Intel i7-8700K Coffee Lake Memory Benchmark Analysis

[W1zzard]

We take a close look at memory speeds, latencies, and command rate on Intel's latest Core i7-8700K with Z370. Scenarios tested include fail-safe 2133 MHz, the platform default of 2666 MHz, and overclocked memory speeds ranging from 3000 MHz to 4000 MHz - at various timings.

Show full review

 

[dcf-joe]

I want to thank you guys for taking the time to do these tests. I found this review very informative.

 

[Gasaraki]

That's a lot of damn speeds and timings to test. Thanks.

 

[EarthDog]

Wow... the LOE to do this was a lot!

Nice to see something we know already proven again with CFL.

EDIT: Something really good to see would be CPU tests in games across multiple levels of cards at multiple resolutions.

 

[Sasqui]

Awesome analysis. Once again, this shows latency is just about as important as speed itself.

 

[mouacyk]

Thanks so much for slogging through this review. 3.5% improvement at best, but I'll always tinker any way -- it's fun.

 

[bug]

Wow... the LOE to do this was a lot!

Nice to see something we know already proven again with CFL.

EDIT: Something really good to see would be CPU tests in games across multiple levels of cards at multiple resolutions.

You never really "know" when it comes to new platforms. We knew CFL hasn't changed the memory subsystem, so we suspected memory speeds won't matter much. Thanks to this article, now we have confirmation.

 

[EarthDog]

I'm glad you are now in the know, bug.

 

[bug]

I'm glad you are now in the know, bug.

I'm happy to have made you glad, dawg

 

[birdie]

So, basically anything above 2666 14-14-34-1T is useless (IOW, the price increase doesn't correspond with performance increase).

 

[HD64G]

Great work @W1zzard! I hope the same test will be done for Ryzen 3,5,7 at least for 5 of the frequecies (2133, 2400, 2933, 3200, 3600). Forgive me if done already and forgot about.

 

[W1zzard]

Great work @W1zzard! I hope the same test will be done for Ryzen 3,5,7 at least for 5 of the frequecies (2133, 2400, 2933, 3200, 3600). Forgive me if done already and forgot about.

https://www.techpowerup.com/reviews/AMD/Ryzen_Memory_Analysis/

No plans for another article, but we'll probably look at it again with AMD's next-gen, whenever that comes out.

 

[hkpolice2]

How about 5Ghz gaming tests /w the uncore maxed out on a GTX 1080Ti? I doubt the stock 4.3Ghz core & 3.7ghz uncore speed is enough to max out the memory @ 3600mhz+

 

[BMfan80]

Thank you for the analysis.

Sorry it's taken me sooo long to say thanx,even the hardware i don't care about is a good read here.

 

[ORLY]

Civilization VI, F1 2017, Fallout 4, Far Cry Primal, Total War: Warhammer?
Would be cool to compare Coffee Lake to Ryzen where the frequencies [could] really matter for Ryzen.

 

[Folterknecht]

Persoanlly I 'd be more interested in min FPS (min 1%) - that aside nice review!

 

[Frick]

Civilization VI, F1 2017, Fallout 4, Far Cry Primal, Total War: Warhammer?
Would be cool to compare Coffee Lake to Ryzen where the frequencies [could] really matter for Ryzen.

If you look at the Ryzen link wiz posted you see Ryzen gains about the same as Coffee Lake, ie slim. At least above 2666Mhz.

 

[Ninjachopstixx]

Thanks for the great article guys, appreciate the hard work on all of your articles!

 

[R-T-B]

If you look at the Ryzen link wiz posted you see Ryzen gains about the same as Coffee Lake, ie slim. At least above 2666Mhz.

We now know however that ram clocks determine infinity fabric speeds, and as such, interthread communication benefits massively from higher speed ram on Ryzen, something not really covered in that review.

 

[TheinsanegamerN]

Seems like low latency 3000 MHz memory would be the best bet, both from peak performance and the variety of choice available. I am curious what the raw bandwidth coffee lake gets from each set speed.

We now know however that ram clocks determine infinity fabric speeds, and as such, interthread communication benefits massively from higher speed ram on Ryzen, something not really covered in that review.

Clearly it doesnt make that much difference. If it did, the multi core encoding benchmarks would have shown massive boosts from higher speed memory. Same with the compression and rendering benches. The performance improvements were in line with what coffee lake gains, suggesting that infinity fabric was not a bottleneck there.

