The difference between the 6000, 7000, 800 and 9000 series Intel processors is ... the numbers 6, 7, 8 and 9. Those are all Skylake-based, and the differences are minimal generation over generation.
I've experiences them all first hand, and therefore I can assure this is wrong.
Between 6700k and 7700k, little difference. There you got a point.
Between 7700k and 8700k, appreciable boost.
Between 8700k and 9700k, appreciable boost, larger than between 7700 and 8700.
All my CPUs are boosted the same way, max turbo speed all the time including with all cores.
I'm only talking about gaming here, as this forum is all about.
You're basically observing that if you take the same cores as before, but add more of them and clock them higher, performance goes up, albeit at the expense of higher power consumption. That's true as far as it goes, but can't drive long-term improvements.
You don't need multiple generations of chips to make that comparison. Just go from a Core i3-8300 to a Core i5-8500 to a Core i5-8600K to a Core i7-8700K and you'll observe about the same thing. Add more cores and higher clock speeds and performance goes up, at the expense of more power consumption.
Intel could have offered 8 core CPUs in their mainstream 7000 series ("Kaby Lake"). They offered a lot more than that in their Xeons and in their HEDT line. They just chose not to. They could have encouraged motherboard vendors to be more aggressive with turbo and ignore the TDP. They just chose not to. Taking off such limitations does allow improved top end performance, at the expense of greatly increased power consumption.
But that's kind of like Nvidia launching a GeForce RTX 3000 series that consists of the 2000 series higher clock speeds and greatly improved coolers to handle all of that extra heat. Faster, yes, but not really the generational uplift that you were hoping for. If the real 3000 series is on 7 nm, then it will be disappointing if they can only improve top end performance by 20%.
Being a process node behind typically means that every time a transistor does something, it uses about 40% more power than if you weren't behind. For decades, AMD has been behind on process nodes. This year, if Zen 2 launches around the middle of this year as AMD says it will, AMD is going to be ahead on process nodes for the first time ever. And that means that if your goal is to keep load power consumption down at a given level of performance, AMD will be your only option for the CPU. Possibly also the GPU, if Navi is good, at least until Nvidia gets to 7 nm. Rumors say that Nvidia won't get there until 2020.
I said total thermal output and that is what I meant, for a given amount of "work" AMD systems have a significantly higher total thermal output.
The laws of physics don't care what logo is on the box that a CPU came in. It turns out that they do care tremendously about things like voltage, current, capacitance, and resistance. What you likely mean is that doing a given amount of work uses more power on an inferior process node than it would on a better one. For decades, AMD has been stuck on process nodes that were inferior to Intel's. That, and not the logo on the box, is probably what led to your observation. This year, that's going to flip around and AMD will be on a superior process node to Intel for the first time ever.
The process node isn't the only thing that affects power output, of course. Some architectures are simply better than others, and AMD's Bulldozer architecture, or going back further, Intel's NetBurst (Pentium 4 and Pentium D) were notorious power hogs. But right now, AMD and Intel are about even on architectural goodness. It's certainly close enough that if either side were a full process node ahead, they'd win by a lot on efficiency.
Unless the issue is that you don't realize that thermal output is the same as power consumption. That's just conservation of energy.
Or if you're talking specifically about chipsets as opposed to CPUs, then you have a point, but probably don't realize what it is.
Thanks for the condescension. Yes I am entirely aware of the physics, probably better than you, having worked in hardware development. But my point is that while both brands are capable of excellent performance, historically systems built with AMD CPU's have a higher measurable thermal output. And this has an impact on room temperature. Glad to hear that AMDs latest generation may reverse that.
Sorry. In my defense, the comment I replied to was kind of baffling as a reply to the comment you quoted.
It's not so much that AMD has greatly improved as that Intel has simply stalled. The move to Ryzen didn't give AMD an architectural lead; it only made them competitive again, and they've been competitive on architecture plenty of times before. The real difference is that AMD has been able to keep advancing on process nodes, while Intel has stalled, as they've been stuck on 14 nm since 2015 and won't be able to move on until around the end of this year.
Sorry, but physical cores in the 9700k are more efficient than in the 8700k.
They're the same architecture on the same process node. The differences are eight cores rather than six, hyperthreading disabled, different clock speeds, and likely different voltages. There might be some slight tweaks for efficiency, as a new die does implicitly let you do a base layer respin. (Or perhaps force you to is the better way to put it.) But the 8700K was already a very mature architecture on a very mature process node;
9700k -- Passmark single thread @ 4.9 - 2822 Passmark multicore 8 (no SMT) @ 3.6 - 17235
So what's the difference here? The 9700 does look more efficient - look, higher multicore score with fewer threads (8 vs 12)!
