Why TDP mismatch between Skylake-X and Ryzen 5

[Bob Hope]

How is it possible that the stock i7-7800X and Ryzen 5 1600X have the same cores, threads, base and boost clocks but the former has a TDP of 140W and the latter only 95W? That is a massive difference.

 

[moinmoin]

The former is a cut down 10 (or was it 12?) core die, the latter an 8 core one. The former is on a process node optimized for high frequencies, the latter on one optimized for low frequencies at high efficiency.

 

[LTC8K6]

How is it possible that the stock i7-7800X and Ryzen 5 1600X have the same cores, threads, base and boost clocks but the former has a TDP of 140W and the latter only 95W? That is a massive difference.

Intel and AMD define TDP differently.

The way to compare the chips is to have them run the same workload at the same clock speed and see. And even then...

 

[scannall]

How is it possible that the stock i7-7800X and Ryzen 5 1600X have the same cores, threads, base and boost clocks but the former has a TDP of 140W and the latter only 95W? That is a massive difference.

Just a difference in process is all. Usually, the higher a process clocks, the more leakage there is. More leakage=more heat=higher TDP.

 

[Topweasel]

Intel and AMD define TDP differently.

The way to compare the chips is to have them run the same workload at the same clock speed and see. And even then...

Not really. Both are a theoretical required cooling measurement on high workloads and not an actual hard fast top power usage. AMD's sits closer in it's burn in testing because since power usage at 90% load is power usage at 90% load in every scenario. Intel on the other hand basically needs to have a standard TDP and AVX2/512 TDP as the TDP that use needs to include the power usage in a heavy load using those systems.

But the end point is SL, SL-X, and Kaby Lake are geared towards clockspeed's which means even at lower clockspeeds they probably leak more power. That and probably a dozen reasons including AMD using a process mostly meant for ARM and other mobile CPU's which would have much less leakage (a ARM CPU that leaks power like a siv would be useless).

So while neither are actually representitive of their actual power usage. There is a still a difference and 140w worth of recommended heat dissipation is 140w and not 90.

 

[wildhorse2k]

https://www.purepc.pl/procesory/test_procesora_intel_core_i97900x_skylakex_witaj_lga_2066?page=0,41

Ryzen 1800X is not as energy efficient as people think. It becomes very inefficient at high clocks too. On the chart you can see the power draw of various CPUs. At least according to that chart Ryzen at 4.1Ghz is less efficient per core than 7900X at 4.5Ghz when running OCCT Linpack. That would mean at same clocks Skylake-X should consume less power (and in the end produce also less heat) per core.

 

[LTC8K6]

Not really. Both are a theoretical required cooling measurement on high workloads and not an actual hard fast top power usage. AMD's sits closer in it's burn in testing because since power usage at 90% load is power usage at 90% load in every scenario. Intel on the other hand basically needs to have a standard TDP and AVX2/512 TDP as the TDP that use needs to include the power usage in a heavy load using those systems.

But the end point is SL, SL-X, and Kaby Lake are geared towards clockspeed's which means even at lower clockspeeds they probably leak more power. That and probably a dozen reasons including AMD using a process mostly meant for ARM and other mobile CPU's which would have much less leakage (a ARM CPU that leaks power like a siv would be useless).

So while neither are actually representitive of their actual power usage. There is a still a difference and 140w worth of recommended heat dissipation is 140w and not 90.

Yes and if you have the two chips run the same workload at the same clocks, you'll get the real world data that you can use to compare, rather than theoretical numbers. You can even get the actual data for the programs you will be using.

http://www.fanlesstech.com/2016/11/the-intel-tdp-vs-amd-tdp-debate.html

 

[Topweasel]

Yes and if you have the two chips run the same workload at the same clocks, you'll get the real world data that you can use to compare, rather than theoretical numbers. You can even get the actual data for the programs you will be using.

http://www.fanlesstech.com/2016/11/the-intel-tdp-vs-amd-tdp-debate.html

I agree. But that still isn't as simple as "Intel and AMD use different definitions on TDP". The Intel solution uses much more power than AMD's legitimately.

