performance trends across CPU generations in terms of energy use
performance trends across CPU generations in terms of energy use
I invested considerable time searching for a thorough evaluation of CPU energy use versus processing capability. Techpowerup offers efficiency data in their latest CPU assessments, but only for chips released after the 7th Intel Core series or the first AMD Ryzen line. How do Haswell chips stack up? What about Sandy-Bridge, Broadwell, Skylake? AMD Bulldozer? All saw major improvements in manufacturing precision, especially with Intel’s shift to 14 nm. Craft computing recently shared a video comparing the E5 2667 v4 Broadwell to a Ryzen 7 3700X in gaming performance, though it didn’t capture power usage. This remains valuable proof that you can enjoy titles like Borderlands 3 or The Witcher 3 without overspending. While processor comparisons are common, understanding efficiency trends is especially relevant today—especially amid energy challenges and hotter summers that raise A/C costs. Do you have any references for such insights? I know cpubenchmark.net exists, but its readings rely on TDP values, which are often unreliable now. Let me know if you think it could help! Thank you for your help!
They’ll all perform worse, particularly Bulldozer. I’d also suspect older CPUs struggle more with low power states; my GTX 970 used 16-20W when idle, while the RX 6600 XT only needed three to four. That’s why I felt more accurate back then. For a rough estimate, TDP works, but expect older chips to consume significantly more power than their newer counterparts. @Lord Nicoll mentioned this back in 2017—remember when the R5 1600AF matched the X5670’s speed in Cinebench while drawing just 69W?
This i7-4770K delivers a solid performance with an average score of 3350 in CBR23, consuming around 100W. In comparison, the newer i5-13600K offers a score of 24000 with only 125W usage, giving it a score/watt ratio of about 240. This means the 4770K is quite efficient, achieving roughly a score per watt of around 6, which is impressive.
Use reference points that track peak power usage and software metrics such as Cinebench. If maximum draw matches, a higher Cinebench score suggests better efficiency. Expect today’s processors to be more efficient thanks to tighter manufacturing. They should also perform better when idle because of improved sleep modes.
Right, but the X5670 has a TDP of 95W so it's not exactly an incredible jump in efficiency, what's that, 50%? I would be hoping for more. Indeed it is! However that is if the score was achieved at 125W, which it was most likely not, closer to 190W according to TechPowerUp, which immediately brings this to 4x. This is still a gigantic difference though and shows that anything older than Haswell is probably not a good idea if power consumption is a problem for you. The E5-2667 v4 though is another story, at 135W TDP (which hopefully is it's actual power draw) has a CBR15 score of 1448 and the 13600K of 3652 while the 4770k gets 709 for slightly less power consumption making the 13600K little more than twice as efficient as the 2667 v4 (somebody should do a CBR23 test on that thing to be sure) and 4x as efficient as the 4770K. So, maybe the magic point actually wasn't Haswell, but Broadwell. And anything older than Broadwell is too inefficient to feasibly buy, especially if you also need to buy RAM for it. That seems to be a good idea since power draw and Cinebench figures seem to be on all CPUs at some point. Thanks! Also, I know that AMD's CPUs seem to have a much better efficiency than Intel right now, at least in multithreaded workloads because according to TechPowerUp, their single-threaded efficiency can sometimes be 2 times lower. Or maybe it has something to do with sleep states?
I made a deliberate choice to contrast two gaming chips. Server chips consistently outperform because they avoid overclocking and boost, achieving a 10% speed boost with only a 50% increase in power use. It seems energy efficiency on those chips matched similar performance gains using smaller silicon manufacturing processes.
Still, 50% isn't anything to scoff at. I could probably use a Wraith Stealth, less so on a X5670 (even besides the socket incompatability). There's also less heat and PSU/VRM strain
That is a good point, however we can use that exact knowledge to our advantage when we see that for gaming these chips are actually very much perfect. Maybe getting 14nm chips is "good enough" for the most part. Or maybe it makes sense to go higher a generation or 2, Skylake is still very popular. Well yes, I was just saying I was expecting more from 7 years of innovation.
It seems the discussion centers on comparing server and desktop performance, highlighting how older 2010 systems can still compete with newer 2017 processors despite being overclocked.