removed that complication and set it to jedec for now, it is low spec'd ram as well, I will use at least XMP spec.

Revised results with fixed bios, higher ambient temp vs earlier.

Revised test done on new 13.02 bios Asrock Z690 Steel Legend, Jedec 2400mhz

New bios has much better default AC/DC load values, and also fixed default of 100C tjmax.
Vcore value fetched from hwinfo as XTU reports VID. But will put VID in brackets.
Note ambient temp 4C higher than earlier test and rose 1.5C between start of these tests and end.

Cinebench 23 multi core.

Almost stock bios settings pl1 125 pl2 253, ram JEDEC ddr4 2400mhz
  With stock vcore hits 1.280v (1.240) under load.
  Max core TDP draw hits 241w.
  Clock speeds start at 5.3ghz then drop to around 4.7 ghz due to thermal limit.
  R20 score 10856
  R23 score 28408

Offset voltage (done at request of community).
  Offset -200mv unstable
  Offset -150mv unstable
  Offset -100mv vcore peaks at 1.304v (1.215)
    Max core TDP hits 228w.
    Clock speeds 5.3ghz then to 4.75ghz due to thermal limit.
  Offset voltage -100mv and Cache offset -75mv, and cache offset -75mv for e-cores vcore peaked at 1.248v
  Cache offset still seems bigger impact than core offset.
    Max core TDP hits 218w
    Clock speeds 5.3ghz thermal throttle gradually down to 4.95ghz at TDP 175w
    R20 score 11309
    R23 score 29524 - what an improvement :)

pl2 175w

  Offset voltage -100mv and Cache offset -75mv, and cache offset -75mv for e-cores vcore peaked at 1.168v (1.122)
  Max core TDP hits 173w
  Clock speed around 5.00ghz, Slight thermal throttling right at end but p-cores stayed 5ghz, e-cores -100mhz at last reading
  R20 score 11006
  R23 score 28663

 

[mahanddeem]

Hi @W1zzard
How much max vcore you needed for 5.5GHz OC in this 13700K? and any LLC changes from stock? and did you change E core values or stock?

 

[chrcoluk]

For reference the above tests I did have been run through all sorts of torture as well as a fair amount of daily use now, and no signs of issues, so its now my daily config on the undervolts. (intel xtu stress tests and bench, long realbench, r23 looping, hours of karhu and gsat - linux is more likely to show instability than windows).

My IMC doesnt seem too great, ram is at XMP speeds of 3200mhz albeit with tightened timings, although after reading some earlier threads I accept me switching to the thermalright bracket (or maybe not) that a mounting issue could be causing my low memory clock ceiling, but regardless I am happy with the upgrade and the oomph out of this system now, certainly feels like a proper upgrade.

Played a few games also and the temps as well as wattage have been consistently lower than when was using my 9900k, so these chip's not having to sweat as hard seem quite efficient for gaming. But do remember I am undervolting, my 9900k was ran at fixed 1.25v so low loads an over volt highest loads an undervolt.

R23 score with proper cooler is 30700 and something.

@dgianstefani with ram at XMP clocks, cinebench scores seem no different.

 

[dgianstefani]

For reference the above tests I did have been run through all sorts of torture as well as a fair amount of daily use now, and no signs of issues, so its now my daily config on the undervolts. (intel xtu stress tests and bench, long realbench, r23 looping, hours of karhu and gsat - linux is more likely to show instability than windows).

My IMC doesnt seem too great, ram is at XMP speeds of 3200mhz albeit with tightened timings, although after reading some earlier threads I accept me switching to the thermalright bracket (or maybe not) that a mounting issue could be causing my low memory clock ceiling, but regardless I am happy with the upgrade and the oomph out of this system now, certainly feels like a proper upgrade.

Played a few games also and the temps as well as wattage have been consistently lower than when was using my 9900k, so these chip's not having to sweat as hard seem quite efficient for gaming. But do remember I am undervolting, my 9900k was ran at fixed 1.25v so low loads an over volt highest loads an undervolt.

R23 score with proper cooler is 30700 and something.

@dgianstefani with ram at XMP clocks, cinebench scores seem no different.

Cinebench doesn't scale with RAM.

 

[chrcoluk]

Cinebench doesn't scale with RAM.

I know now, I tagged you as you asked about me using jedec for that testing.