Does sustained power consumption look similar to peak power consumption? Why does your article lead with "Efficiency at 65 Watts" when the peak power is 90 Watts?
Because the Platform Power Target (PPT) for '65W TDP' models from AMD has always been around 87W (~90 give or take). The TDP is thermal watts for cooling systems not sustained electrical watts for power usage. I assume the author is implying that the models are the AMD-rated 65W TDP and are efficient, not that they specifically use 65 watts, but I suppose it's kinda misleading.
All of the electrical energy going into the chip does become heat aside from a tiny bit that leaves as bits on the various buses (which is mostly canceled out by energy coming in from other chips on the same bus). The difference between TDP and peak power is that the CPU can thermally throttle if needed.
A CPU is not a resistor. So not all the energy is converted into heat. For example, some of it goes down the PCIe bus, and heats up the PCB traces and PCIe card. Likewise with the RAM.
Intel and AMD have dramatically different packaging. The TDP designation in AMD CPU's applies to the CPU portion of the package ie ZEN core chiplet(s) while the IO CCD is rolled into the PPT as it is effectively the rest of the "platform."
If we were to try comparing fruits, AMD's PPT would be comparable to Intel's TDP + chipset (Z690 is 6-watts) + other external components that would otherwise be integrated in a Ryzen IO but as these architectures are vastly different a direct comparison is impossible; for example, AMD's IO is far more complex than Intel's, supporting a wider range of platform designs and IO types. While Intel and AMD both support DDR4\5 at this point within their memory controllers, PCIe 3.x\4.x\5.x is handled inside of AMD's IO while Intel offloads much of that to the chipset, only supporting some lanes inside the CPU, and many CPU's not supporting 5.x. To make matters worse, the Z690 is manufactured on a 14nm process while AMD's IO CCD is 7nm (or something like that) so while the Z690 would use much more power if it did the same thing as the AMD IO CCD, it doesn't. In fact it does very little and the package itself is tiny like 25mm2. This means the AMD IO handles a lot more while being 7nm while Intel integrates all of that into the CPU die which is a more refined 10nm process that efficiency is somewhere between TSMC 5nm and 7nm.
Basically AMD's CPU CCD (5nm?) is more efficient while their IO (7nm?) is less efficient and this doesn't entirely balance out compared to Intel's monolithic 10nm (which more efficient than TSMC's 7nm.)
Also worth noting is that the power dissipated is related to the resistance of the component, not the amount of power going into the component. You should recall that from any basic series circuit you have worked with/on.
My annoyance is mostly that it is misleading - while I understand that companies market these things this way the fact of the matter is that the efficiency is not at 65 Watts, its at 90 Watts. Sure they are spectacularly efficient compared to when running them without power limits but the power limits given by marketing are useless for talking about efficiency because they are just numbers on a slide. AMD could call it TDP 50 Watts and it would be just as meaningful because the real number is so far off.
That's definitely wrong. The PPT number, as the name implies (package power tracking) certainly includes the IOD, it's a limit which applies to the socket. (It does of course not include the B650 (or whatever) chipset but that's of course the same with intel.)
There's no perfect solution. The gap between what "you think" 65W means in case of AMD, and the reality, is still much smaller than in case of Intel's TDP. Inte'ls 125W CPU can reach well beyond 200W, which is insane (nearly 2x). In case of AMD the difference (explained quite clearly) is much lower (30~40%).
PL2 (boost power level) for Ryzen is tdp x 1.35. So, in this case: 65x1.35 = 87.75. The IOD consumes more than what ll be left here. In any case, the best way to measure these types consumption, from my observations, is to measure from the wall. It usually results in far greater discrepancies than software reports.
"Looking at the Ryzen 9 7900, it marginally beats out the Ryzen 9 _7950X_ and sits" - Typo, should be 5950X, above borderlands 3 test on conclusion page.
I am concerned about Zen4 minimum power draw (idle). Judging by graphs on page 2, it seems like CPU consumes 20 watt on idle???
