Intel Core i9-14900K Raptor Lake Tested at Power Limits Down to 35 W

[W1zzard]

What happened to the 95W undervolt results?

ugh .. fail .. i forgot the UV when I added 200 W. . i'll rerender the charts

edit: should be fixed now

 

[pexxie]

Not bad with the undervolt , they better release an i9-14900 non K , it would be a better buying option

Or even the i9-14900T. It's a pity TPU stopped reviewing non-K i7s & i9s. Much better efficiency. The i9-13900T efficiency is unreal.

 

[W1zzard]

It's a pity TPU stopped reviewing non-K i7s & i9s

Uh, I reviewed 13400F this year, nothing else was interesting enough to spend my own money on, for a review. I definitely plan on buying a couple of 14th Gen non-K SKUs when they are released

 

[pexxie]

I reviewed 13400F this year

True story, although it's an i5, that's why I carefully wrote "i7s & i9s."

 

[Lew Zealand]

Or even the i9-14900T. It's a pity TPU stopped reviewing non-K i7s & i9s. Much better efficiency. The i9-13900T efficiency is unreal.

Because most people don't want to spend $550 on a 35W processor. Instead buying the $590 i9-13900K and afterwards testing at 35W or 65W (to match the i9-13900) is a reasonable facsimile and gets a review out on the processor most people want, with information relevant to the other 2 for those interested.

As time and money efficient for reviewing as the 13900T is power-efficient.

 

[THU31]

Wouldn't it be logical to assume that K SKUs have better binned dies than non-K ones? So in theory you should get better efficiency at whatever power limit you decide to go for.

I don't think 7% is a significant cost saving when buying something that you'll use for years most likely, at least for a single PC. If you were equipping an office with dozens of them, then maybe.

 

[pexxie]

Because most people don't want to spend $550 on a 35W processor. Instead buying the $590 i9-13900K and afterwards testing at 35W or 65W (to match the i9-13900) is a reasonable facsimile and gets a review out on the processor most people want, with information relevant to the other 2 for those interested.

As time and money efficient for reviewing as the 13900T is power-efficient.

Actually that could also be an interesting comparison! Factory-limited vs user-limited. Impractical and useless, sure, but for fun and curiosity's sake anyway, especially when limiting down to the -T TDP levels. It's hard/impossible finding bench/review data on the -T, it being so niche. There's also very limited data on the 13th high-end non-Ks.

 

[soPvN]

Any chance you could do this for the 14700k as well, where you push the P cores for performance, and make the E cores more efficient by UV and UC. Showing points of diminishing returns possibly ? Would love to see how you could push performance in games, while also eating any background tasks that comes it's way. People tout the 7800x3d as the king of gaming, but I'd like to know how well it would handle when there's some tasks running in the background. I've seen posts where the 7950x3d clearly gets more/sustains performance due to this.

 

[fevgatos]

Any chance you could do this for the 14700k as well, where you push the P cores for performance, and make the E cores more efficient by UV and UC. Showing points of diminishing returns possibly ? Would love to see how you could push performance in games, while also eating any background tasks that comes it's way. People tout the 7800x3d as the king of gaming, but I'd like to know how well it would handle when there's some tasks running in the background. I've seen posts where the 7950x3d clearly gets more/sustains performance due to this.

It's not the 14700k but here you go, stock out of the box and tuned for efficiency. Im throwing in a 12900k just for comparison

Stock 14900k

Spoiler

Tuned 14900k

Spoiler

Tuned 12900k

Spoiler

This is the heaviest game in terms of power draw. The 7800x 3d is nowhere near the 14900k in performance.

 

[Lew Zealand]

The 7800x 3d is nowhere near the 14900k in performance.

So close and dropped the ball in the last sentence. Nowhere is that shown in the videos, so yet another empty fanboy claim. Go ahead and tune the 7800X3D similarly and do the tests under the same conditions.

 

[fevgatos]

So close and dropped the ball in the last sentence. Nowhere is that shown in the videos, so yet another empty fanboy claim. Go ahead and tune the 7800X3D similarly and do the tests under the same conditions.

I have, the 7800x 3d with 6200c28 tuned ram is a little behind the stock 12900k. Also tested a 5800x 3d which...yeah well, let's just say its not competitive at all.

 

[soPvN]

It's not the 14700k but here you go, stock out of the box and tuned for efficiency. Im throwing in a 12900k just for comparison

Stock 14900k

Spoiler

Tuned 14900k

Spoiler

Tuned 12900k

Spoiler

This is the heaviest game in terms of power draw. The 7800x 3d is nowhere near the 14900k in performance.

