Intel Coffee Lake: 8700 vs 9700

Besides $50 plus the increased price for the 390 mobo it looks like the slightly newer technology gives slightly better result - probably not something 99.72% of users would notice. However, the 9700 gets rid of hyperthreading. I’m assuming that’s a good thing but what exactly is the difference with and without hyperthreading, both now and in the future?

And are there any other difference between the two I should be aware of when planning out my new build?

Hyperthreading is a funny thing. In principle it gets you a very slight improvement in performance if you have lots of threads. The idea being that there are execution units in the pipeline that are idle some cycles, and so having a couple of threads co-existing means you can get better use of resources, and get a slight overall improvement in throughput. The down side is that the threads compete for some resources, and it you have workloads that do make really good use of the pipeline, you get zip advantage. Years ago, when hyperthreading was new, we disabled it because it slowed down the processor on intensive numeric workloads. In the modern world, hyper-threading can provide a mechanism for meltdown like exploits. Not something that you are going to worry about on a home gaming machine. But something that worries some users.
Overall hyper-threading is a mixed bag. For straight line compute intensive tasks at best it doesn’t help. For servers with many server threads, it can deliver a free 10% (or so) performance boost by improving utilisation of resources. I suspect it is a wash for gaming machines.

8700 has 6 cores, 9700 has 8 cores. So even without hyperthreading, 9700 may have higher performance for multi-threaded applications. It depends.

My understanding is it’s mainly a marketing (price discrimination) decision. Previously the “mainstream” desktop processors were i3, i5 and i7. So i7-8700 was the top of the line, and hyperthreading was one of the main differences between i5 and i7. There was an i9, but that was in a completely different high-end lineup and had a different socket. With the 9th generation (9000 series), the i9 is now part of the mainstream desktop lineup, so now i9-9900 gets hyperthreading and i7-9700 doesn’t.

One thing needs clarification. There’s a Core i7-8700 and a Core i7-8700K but only a Core i7-9700K (no non-K version). Which of the 8700-series chips are you looking at? For games, clock speed is king, and going much over 4 cores is generally diminishing returns. The removal of hyperthreading is no loss in the 9700K because, as Francis said, its overall multi-core speed improvement is fairly small. For video processing I think it can give about a 20% advantage at best, so more cores are always better than hyperthreading at least until we’re getting into server and render farm territory with 18-core and up processors.

All that said, the 9700K has the best benchmark scores both in single-core and multi-core performance, though it’s a smaller advantage over the 8700K than over the 8700.

Core i7-8700: 3.2 - 4.6 GHz, 6/12 Cores/Threads, Geekbench 5301 Single-Core, 23011 Multi-Core
Core i7-8700K: 3.7 - 4.7 GHz, 6/12 Cores/Threads, Geekbench 5885 Single-Core, 25974 Multi-Core
Core i7-9700K: 3.6 - 4.9 GHz, 8/8 Cores/Threads, Geekbench 6088 Single-Core, 28869 Multi-Core

Was looking at the 8700K. My builds are normally about $2000 and last around 5 years before I need to buy a new one* so part of my build is what I need 5 years from now (exempting adding more RAM ad usually a video card upgrade in a few years).

  • The old ones still works great but a bit sluggish so they are hand-me-downs for family.

Intel’s SMT/Hyperthreading has never been a stellar performer (e.g. 10% to 20% depending on workload), but IBM’s SMT on PowerX cpu’s provides significant gains (2 way about 60%, 4 way and 8 way can be more depending on workload).

But IBM designed/added low level resources to support this and avoid contention (which probably added cost).

If it were me I’d go with the 9700K considering the small difference in price. The better turbo boost speed and two extra real cores more than makes up for the loss of hyperthreading or the slightly lower base clock, which on a desktop machine is unlikely to be a factor (I’ll get to that in a moment).

It’s gotten much more difficult in recent years to adequately compare processors because of the differences in single-core vs multi-core clock speed, turbo boost, and hyperthreading. Not to mention the “megahertz myth” from days of yore that is still true today, wherein a modern processor at the same clock speed as one several years older (or of a different architecture altogether such as X86 vs PowerPC) will almost certainly perform better. It’s the turbo boost and multi-core numbers that really get confusing because it’s not clear without actual benchmarks how the clock rate will throttle down under different load characteristics. In post-Haswell chips different cores can even boost to different clock rates depending on the type of instruction sets they’re processing.

I like to compare the performance of my old 2008 Mac Pro which has (2) 4-core 2.8 GHz Xeon E5462 Harpertown processors. They run at 2.8 GHz under any load no matter what, there’s no turbo boost, no hyperthreading, strictly eight cores and eight threads, though I do believe they can still throttle down under low-load conditions to save energy. My 2017 iMac Pro however has an 8-core 3.2 GHz Xeon W-2140B Skylake processor. On the surface those numbers don’t look all that different, only 14% higher clock speed for the iMac. However, the iMac can theoretically turbo boost to 4.2 GHz which is 50% faster. Nonetheless, the single-core Geekbench score of 5036 for the iMac blows away the Mac Pro’s 1854, showing that the newer processor is getting a lot more done per clock cycle than 10 years ago. The multi-core numbers are also much different, 30842 for the iMac versus 9091 for the Mac Pro.

The Xeon W-2140B appears to be a variant of the W-2145. I’ve never seen it reach 4.2 GHz but with single-threaded instructions it’ll hit 4.0 GHz. Nevertheless, if I throw everything I can at the computer, maxing out a total of 16 threads, it will still hum along at 3.9 GHz with only brief one-second dips to 3.8 GHz (which has been corroborated by others). That’s not bad, and it shows that the thermal design of the computer is doing a good job when it can maintain a speed 700 MHz higher than its rated base clock at full load. It’s actually the same numbers as the W-2145 under an AVX2 workload. Mobile systems can have a much harder time maintaining any turbo boost at all. I’ve observed that any multi-core workload will usually drop them back to their base clock very quickly, even if brief single-core executions can boost up to the processor’s maximum. So for desktop systems with a good cooling system, even the base clock rating is somewhat useless. Confused yet?