Going back to the Zen4 X comparison vs 13th gen, Intel seems to have advantage at idle. Planning to update home server that sits on idle or very low load 80% of the day, but needs to be on 24/7. I eventually am planing to run VMs on it (rare gaming with GPU pass thru). Thoughts?
thoughts? you are creating a wacko system. one site you want to have low low idle, and than intel is better although the latest generation is less difference... its not the cpu, its the total package with pci-e5 etc that is upping the idle power. on the other side you want GPU power which eats already 50% of your total idle power. so the diff between intel and amd will now be only way less vs total idle.. on top for VM you should have an AMD multicore for sure. my advise? wait for the 7000 APU it might fit both your needs.
Is it just me or do the 7600x and 7700x just seem like bad deals compared to the newly released chips. $30 moves you up to the 7700 or 7900 respectively with the same rated boost clocks for ST task. Lower base clocks will hurt but the additional cores will help compensate in most MT situations. The 7700 beats the 7600x in many of the multi-threaded tests and does it for a small premium and while using a lot less power in the process.
X is now the "founders" edition where rather than get discounts like kickstarter, you pay for the privilege for getting it first. Nvidia did this with the 2080Ti FE being $1200 whereas msrp was $999. They saw it work well so everyone taking a page. Ryzen traditionally launched with non-X and that was the go to. The X is mostly for emotional reasons which plays a bigger factor in marketing than you think. Most people will never lower themselves to a x700 series card for instance. That is why nvidia did two 4080s in the beginning but backtracked. Wrong move.
Hi. For "(3-3) Dolphin 5.0 Render Test" you have the test results in your Bench different to the test results in the graph.
In the graph: Intel Core i9-12900KS = 182.2
In your Bench: Intel Core i9-12900KS with DDR5 = 169 Intel Core i9-12900K with DDR4 = 181
Also, according to your Bench, for "(3-4c) Factorio v1.1.26 Test, 20K Hybrid," the i3-12300 with a score of 72 beats *ALL* the new Ryzen 7xxx CPUs. Can you confirm that please?
I haven't checked any of the other results to your bench but suspect that more might also be different. Also, are efforts underway to update the Bench to include these results, please?
The multiple tests where the 7900 is beating the 7950X by large (>5%) differences in game tests makes no sense and makes me concerned for the repeatability of the test suite. There is nothing (cache, clock speed, architecture, TDP, NUMA) that is inferior on the 7950X compared to the 7900 but somehow it loses by a large margin in many game tests.
While you can make the argument that a "65W" AMD CPU consuming 90W is misleading, at least it's going to be consistently 90W regardless of which model you choose...
Doesn't it become really weird to talk about efficiency while only comparing to TDP and not actual power consumption for the load? Not everything hits TTP as yCruncher does either. Preferrably the score in each result should be normalized to actual power consumption, or something similar. Even that has its issues though, since the balance between performance and efficiency is somewhat tuneable.
Real question: Why do CPUs no longer idle in the 800MHz-1600MHz range? Is there too great a change in the multiplier to hit max turbo at this point? Otherwise, what's the point at idling around 3.6GHz?
Linus Tech Tips actually did a review on these chips with benchmarks. The 7900 and 7900X are basically identical chips with one being 65W and the other being 170W, but the 7900 is also unlocked and if you use PBO, it can match its 7900X sibling with marginal performance differences on some programs and the upside being the 7900 is cheaper. I for one think I may pick the 7900 up for a build but I am also waiting on pricing for the 7900X3D and 7950X3D parts for my big gaming PC build that will replace my current 5900X.
Though not with the bundled heatsink of course haha. AMD included inadequate cooling for these CPU running at 99C under sustained workloads at factory clocks.
They can't run at 99C since AMD uses temperature as the basis for how high the clock speed will go. The default for the 7900X and 7950X was at 95 degrees C, and the chip won't go more than a degree above that point. With the temperature as the key, the chip will clock itself as high as cooling works, so better cooling results in higher clock speeds, and worse cooling will result in lower speeds, but you will never see temperatures above the max temperature.
Bundling coolers with its more affordable CPUs means users can spend their budget on a premium AIO cooler through better memory, storage, or graphics, which will have a positive impact on performance. What do u mean by that isnt aio cooler just cpu cooler liquid obviously but isnt it a cpu cooler or are you refering to fans ?