 

[tamz_msc]

https://www.purepc.pl/procesory/test_procesora_intel_core_i97900x_skylakex_witaj_lga_2066?page=0,41

Ryzen 1800X is not as energy efficient as people think. It becomes very inefficient at high clocks too. On the chart you can see the power draw of various CPUs. At least according to that chart Ryzen at 4.1Ghz is less efficient per core than 7900X at 4.5Ghz when running OCCT Linpack. That would mean at same clocks Skylake-X should consume less power (and in the end produce also less heat) per core.

First of all, OCCT Linpack isn't a software you run to do actual things with. Secondly, every piece of silicon has it's sweetspot.

Here's the power consumption and efficiency results running actual jobs like x264 transcoding:

So the 1800X is slightly behind while the 1700X matches the 7900X in efficiency. 7900X is behind it's direct replacement in terms of core count, the 6950X, in efficiency. The 1700 is on another level.

So yes, that Ryzen is extremely efficient running software that actually gets work done isn't something that people make up without validity.

 

[DaveSimmons]

It's too bad the 65 watt TDP 7700 non-K isn't on those charts, I suspect it's up there with the 1700 in efficiency, and the 3.6 - 4.2 clock makes it 20%+ faster for less heavily threaded workloads.

I see the 7700K and these X chips as "factory overclocked" like the RX 580. Trading a lot of power efficiency for a bit more speed.

 

[frozentundra123456]

First of all, OCCT Linpack isn't a software you run to do actual things with. Secondly, every piece of silicon has it's sweetspot.

Here's the power consumption and efficiency results running actual jobs like x264 transcoding:

So the 1800X is slightly behind while the 1700X matches the 7900X in efficiency. 7900X is behind it's direct replacement in terms of core count, the 6950X, in efficiency. The 1700 is on another level.

So yes, that Ryzen is extremely efficient running software that actually gets work done isn't something that people make up without validity.

Can you link the source for that data?

 

[tamz_msc]

Can you link the source for that data?

It's hardware.fr

 

[wildhorse2k]

So the 1800X is slightly behind while the 1700X matches the 7900X in efficiency. 7900X is behind it's direct replacement in terms of core count, the 6950X, in efficiency. The 1700 is on another level.

So yes, that Ryzen is extremely efficient running software that actually gets work done isn't something that people make up without validity.

So lets take your data then.

7900X - 150W max, 10 cores, 15W per core
Ryzen 1800X - 129W max, 8 cores, 16.1W per core

So 7900X wins and that is stock Ryzen with its low clocks.

But we didn't consider the performance in x264 encoding (more points = faster):

Skylake-X 7900X - 27.97 points = 2.79 per core
Ryzen 1800X - 21.62 = 2.7 per core

So 7900X is slightly faster than 1800X at slightly less power consumption per core.

Your last chart confirms the following:
1. Ryzen 1700 is most efficient with its low clocks
2. Broadwell-E is more efficient than Skylake-X
3. Skylake-X is more efficient than Ryzen 1800X

That makes two independent websites that confirm my claim that Skylake-X 7900X is more efficient than Ryzen 1800X despite the opposite being believed.

 

[JimmiG]

So lets take your data then.

7900X - 150W max, 10 cores, 15W per core
Ryzen 1800X - 129W max, 8 cores, 16.1W per core

So 7900X wins and that is stock Ryzen with its low clocks.

But we didn't consider the performance in x264 encoding (more points = faster):

Skylake-X 7900X - 27.97 points = 2.79 per core
Ryzen 1800X - 21.62 = 2.7 per core

So 7900X is slightly faster than 1800X at slightly less power consumption per core.

Your last chart confirms the following:
1. Ryzen 1700 is most efficient with its low clocks
2. Broadwell-E is more efficient than Skylake-X
3. Skylake-X is more efficient than Ryzen 1800X

That makes two independent websites that confirm my claim that Skylake-X 7900X is more efficient than Ryzen 1800X despite the opposite being believed.

