6 ms·
>Add to all of that, this latest batch of CPUs are just stupid power hungry - like 240w+ under load (except for the non-x variants of AMD 7000 series, just rele
by gruez 4y ago
>Add to all of that, this latest batch of CPUs are just stupid power hungry - like 240w+ under load (except for the non-x variants of AMD 7000 series, just released last month).
That's because in the race to get the highest benchmark scores, both companies have set the stock clocks to a level that's way beyond what's optimal (eg. adding 100W of power consumption to get 5% higher benchmark scores). The CPUs themselves are fine, you just have to adjust the power/clock limit lower.
- chx 4y ago^^^^ this Someone on reddit did a power analysis of the 13900K and I reposted it here https://news.ycombinator.com/item?id=34404683 https://news.ycombinator.com/item?id=34404683 https://www.reddit.com/r/hardware/comments/10bna5r/13900k_power_scaling_analysis/ https://www.reddit.com/r/hardware/comments/10bna5r/13900k_po... and it shows the CPU at 100W has 75% of the performance at 40% of the power consumption...
- oblak 4y agoIt's still pretty bad compared to Ryzen 7000. The non-x models are just the regular guys at 65W, like my good ol 3700x. That said, Intel sure has magic dust given how far behind their node disadvantage. Serves them right, though. What a horrible company.
- reisse 4y agoTheir node disadvantage is mostly good marketing from TSMC though. Intel 10 (10nm, but that's not the real nanometers) is ~100M transistors per mm2, while TSMC 5 (5nm, but again, that's not the real nanometers) is ~130M transistors per mm2. Sure, 30% better, but not a 2x improvement as marketing suggests.
- megous 4y ago10nm to 5nm feature size should be [naively thinking] 4x improvement in density, so it's even worse. :)
- paulmd 4y agoActual feature size stopped shrinking a long time ago, though. First they flipped the transistors on their side (FinFET) and now they're embedding them inside a block of gate (GAAFET) to get more out of the same size. But the transistors have not just diverged from their actual names (which happened a long time ago) but have not even really shrunk since like 28nm, there is no dimension of a 5nm chip that is actually 5nm. ;) I know you're poking fun at it, but, node names are completely made up and the points don't matter these days. It is a "5nm-class node", if nodes had continued to shrink according to the original plans in 2000 or whatever.
- Dylan16807 4y agoI will nitpick because the naming is a giant mess: There is no "Intel 10". Intel's old naming scheme had "nm" after every number, and the one you're describing is "10nm". Then they renamed everything to use smaller numbers, and no "nm". Newer versions of "10nm" became "Intel 7", still at ~100M transistors per mm². Similarly "7nm" became "Intel 4", with an expected density of ~160M. And so on.
- 7speter 4y agoA bit of an efficiency advantage though, as seen when you max out zen4 vs alderlake/raptorlake cpus.
- paulmd 4y ago> It's still pretty bad compared to Ryzen 7000. The non-x models are just the regular guys at 65W, like my good ol 3700x. * 65W TDP, 88W PPT This little bit of sleight-of-hand from AMD was really really effective, boost TDP used to be called boost TDP, and Intel expected their TDPs to cover full load when boosting or under AVX. Then AMD made up this new number but it's not TDP it's uh, PPT, and marketed all their chips under a 30% lower number. It's obviously worked because people do this "my CPU is only 65W!" thing all the time and no, it's not, it's 88W, unless you've turned off boost.
- Gigachad 4y agoIn a desktop, hardly anyone cares. Everyone is using AIO coolers that can handle these insane temps. So the 15% improvement is usually worth it, considering the CPU will not even use these powers unless it's actually being used to the fullest.
- chx 4y ago> Everyone is using AIO coolers [citation needed]
- mastax 4y agoIn the summer, at least, I appreciate not having my room heat up so much.
- greggyb 4y agoAIOs do not significantly outperform tower coolers. Their benefit is flexibility of placement. Tower coolers require room for ... a tower. And that placement is both inflexible and in need of airflow through most of the case.
- kerpotgh 4y ago[dead]
- BoorishBears 4y agoHonestly you don't have to adjust the limit: the "power under load" angle just gets completely overblown because people go based on a reviewer's definition of load They might be 240W under extreme load, but I can play AAA titles on my i9 at 240hz barely cracking 50% CPU load. And that's with a 3090, so not exactly a mismatched CPU/GPU situation. At those types of loads the CPU doesn't even try to hit boost clocks most of the time, so you're nowhere near the figures you often see touted based on benchmarks.
- 015a 4y agoYeah; games to this day are still pretty bad at utilizing multiple cores. Any idiot can do the math; the i7 13700k has 16 cores, and an advertised max TDP of 250W. That's a ton of power; but if you're only really stressing 2 or 3 cores the real power draw isn't that high. These chips are so powerful that your bottleneck is almost always GPU, unless you're playing CS:GO at 1080p and aiming for 800fps, so realistically its common to see 25-60% utilization on 1 or 2 P-cores, and the rest just running Windows background shit. This is proven by any outlet which takes the time to measure Performance Per Watt (e.g. https://www.hwcooling.net/en/intel-core-i7-13700k-efficient-choice-for-a-gaming-workhorse-review/37/ https://www.hwcooling.net/en/intel-core-i7-13700k-efficient-...). Intel has consistently driven higher PPW with every generation, when you're comparing like-for-like binned chips. AMD, on the other hand, has been a bigger victim of what the OP is describing; while their raw PPW is generally higher than Intel, so they have room to fall, their Ryzen 7x chips aren't consistently posting higher PPW numbers over Ryzen 5x. In other words; both Intel and AMD are doing well here. If you're stressing every core on the CPU at 100%, you're going to draw a lot of power, but you're also going to be completing workloads much faster than on 12th or 11th gen, so your aggregate power draw will be lower. The low-tier media outlets that post "OMG 250W" aren't doing research, and also don't care to, because they get clicks from tons of people like the OP who eat outrage at face value.
- BoorishBears 4y agoIt's not just low-tier media outlets (at least in terms of reach), this is largely driven by large names like LTT and Gamer's Nexus People don't realize that mid range mobile chips from both Intel and AMD out perform top of the range desktop SKUs from a few years ago because they're constantly bombarded by hot takes based on things like CPU rendering and unzipping files... At this point I'm convinced it's just an informal cycle, where if they actually reported in a realistic manner, no release would be exciting. If they didn't pair bottom of the line CPUs with top end GPUs and odd settings configurations under the guise of "not wanting to reflect a GPU bottleneck", it'd be a lot clearer how badly needs have stagnated vs the speed of these new SKUs