But... look at single core score - pretty much in line. 4% increase in clock speed, 4% increase in bench score.
SMT cores are traditionally worth about 40% of a single full core.. so we would expect the 8700k to act about like an 8 or 9 core CPU without SMT, but we are still seeing higher scores with the 9700k.
Remember, on the 9000 series, Intel changed the TDP rules, so the 9700k is able to pull a lot more power than the 8000 for same workloads. This additional power under fully loaded conditions due to change in power profile/TDP is what is accounting for the big difference. The 9700k 95W TDP is actually pulling closer to 140W in this case - a sizable jump in power over what the 8700k would be allowed to pull under stock conditions.
That's why the 9700k looks like it's more "efficient" - it's not, it's just much better at auto-overclocking. The per-core architecture and IPS is virtually identical between the two.
That's why the 9700k looks like it's more "efficient" - it's not, it's just much better at auto-overclocking. The per-core architecture and IPS is virtually identical between the two.
I certainly used the wrong term with "efficient", but the 9700k is just a more powerful CPU than the 8700k. And not just because he can clock higher.
It is what I said it was: they got more performance by having more cores clocked higher, and it came at the expense of burning more power. That's all that there is to it, as other than that they disabled hyperthreading on the 9700K, the cores themselves are identical. If you were to disable hyperthreading on the 8700K, disable two cores on the 9700K, disable turbo on both, and fix them at the same clock speed, you'd get identical performance.
Comments
You don't need multiple generations of chips to make that comparison. Just go from a Core i3-8300 to a Core i5-8500 to a Core i5-8600K to a Core i7-8700K and you'll observe about the same thing. Add more cores and higher clock speeds and performance goes up, at the expense of more power consumption.
Intel could have offered 8 core CPUs in their mainstream 7000 series ("Kaby Lake"). They offered a lot more than that in their Xeons and in their HEDT line. They just chose not to. They could have encouraged motherboard vendors to be more aggressive with turbo and ignore the TDP. They just chose not to. Taking off such limitations does allow improved top end performance, at the expense of greatly increased power consumption.
But that's kind of like Nvidia launching a GeForce RTX 3000 series that consists of the 2000 series higher clock speeds and greatly improved coolers to handle all of that extra heat. Faster, yes, but not really the generational uplift that you were hoping for. If the real 3000 series is on 7 nm, then it will be disappointing if they can only improve top end performance by 20%.
It's not so much that AMD has greatly improved as that Intel has simply stalled. The move to Ryzen didn't give AMD an architectural lead; it only made them competitive again, and they've been competitive on architecture plenty of times before. The real difference is that AMD has been able to keep advancing on process nodes, while Intel has stalled, as they've been stuck on 14 nm since 2015 and won't be able to move on until around the end of this year.
Passmark single thread @ 4.7 - 2703
Passmark multicore 6 w/SMT @ 3.7 - 15959
9700k --
Passmark single thread @ 4.9 - 2822
Passmark multicore 8 (no SMT) @ 3.6 - 17235
So what's the difference here? The 9700 does look more efficient - look, higher multicore score with fewer threads (8 vs 12)!
But... look at single core score - pretty much in line. 4% increase in clock speed, 4% increase in bench score.
SMT cores are traditionally worth about 40% of a single full core.. so we would expect the 8700k to act about like an 8 or 9 core CPU without SMT, but we are still seeing higher scores with the 9700k.
Remember, on the 9000 series, Intel changed the TDP rules, so the 9700k is able to pull a lot more power than the 8000 for same workloads. This additional power under fully loaded conditions due to change in power profile/TDP is what is accounting for the big difference. The 9700k 95W TDP is actually pulling closer to 140W in this case - a sizable jump in power over what the 8700k would be allowed to pull under stock conditions.
That's why the 9700k looks like it's more "efficient" - it's not, it's just much better at auto-overclocking. The per-core architecture and IPS is virtually identical between the two.
https://m.hardocp.com/article/2019/01/13/battlefield_v_nvidia_ray_tracing_i99900k_cpu_testing/1
Sure, there are SKUs with more cores, different HT configurations, etc. But it's the same core architecture.
You can go back even farther than that - the core architecture differences between a 6000 and 9000 are miniscule.