I am disappointed that the benchmarks do not include more samples from the Zen 3 family, aside from the 5800X3D. With many of us rocking Ryzen 3xxx and 5xxx, such a direct comparison would be much more useful than all the catalogue of horribly power-inefficient Intel chips that most people are not likely to buy anyway...
OK, so I see a CPU that's a bit less power draw with a price close to last AMD generation. Would change my 5-5600X IF there was a descent motherboard that fits my mATX case! LOUSY selection so far of mATX motherboards from ASUS, Gigabyte, MSI, ASRock, etc. Upgrade needs AM5 motherboard and DDR5 (probably) in addition to 7000-series CPU. B650M expensive or crippled in capability and X670M boards are basically non-existent. Need to wait out until manufacturers decide offering to us need some degree of reasonableness in price and features. O(old)MHO. Thanks for good articles...I hope MB manufacturers read reviews AND comments.
How do you get V-Ray benchmark values which are 35 to 50% higher than anything that can be seen on the results page? What version of the benchmark are you using?
It isn't the latest iteration. It's the previous version (4). The critical area to consider is the consistency of the results within the same version.
Of course, we will update this benchmark in our next wave to the latest version. If we did this for every benchmark, I would spend 100% of my time testing and 0% writing.
I'd like to know the idle power consumption of the CPUs. Apparently there is quite a difference at idle between the latest AMD and Intel options but can't seem to find much in the way of tests.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
55 Comments
Back to Article
bcortens - Monday, January 9, 2023 - link
Does sustained power consumption look similar to peak power consumption?Why does your article lead with "Efficiency at 65 Watts" when the peak power is 90 Watts?
AshlayW - Monday, January 9, 2023 - link
Because the Platform Power Target (PPT) for '65W TDP' models from AMD has always been around 87W (~90 give or take). The TDP is thermal watts for cooling systems not sustained electrical watts for power usage. I assume the author is implying that the models are the AMD-rated 65W TDP and are efficient, not that they specifically use 65 watts, but I suppose it's kinda misleading.meacupla - Monday, January 9, 2023 - link
yeah, not all the electrical energy going into the chip ends up as heat energy.saratoga4 - Monday, January 9, 2023 - link
All of the electrical energy going into the chip does become heat aside from a tiny bit that leaves as bits on the various buses (which is mostly canceled out by energy coming in from other chips on the same bus). The difference between TDP and peak power is that the CPU can thermally throttle if needed.ballsystemlord - Monday, January 9, 2023 - link
A CPU is not a resistor. So not all the energy is converted into heat.For example, some of it goes down the PCIe bus, and heats up the PCB traces and PCIe card. Likewise with the RAM.
Samus - Wednesday, January 11, 2023 - link
Intel and AMD have dramatically different packaging. The TDP designation in AMD CPU's applies to the CPU portion of the package ie ZEN core chiplet(s) while the IO CCD is rolled into the PPT as it is effectively the rest of the "platform."If we were to try comparing fruits, AMD's PPT would be comparable to Intel's TDP + chipset (Z690 is 6-watts) + other external components that would otherwise be integrated in a Ryzen IO but as these architectures are vastly different a direct comparison is impossible; for example, AMD's IO is far more complex than Intel's, supporting a wider range of platform designs and IO types. While Intel and AMD both support DDR4\5 at this point within their memory controllers, PCIe 3.x\4.x\5.x is handled inside of AMD's IO while Intel offloads much of that to the chipset, only supporting some lanes inside the CPU, and many CPU's not supporting 5.x. To make matters worse, the Z690 is manufactured on a 14nm process while AMD's IO CCD is 7nm (or something like that) so while the Z690 would use much more power if it did the same thing as the AMD IO CCD, it doesn't. In fact it does very little and the package itself is tiny like 25mm2. This means the AMD IO handles a lot more while being 7nm while Intel integrates all of that into the CPU die which is a more refined 10nm process that efficiency is somewhere between TSMC 5nm and 7nm.
Basically AMD's CPU CCD (5nm?) is more efficient while their IO (7nm?) is less efficient and this doesn't entirely balance out compared to Intel's monolithic 10nm (which more efficient than TSMC's 7nm.)