I'm pretty sure AMD would have pushed the multiplier up if there was that much performance left on the floor.

How about 5Ghz gaming tests /w the uncore maxed out on a GTX 1080Ti? I doubt the stock 4.3Ghz core & 3.7ghz uncore speed is enough to max out the memory @ 3600mhz+

Based on? Games have not been memory bandwidth limited for a very, very long time. DDR4 finally pushed the last holdouts, games like supreme commander, to their limits, and it is highly unlikely going from 4.3 to 5 GHz would significantly increase memory bandwidth demand to make the jump from 3000 to 3600Mhz memory noticeable. Going from 2666 to 3000 already shows practically no performance gain.

Game engines are simply not built to handle that much power. to benefit, you would need a game engine built to use that much bandwidth effectively, and such an engine would not work on consoles. Outside of perhaps cloud imperium, I cant imagine anybody modding a PC engine that much.

 

[Manu_PT]

Altho I respect the reviewer for this immense work, the gaming benchmarks are flawed. Dishonored is well known for being limited and you can see it on 720p and 1080p tests where it has basically the same framerate. Hitman also isn´t the best game to bench CPU/RAM. Battlefield 1 and Witcher 3 on the other hand are useful. I would suggest you bench other intensive CPU games that are played on 144hz/240hz like Pubg and Quake Champions for example.

From the tests around the web, never a 3200mhz low latency kit would be on par with a 4000mhz CL19 one. Not when the GPU isn´t the bottleneck (aka with usage below 80%). E-sports players usually get the higher frequency ram kits to sustain better minimum fps. The gaming tests ended up confusing because 2 games shouldn´t be tested at all. And you should provide numbers about minimum fps/0,1%, wich are the most affected by increasing Ram Speed.

 

[EarthDog]

I'm pretty sure AMD would have pushed the multiplier up if there was that much performance left on the floor.

Dont hesitate to go look for testing on the web.

It depends on testing, but there can be no to significant gains... here is an excerpt from the recent anand article in ryzen memory scaling..

It is pretty clear to see that Ryzen can be fairly dependant on memory frequency, but it depends very much on the sort of test and the nature of the workload on memory accesses. On the benchmarks where it matters, our memory kit was above to push performance up and over 20%, although despite the few benchmarks where this happened, it was outnumbered by benchmarks that had zero or a very minor effect. Some gaming titles had up to a 5-10% difference in average frame rates, but others had zero change.

I would suggest you bench other intensive CPU games that are played on 144hz/240hz like Pubg and Quake Champions for example

testing online only games is a terrible idea....

Outside of that, i agree.

 

[Sasqui]

Altho I respect the reviewer for this immense work, the gaming benchmarks are flawed. Dishonored is well known for being limited and you can see it on 720p and 1080p tests where it has basically the same framerate. Hitman also isn´t the best game to bench CPU/RAM. Battlefield 1 and Witcher 3 on the other hand are useful. I would suggest you bench other intensive CPU games that are played on 144hz/240hz like Pubg and Quake Champions for example.

Not saying you're wrong, just that the review is meant to cover a cross section of popular games, not just the ones that are really CPU bound. I think that's fair and also shows which ones are and aren't.

 

[R-T-B]

Clearly it doesnt make that much difference. If it did, the multi core encoding benchmarks would have shown massive boosts from higher speed memory. Same with the compression and rendering benches. The performance improvements were in line with what coffee lake gains, suggesting that infinity fabric was not a bottleneck there.

Clearly? Clearly you are accepting the results without thinking too in detail about them. There's a reason W1zzard himself contradicts them in this very review.

Hint: Encode doesn't use intercore communication much at all.

Other hint: What does infinity fabric do?

 

[Manu_PT]

Not saying you're wrong, just that the review is meant to cover a cross section of popular games, not just the ones that are really CPU bound. I think that's fair and also shows which ones are and aren't.

Fair enough, but we can´t conclude that higher frequency RAM isn´t worth for gaming. Depends on your usage, and again, minimumfps/0,1% are too important on a RAM Gaming test. Also you can´t get CAS 14 kits that easy, not at 3200mhz for sure. You can downclock 3800mhz CAS19 to 3200mhz CAS 14 tho, but that would cost the same money.