Thank you for this !!

Edit: can you do some roaming around Hogwarts legacy or cyberpunk? I know those to like the CPU quite a bit.
With temps, power, fps avg lows etc.
Also more info on what exactly you did for the tuning of efficiency of ecores. How low did you get overall wattage output while gaming. It's not so much about performance only, as it's about getting a nice overall performance to power ratio, aiming to get close to the 7950x3D. As with 2 CCD's, one can never completely use no power. I honestly don't believe 7800x3d should be compared to the 14700k, prolly the 14600k. If AMD had a brain they would've made the 7900x3d have 8 cores on the first ccd and 4 on the 2nd to create a ideal gaming CPU, while avoiding a hit from background activities. Honestly a waste of silicone. Just seems like greed, cheap out on the 7800x3d, create the 7900x3d useful for no one, then create the 7950x3d which has a great purpose but is overkill for someone who wants gaming performance with the ability to not hinder performance from having stuff on in the background.

 

[fevgatos]

Thank you for this !!

Edit: can you do some roaming around Hogwarts legacy or cyberpunk? I know those to like the CPU quite a bit.
With temps, power, fps avg lows etc.
Also more info on what exactly you did for the tuning of efficiency of ecores. How low did you get overall wattage output while gaming. It's not so much about performance only, as it's about getting a nice overall performance to power ratio, aiming to get close to the 7950x3D. As with 2 CCD's, one can never completely use no power. I honestly don't believe 7800x3d should be compared to the 14700k, prolly the 14600k. If AMD had a brain they would've made the 7900x3d have 8 cores on the first ccd and 4 on the 2nd to create a ideal gaming CPU, while avoiding a hit from background activities. Honestly a waste of silicone. Just seems like greed, cheap out on the 7800x3d, create the 7900x3d useful for no one, then create the 7950x3d which has a great purpose but is overkill for someone who wants gaming performance with the ability to not hinder performance from having stuff on in the background.

I have some videos on cyberpunk with the 12900k if you are interested, it's not as heavy as TLOU in terms of power draw, and in terms of performance the 12900k already gets over 130 fps average, the 14900k is closer to 160+. Worst case scenario is toms dinner were the 12900k drops to around 90-100. Power draw is lower than in TLOU

Hogwarts is tricky to benchmark, I get over 130 fps almost everywhere except the village of hogsmeade where it drops to the 80s. Power draw in this one is negligible, 50 to 70w.

 

[Gica]

PL 125 versus PL 253
Games
-1.4% in 1440p (TPU review, of course, with the most powerful video card on the planet. For the others, the difference is ZERO!)
0% in 4K (idem)

For Content Creation, see here

 

[Clanmaster21]

Anand techs numbers are wrong. He is using TDP on the amd cpus, not actual power draw.

I'm a bit late to the discussion, but I found a page here that does a similar benchmark to anandtech but with the correct numbers for power limit. I'm uncertain what benchmark is being used exactly for that section, but if you use those corrected power limits with the results from anandtech then you get the results in this graph. Since the results match up well between the two pages, I've included the result at 65W from the computerbase page, allowing a direct comparison between the three at a 65W power limit. I've used the power limits on the x axis since the other two sources didn't measure actual socket power.

I tried to scale up the 14900K results from this post by multiplying the "points per watt" cinebench values by the power usage measured in blender, since the power usage in cinebench isn't given, but this gave results with the 14900K below the 13900K at basically every value, so I've instead just bumped up the 13900K results by 5% to give a crude estimate. It would be helpful to have the raw cinebench power usage (or performance) numbers so I can compare properly.

Going by that estimate it seems the 14900K has better performance at >180W, but the 7950X wins below that. I've included a second graph to show efficiency too.

 

[KaitouX]

I'm a bit late to the discussion, but I found a page here that does a similar benchmark to anandtech but with the correct numbers for power limit. I'm uncertain what benchmark is being used exactly for that section, but if you use those corrected power limits with the results from anandtech then you get the results in this graph. Since the results match up well between the two pages, I've included the result at 65W from the computerbase page, allowing a direct comparison between the three at a 65W power limit. I've used the power limits on the x axis since the other two sources didn't measure actual socket power.

I tried to scale up the 14900K results from this post by multiplying the "points per watt" cinebench values by the power usage measured in blender, since the power usage in cinebench isn't given, but this gave results with the 14900K below the 13900K at basically every value, so I've instead just bumped up the 13900K results by 5% to give a crude estimate. It would be helpful to have the raw cinebench power usage (or performance) numbers so I can compare properly.