This chart breaks it down better than I can - scroll down to density comparison: https://www.granitefirm.com/blog/us/2021/12/28/tsm...
ballsystemlord - Thursday, January 12, 2023 - link
It's unclear to me how this relates to my post.I was saying that not all the power is turned into heat, therefore, the 90w figure is not misleading.
ballsystemlord - Monday, January 9, 2023 - link
Also worth noting is that the power dissipated is related to the resistance of the component, not the amount of power going into the component. You should recall that from any basic series circuit you have worked with/on.escksu - Tuesday, January 10, 2023 - link
It's true that it's related to resistance but in the case of the cpu, it's mostly due to the rapid switching of transistors.escksu - Tuesday, January 10, 2023 - link
Not all but most of becomes heat. It's due to the rapid switching on/off of the transistors.bcortens - Monday, January 9, 2023 - link
My annoyance is mostly that it is misleading - while I understand that companies market these things this way the fact of the matter is that the efficiency is not at 65 Watts, its at 90 Watts. Sure they are spectacularly efficient compared to when running them without power limits but the power limits given by marketing are useless for talking about efficiency because they are just numbers on a slide. AMD could call it TDP 50 Watts and it would be just as meaningful because the real number is so far off.Zucker2k - Monday, January 9, 2023 - link
The 90 watts also doesn't account for iod consumption. So, overall chip consumption is higher than 90 watts.mczak - Monday, January 9, 2023 - link
That's definitely wrong. The PPT number, as the name implies (package power tracking) certainly includes the IOD, it's a limit which applies to the socket. (It does of course not include the B650 (or whatever) chipset but that's of course the same with intel.)Zucker2k - Monday, January 9, 2023 - link
PPT is AMD's version of Intel's PL2 or boosting power level.The power limits reported for the chip and iod are separate.
Gigaplex - Monday, January 9, 2023 - link
PPT is for the whole package. The CCD and IOD do report separate power, but the total package power is also reported and is what is governed by PPT.Gigaplex - Monday, January 9, 2023 - link
PPT does include the IOD, but the TDP is pretty close to the actual power consumption of the CCD.yankeeDDL - Wednesday, January 11, 2023 - link
There's no perfect solution. The gap between what "you think" 65W means in case of AMD, and the reality, is still much smaller than in case of Intel's TDP.Inte'ls 125W CPU can reach well beyond 200W, which is insane (nearly 2x). In case of AMD the difference (explained quite clearly) is much lower (30~40%).
Gigaplex - Monday, January 9, 2023 - link
Another thing to add is that the 65W chips use about 65W for the cores, the rest of the wattage is for the IO.Zucker2k - Tuesday, January 10, 2023 - link
PL2 (boost power level) for Ryzen is tdp x 1.35. So, in this case: 65x1.35 = 87.75. The IOD consumes more than what ll be left here. In any case, the best way to measure these types consumption, from my observations, is to measure from the wall. It usually results in far greater discrepancies than software reports.Atari2600 - Tuesday, January 10, 2023 - link
Source please zucker.Zucker2k - Wednesday, January 11, 2023 - link
https://www.anandtech.com/show/17418/amd-corrects-...65x1.35 =87.75ppt
105x1.35=141.75ppt
120x1.35=162ppt
125x1.35=168.75ppt
170x1.35=229.5ppt
juhatus - Monday, January 9, 2023 - link
"Looking at the Ryzen 9 7900, it marginally beats out the Ryzen 9 _7950X_ and sits" - Typo, should be 5950X, above borderlands 3 test on conclusion page.Gavin Bonshor - Monday, January 9, 2023 - link
Thank you for spotting that!MadAd - Monday, January 9, 2023 - link
Did I miss the single core performance test?djsvetljo - Monday, January 9, 2023 - link
I am concerned about Zen4 minimum power draw (idle). Judging by graphs on page 2, it seems like CPU consumes 20 watt on idle???Going back to the Zen4 X comparison vs 13th gen, Intel seems to have advantage at idle. Planning to update home server that sits on idle or very low load 80% of the day, but needs to be on 24/7. I eventually am planing to run VMs on it (rare gaming with GPU pass thru). Thoughts?