Here's an example where Ryzen is more power efficient:
http://cdn.sweclockers.com/artikel/diagram/13604?key=05d25a7e529212e7cd9580eb0710b2ce
(http://www.sweclockers.com/test/23937-intel-core-i9-7900x-skylake-x/22)
Total power usage while rendering Island in Blender

It all depends on the workload. The 7900X is extremely power hungry, but it also has 10 very fast cores, which can sometimes make up for the high power consumption by simply finishing the job faster.

The 1800X clock speed is definitely past peak efficiency of the Zen architecture and 14nm LP process, but it's not intended to be their most power efficient part either. AMD are really aiming for the data center, and you'll notice the fastest Epyc only has a max boost of 3.2 GHz (all-core 2.7 GHz). Nobody in their right mind would put something as power hungry as a 7900X in a data center, it's more of an expensive toy, essentially a factory overclocked Xeon.

 

[moinmoin]

Ryzen 1800X is not as energy efficient as people think. It becomes very inefficient at high clocks too. On the chart you can see the power draw of various CPUs. At least according to that chart Ryzen at 4.1Ghz is less efficient per core than 7900X at 4.5Ghz when running OCCT Linpack. That would mean at same clocks Skylake-X should consume less power (and in the end produce also less heat) per core.

The energy efficient Ryzen 7 is called 1700. The process node used is at its most energy efficient at 3.3 Ghz. Everything above that is quickly losing efficiency. Depending on binning 3.9-4.1 Ghz is the reasonable upper limit of the process node. Both 1700X and 1800X are essentially factory overclocked chips through binning selection.

Image from The Stilt's excellent Ryzen: Strictly technical thread

 

[FIVR]

Intel architecture is a power hog. Even with a more advanced process, they use way more power. Skylake-X is basically Intel's version of Piledriver. They will improve it... hopefully.

 

[frozentundra123456]

So lets take your data then.

7900X - 150W max, 10 cores, 15W per core
Ryzen 1800X - 129W max, 8 cores, 16.1W per core

So 7900X wins and that is stock Ryzen with its low clocks.

But we didn't consider the performance in x264 encoding (more points = faster):

Skylake-X 7900X - 27.97 points = 2.79 per core
Ryzen 1800X - 21.62 = 2.7 per core

So 7900X is slightly faster than 1800X at slightly less power consumption per core.

Your last chart confirms the following:
1. Ryzen 1700 is most efficient with its low clocks
2. Broadwell-E is more efficient than Skylake-X
3. Skylake-X is more efficient than Ryzen 1800X

That makes two independent websites that confirm my claim that Skylake-X 7900X is more efficient than Ryzen 1800X despite the opposite being believed.

Yes, the 7900x uses a lot of power, but it is also very fast. The efficiency varies of course by workload, but it is not as bad as it is being portrayed in these forums, as tamz's data shows. The real problem with the 7900x is not so much power consumption, but temperatures (at least partially due to the use of TIM instead of solder), and the possible problems with cooling the VRMs on the motherboard. Whether the latter issue is a real problem or a sensationalized issue that occurs only vary rarely (if ever) in real use still remains to be seen. Ironically, the 7900x's ability to overclock past the already high stock clocks, while a good thing, pushes power consumption up very fast, a good way to prove the cpu is a "power hog" if one wishes to portray it as such.

 

[tamz_msc]

So lets take your data then.

7900X - 150W max, 10 cores, 15W per core
Ryzen 1800X - 129W max, 8 cores, 16.1W per core

So 7900X wins and that is stock Ryzen with its low clocks.

But we didn't consider the performance in x264 encoding (more points = faster):

Skylake-X 7900X - 27.97 points = 2.79 per core
Ryzen 1800X - 21.62 = 2.7 per core

So 7900X is slightly faster than 1800X at slightly less power consumption per core.

Your last chart confirms the following:
1. Ryzen 1700 is most efficient with its low clocks
2. Broadwell-E is more efficient than Skylake-X
3. Skylake-X is more efficient than Ryzen 1800X

That makes two independent websites that confirm my claim that Skylake-X 7900X is more efficient than Ryzen 1800X despite the opposite being believed.

Noboby mentions the 1800X when discussing efficiency. Did you just ignore the 1700X?

The 1800X in your chart is being operated at the maximum voltage and frequency the CPU will tolerate without LN2. What happens when you do the same with the 7900X?