Going by that estimate it seems the 14900K has better performance at >180W, but the 7950X wins below that. I've included a second graph to show efficiency too.

Core i9-13900K, i7-13700K & i5-13600K: Gaming-Könige im Test: Leistungsaufnahme und Effizienz

Intel Raptor Lake im Test: Leistungsaufnahme und Effizienz / Leistungsaufnahme in Anwendungen ab Werk / Der Verbrauch steigt

www.computerbase.de

This has the numbers from both Raptor Lake and Zen 4. On their Cinebench tests the 13900K is faster than the 7950X at stock and 45W.


The stock power limit of the entire "7000X" line up is dumb, nearly doubling the power consumption for at most 8% improvement is kinda ridiculous. It somehow makes Raptor Lake stock power limits look "good".

 

[Clanmaster21]

Core i9-13900K, i7-13700K & i5-13600K: Gaming-Könige im Test: Leistungsaufnahme und Effizienz

Intel Raptor Lake im Test: Leistungsaufnahme und Effizienz / Leistungsaufnahme in Anwendungen ab Werk / Der Verbrauch steigt

www.computerbase.de

This has the numbers from both Raptor Lake and Zen 4. On their Cinebench tests the 13900K is faster than the 7950X at stock and 45W.


The stock power limit of the entire "7000X" line up is dumb, nearly doubling the power consumption for at most 8% improvement is kinda ridiculous. It somehow makes Raptor Lake stock power limits look "good".

Thanks, I had seen that website but didn't see you could change the tabs in the table. I've redone the graph using that Cinebench R23 data and it's a similar but closer result, the 14900K does better above 160W and below 60W, but the 7950X does better between the two.

Yeah it seems the default power limit for most hardware is way too high, I wish more people tested how power limits affected performance because there's no way I'm running any modern CPU or GPU at stock settings. I would like to compare the two above to CPUs designed to run in the <60W range like a 13900T or a 7945HX or something, but the benchmarks for those processors let them boost to like 100W so I can't find any useful data points.

 

[AnotherReader]

Core i9-13900K, i7-13700K & i5-13600K: Gaming-Könige im Test: Leistungsaufnahme und Effizienz

Intel Raptor Lake im Test: Leistungsaufnahme und Effizienz / Leistungsaufnahme in Anwendungen ab Werk / Der Verbrauch steigt

www.computerbase.de

This has the numbers from both Raptor Lake and Zen 4. On their Cinebench tests the 13900K is faster than the 7950X at stock and 45W.


The stock power limit of the entire "7000X" line up is dumb, nearly doubling the power consumption for at most 8% improvement is kinda ridiculous. It somehow makes Raptor Lake stock power limits look "good".

In tests that stress the CPU the most such as Blender, the 7950X is faster than the 13900K at all power levels.

 

[fevgatos]

Thanks, I had seen that website but didn't see you could change the tabs in the table. I've redone the graph using that Cinebench R23 data and it's a similar but closer result, the 14900K does better above 160W and below 60W, but the 7950X does better between the two.

Yeah it seems the default power limit for most hardware is way too high, I wish more people tested how power limits affected performance because there's no way I'm running any modern CPU or GPU at stock settings. I would like to compare the two above to CPUs designed to run in the <60W range like a 13900T or a 7945HX or something, but the benchmarks for those processors let them boost to like 100W so I can't find any useful data points.

Who would have thought that amd does not actually have a lead in efficiency huh? Not like I've been saying it for the last couple of years.

Especially the lower you go on the stack, the more atrocious it becomes. I7 13700k vs R7 7700x is a slaughter.

In tests that stress the CPU the most such as Blender, the 7950X is faster than the 13900K at all power levels.

View attachment 323145

Out of 10 or so tests you picked the one that amd wins. Hard copium.

 

[Clanmaster21]

Who would have thought that amd does not actually have a lead in efficiency huh? Not like I've been saying it for the last couple of years.

Especially the lower you go on the stack, the more atrocious it becomes. I7 13700k vs R7 7700x is a slaughter.


Out of 10 or so tests you picked the one that amd wins. Hard copium.

Yeah given how much everyone says AMD is more efficient I was expecting a massive win for the 7950x, but it really depends on the task and power limit.

I think the 13700k is more comparable to the 7900x though, they have the same number of threads at least and are priced similarly here.

 

[fevgatos]

Yeah given how much everyone says AMD is more efficient I was expecting a massive win for the 7950x, but it really depends on the task and power limit.

I think the 13700k is more comparable to the 7900x though, they have the same number of threads at least and are priced similarly here.