duploxxx - Tuesday, January 10, 2023 - link
thoughts? you are creating a wacko system. one site you want to have low low idle, and than intel is better although the latest generation is less difference... its not the cpu, its the total package with pci-e5 etc that is upping the idle power. on the other side you want GPU power which eats already 50% of your total idle power. so the diff between intel and amd will now be only way less vs total idle.. on top for VM you should have an AMD multicore for sure. my advise? wait for the 7000 APU it might fit both your needs.kpb321 - Monday, January 9, 2023 - link
Is it just me or do the 7600x and 7700x just seem like bad deals compared to the newly released chips. $30 moves you up to the 7700 or 7900 respectively with the same rated boost clocks for ST task. Lower base clocks will hurt but the additional cores will help compensate in most MT situations. The 7700 beats the 7600x in many of the multi-threaded tests and does it for a small premium and while using a lot less power in the process.steveofwa - Monday, January 9, 2023 - link
Seems like there is no reason for the "x" versions to exist. They can have their non X with good power efficiency, and the x3d for performance.Byte - Tuesday, January 10, 2023 - link
X is now the "founders" edition where rather than get discounts like kickstarter, you pay for the privilege for getting it first. Nvidia did this with the 2080Ti FE being $1200 whereas msrp was $999. They saw it work well so everyone taking a page. Ryzen traditionally launched with non-X and that was the go to. The X is mostly for emotional reasons which plays a bigger factor in marketing than you think. Most people will never lower themselves to a x700 series card for instance. That is why nvidia did two 4080s in the beginning but backtracked. Wrong move.AndrewJacksonZA - Monday, January 9, 2023 - link
Hi. For "(3-3) Dolphin 5.0 Render Test" you have the test results in your Bench different to the test results in the graph.In the graph:
Intel Core i9-12900KS = 182.2
In your Bench:
Intel Core i9-12900KS with DDR5 = 169
Intel Core i9-12900K with DDR4 = 181
Also, according to your Bench, for "(3-4c) Factorio v1.1.26 Test, 20K Hybrid," the i3-12300 with a score of 72 beats *ALL* the new Ryzen 7xxx CPUs. Can you confirm that please?
I haven't checked any of the other results to your bench but suspect that more might also be different. Also, are efforts underway to update the Bench to include these results, please?
Thank you
The Von Matrices - Monday, January 9, 2023 - link
The multiple tests where the 7900 is beating the 7950X by large (>5%) differences in game tests makes no sense and makes me concerned for the repeatability of the test suite. There is nothing (cache, clock speed, architecture, TDP, NUMA) that is inferior on the 7950X compared to the 7900 but somehow it loses by a large margin in many game tests.ag10n - Monday, January 9, 2023 - link
in your conclusion "Of course, users on a budget may want to pair up a Ryzen 5 7600 with a card such as an AMD Radeon RTX 6600"no RTX on the 6600 afaik ;)
boozed - Monday, January 9, 2023 - link
Those power consumption numbers are amazing.While you can make the argument that a "65W" AMD CPU consuming 90W is misleading, at least it's going to be consistently 90W regardless of which model you choose...
thulle - Monday, January 9, 2023 - link
Doesn't it become really weird to talk about efficiency while only comparing to TDP and not actual power consumption for the load? Not everything hits TTP as yCruncher does either.Preferrably the score in each result should be normalized to actual power consumption, or something similar. Even that has its issues though, since the balance between performance and efficiency is somewhat tuneable.
t.s - Monday, January 16, 2023 - link
Seconded! Or write the AVG power for the task. Ex: Cinebench: 7950X (214W) xxx.xxx pointHarry_Wild - Monday, January 9, 2023 - link
Performance difference is not that much! I going with the 7600 for internet surfing, watching streaming videos and email!😁👍LuxZg - Tuesday, January 10, 2023 - link
I was expecting that all along, so I'm glad it's confirmed. Now just to find sensible AM5 MBO at the right price :-/James5mith - Tuesday, January 10, 2023 - link
Real question: Why do CPUs no longer idle in the 800MHz-1600MHz range? Is there too great a change in the multiplier to hit max turbo at this point? Otherwise, what's the point at idling around 3.6GHz?It seems like a waste of power.