Originally the r7 was meant to compete against the i7, hence the names. Even prices were close together on release, the 7900x was more expensive than both of them.

 

[chrcoluk]

Wouldn't it be logical to assume that K SKUs have better binned dies than non-K ones? So in theory you should get better efficiency at whatever power limit you decide to go for.

I don't think 7% is a significant cost saving when buying something that you'll use for years most likely, at least for a single PC. If you were equipping an office with dozens of them, then maybe.

At my current energy costs I have estimated about £36 a year in savings from the current measure's I place on my CPU. Big enough over multiple years to give me something meaningful I suppose, but also not great once I actually calculated it for this post.

A little rundown.

I have excluded the higher power savings under full load (which are higher) as I have yet to do any of that kind of load on my chip outside of testing. So to make it a fair comparison it wasnt considered.
I included the savings from undervolting, these exist at moderate loads and increase the higher you go.
I include the savings from my custom power schema's these actually provide a saving bigger than my undervolting in a lot of my use cases, in other use cases the undervolt provides the bulk.

I calculated based on an average 30w saving over 16 hours a day for 365 days of year.

What I categorised is schema settings includes my affinity adjustments. As an example if I block turbo clocks when watching media like youtube, twitch, netflix etc. especially full screen it saves about 20-30w. If the browser is using p-cores and I have all cores unparked it will use another 10-20w on top of that, so my media schema only has the 2 preferred p-cores always unparked, whilst others are not blocked from unparking the OS wont do so in this workload. Using software p-states in the schemas has an effect also.

If the only thing I did on my CPU was to lower the power limit, no undervolt, no custom schema's I probably wouldnt actually be saving much power unless it was really low like 35w, as in day to day use the chip will not often exceed 125w even at stock. Downloading at gigabit speeds on steam with p-cores, I expect is similar load to a heavily threaded game (steam really makes these intel chips ramp up and use a lot of power on high speed downloading) I have seen it clear 100w untamed. I dont put steam on e-cores though as then any game launched by it would inherit the affinity so I would then need to set manual affinity for every game.

At full pelt though my undervolt has a much bigger impact than 7% on CPU power, as a % of system power though it will be lower. But I dont compile software, software encode, or other similar use case on this machine.

For games, I also have custom GPU profiles that manipulate the clock speed the GPU will run at and the GPU is undervolted, that can save me 100w when playing a game, more realistically usually 50-80w. Combined with CPU savings probably exceed 100w. Some days I will be playing games all day, but I can go several days without playing any. I think if I factored this in more I am saving more than £36 a year. I not that long ago posted a afterburner screenshot on the map screen in dune spice wars in a thread, and it showed me instant dropping GPU by about 70w just from switching to a profile that caps my GPU clocks to 1500mhz, with no loss of performance.

 

[Gica]

At my current energy costs I have estimated about £36 a year in savings from the current measure's I place on my CPU. Big enough over multiple years to give me something meaningful

- You smoke?
- Yes.
– How many packages per day?
– Three packages.
– How much does a package cost?
– 18 lei (Romanian currency).
- How long have you been smoking?
- 15 years.
- So, if a pack costs 18 lei, you smoke three a day, it means that you spend 1620 lei every month. In a year they make about 19,440, correct?
- Correct.
- So, if you spend 19,440 per year, not taking inflation into account, in the last 15 years, you spent over 291,600 lei, right?
- Correct.
– Did you know that if you hadn't smoked and had saved this money in a savings account, adding up the interest for 15 years, you could have bought yourself a Ferrari?
- You smoke?
- Not.
- Then where is that damn Ferrari of yours?!

 

[chrcoluk]

- You smoke?
- Yes.
– How many packages per day?
– Three packages.
– How much does a package cost?
– 18 lei (Romanian currency).
- How long have you been smoking?
- 15 years.
- So, if a pack costs 18 lei, you smoke three a day, it means that you spend 1620 lei every month. In a year they make about 19,440, correct?
- Correct.
- So, if you spend 19,440 per year, not taking inflation into account, in the last 15 years, you spent over 291,600 lei, right?
- Correct.
– Did you know that if you hadn't smoked and had saved this money in a savings account, adding up the interest for 15 years, you could have bought yourself a Ferrari?
- You smoke?
- Not.
- Then where is that damn Ferrari of yours?!

A weird way of asking me if I smoke? I dont.

 

[Apart]

If power limited and undervolted, does it still support Intel APO? Would be interesting to see a test like this but with APO enabled games. Especially if you're possible to do it with all K 14th gen CPUs.