qwertymac93 - Tuesday, January 10, 2023 - link
The cores are gated such that they are at 0hz when idle. The clock shown in Windows is just the speed the core ran at when it last reported.fallaha56 - Tuesday, January 10, 2023 - link
Goodness Anandtech do better, ditch the bizarre memory policy and do some PBO testing…Maverick009 - Friday, January 13, 2023 - link
Linus Tech Tips actually did a review on these chips with benchmarks. The 7900 and 7900X are basically identical chips with one being 65W and the other being 170W, but the 7900 is also unlocked and if you use PBO, it can match its 7900X sibling with marginal performance differences on some programs and the upside being the 7900 is cheaper. I for one think I may pick the 7900 up for a build but I am also waiting on pricing for the 7900X3D and 7950X3D parts for my big gaming PC build that will replace my current 5900X.blzd - Friday, January 13, 2023 - link
I don't get how 7600 is beating 7600X in your F1 2022 and RDR2 gaming tests. By a large margin.It's identical other than lower clock speed and power limit yet performing faster somehow?
Kangal - Tuesday, January 31, 2023 - link
I noticed that too, and frankly, I cannot accept any of their gaming results here.blzd - Friday, January 13, 2023 - link
OC test results would be nice addition.Though not with the bundled heatsink of course haha. AMD included inadequate cooling for these CPU running at 99C under sustained workloads at factory clocks.
Thank you for the review!
Targon - Monday, January 23, 2023 - link
They can't run at 99C since AMD uses temperature as the basis for how high the clock speed will go. The default for the 7900X and 7950X was at 95 degrees C, and the chip won't go more than a degree above that point. With the temperature as the key, the chip will clock itself as high as cooling works, so better cooling results in higher clock speeds, and worse cooling will result in lower speeds, but you will never see temperatures above the max temperature.dicobalt - Wednesday, January 18, 2023 - link
Now if only it was possible to get a modern GPU will a low space heater factor that isn't based on an architecture that's positively ancient.Roy8oh - Friday, January 20, 2023 - link
Bundling coolers with its more affordable CPUs means users can spend their budget on a premium AIO cooler through better memory, storage, or graphics, which will have a positive impact on performance. What do u mean by that isnt aio cooler just cpu cooler liquid obviously but isnt it a cpu cooler or are you refering to fans ?AnitaPeterson - Sunday, January 22, 2023 - link
I am disappointed that the benchmarks do not include more samples from the Zen 3 family, aside from the 5800X3D. With many of us rocking Ryzen 3xxx and 5xxx, such a direct comparison would be much more useful than all the catalogue of horribly power-inefficient Intel chips that most people are not likely to buy anyway...mikato - Thursday, January 26, 2023 - link
"The Ryzen 7 7700 also performs well, but with just 8C/12T"Looks like that should be 16 threads, page 5.
dennphill - Monday, January 30, 2023 - link
OK, so I see a CPU that's a bit less power draw with a price close to last AMD generation. Would change my 5-5600X IF there was a descent motherboard that fits my mATX case! LOUSY selection so far of mATX motherboards from ASUS, Gigabyte, MSI, ASRock, etc. Upgrade needs AM5 motherboard and DDR5 (probably) in addition to 7000-series CPU. B650M expensive or crippled in capability and X670M boards are basically non-existent. Need to wait out until manufacturers decide offering to us need some degree of reasonableness in price and features. O(old)MHO. Thanks for good articles...I hope MB manufacturers read reviews AND comments.iRacer - Wednesday, February 1, 2023 - link
How do you get V-Ray benchmark values which are 35 to 50% higher than anything that can be seen on the results page?What version of the benchmark are you using?
leavenfish - Sunday, February 5, 2023 - link
Ummm....don't think he ever did the testing update he said he would try to do (but was busy) the weekened after the article came out?Gavin Bonshor - Monday, February 6, 2023 - link
It's coming, don't worry.Gavin Bonshor - Monday, February 6, 2023 - link
It isn't the latest iteration. It's the previous version (4). The critical area to consider is the consistency of the results within the same version.Of course, we will update this benchmark in our next wave to the latest version. If we did this for every benchmark, I would spend 100% of my time testing and 0% writing.
blueryu - Tuesday, February 7, 2023 - link
I'd like to know the idle power consumption of the CPUs. Apparently there is quite a difference at idle between the latest AMD and Intel options but can't seem to find much in the way of tests.