8700K vs 2700X on with 2080Ti [computerbase]

[epsilon84]

You'll likely see bigger swings on Ryzen just going from DDR4-2133 to DDR4-3200. Getting DDR4-4000 to run on any AM4 setup is (currently) nearly impossible.

No doubt Ryzen benefits greatly from faster memory (plus tight timings such as 'Stilts' presets) whereas Intel seems to benefit more from sheer bandwith, as shown by the DDR4-4000 numbers.

A 8700K / 9900K @ 5GHz with DDR4-4000 would be the ultimate combination to go with a 2080 Ti, actually I wouldn't be surprised to see *that* particular setup pull 30% ahead of a Ryzen setup, but you'll never see that in a review because (rightly or wrongly) reviewers will usually stick to DDR4-3200 for both platforms for 'fairness'.

 

[tamz_msc]

No doubt Ryzen benefits greatly from faster memory (plus tight timings such as 'Stilts' presets) whereas Intel seems to benefit more from sheer bandwith, as shown by the DDR4-4000 numbers.

A 8700K / 9900K @ 5GHz with DDR4-4000 would be the ultimate combination to go with a 2080 Ti, actually I wouldn't be surprised to see *that* particular setup pull 30% ahead of a Ryzen setup, but you'll never see that in a review because (rightly or wrongly) reviewers will usually stick to DDR4-3200 for both platforms for 'fairness'.

Getting DDR-4000 to work even on Intel platform is not a matter of joke. You can get decent primary timings, but tweaking secondary ones is not easy. You need the right memory kit, right motherboard, a CPU with a good IMC and the stars to align to get decent latency at 4000MHz and above. That's why reviewers don't usually bother with trying to achieve 4000MHz because 4000CL17 is still worse than 3600CL15 in terms of latency, while being far more difficult to achieve. After a point bandwidth ceases to matter and it only depends on latency - that's why 3600CL15 is more sensible than 4000 with comparatively looser timings.

 

[DrMrLordX]

Getting DDR-4000 to work even on Intel platform is not a matter of joke. You can get decent primary timings, but tweaking secondary ones is not easy. You need the right memory kit, right motherboard, a CPU with a good IMC and the stars to align to get decent latency at 4000MHz and above. That's why reviewers don't usually bother with trying to achieve 4000MHz because 4000CL17 is still worse than 3600CL15 in terms of latency, while being far more difficult to achieve. After a point bandwidth ceases to matter and it only depends on latency - that's why 3600CL15 is more sensible than 4000 with comparatively looser timings.

Also true. On AMD systems, at least you have the benefit of faster IF performance at higher memory clocks, though total system performance will usually suffer if you achieve those clocks with higher overall latency. For most Ryzen and Ryzen2 users, the peak performance is somewhere around DDR4-3466 depending on your kit.

 

[Elfear]

Nope

My 2700X turbos up to 4.35Ghz on multiple cores while gaming. Just fired up Witcher 3 with HWInfo running. That's bone stock by the way except for the Stilt's memory timing settings.

 

[Zucker2k]

My 2700X turbos up to 4.35Ghz on multiple cores while gaming. Just fired up Witcher 3 with HWInfo running. That's bone stock by the way except for the Stilt's memory timing settings.

I believe this is fast switching by the scheduler giving the impression that the cpu is running at an all-core turbo of 4.35Ghz. It is not. The switching is happening faster than the software polling. I see this phenomenon on my Lenovo Y700. Are FXR2 turbo clocks not fixed?

 

[Abwx]

I believe this is fast switching by the scheduler giving the impression that the cpu is running at an all-core turbo of 4.35Ghz. It is not. The switching is happening faster than the software polling. I see this phenomenon on my Lenovo Y700. Are FXR2 turbo clocks not fixed?

That s possible, Hardware.fr said that they had all cores at up to 4.075GHz and they are talking of heavy loading, not of games that use cores at lower than 100% usage, FTR they measured 4.05GHz all core when loading the chip with Prime 95 and 128K FFTs.

 

[beginner99]

If you select the games that DO stress the CPU heavily rely on 1 main thread

fixed. Everyone knows crappy games that are mostly single-threaded will of course run faster on Intel. It's not games that stress the CPU it's games that are poorly coded. But yeah this is one reason I also recommend and chose intel for gaming.

 

[epsilon84]

fixed. Everyone knows crappy games that are mostly single-threaded will of course run faster on Intel. It's not games that stress the CPU it's games that are poorly coded. But yeah this is one reason I also recommend and chose intel for gaming.

That's incorrect, basically *any* game that is CPU bound will generally run faster on Intel, for example:
https://www.techspot.com/review/1655-core-i7-8700k-vs-ryzen-7-2700x/

You can't say AC:O is mainly bound by single thread performance since older gen 4C/4T i5s like the 6600K/7600K struggle with this game.

 

[Zucker2k]

fixed. Everyone knows crappy games that are mostly single-threaded will of course run faster on Intel. It's not games that stress the CPU it's games that are poorly coded. But yeah this is one reason I also recommend and chose intel for gaming.

I see. The underlying uarch has no influence on games, according to you.

 

[epsilon84]

Getting DDR-4000 to work even on Intel platform is not a matter of joke. You can get decent primary timings, but tweaking secondary ones is not easy. You need the right memory kit, right motherboard, a CPU with a good IMC and the stars to align to get decent latency at 4000MHz and above. That's why reviewers don't usually bother with trying to achieve 4000MHz because 4000CL17 is still worse than 3600CL15 in terms of latency, while being far more difficult to achieve. After a point bandwidth ceases to matter and it only depends on latency - that's why 3600CL15 is more sensible than 4000 with comparatively looser timings.

Generally 3600 CL15 kits can be overclocked to 4000 speeds anyway (albeit at looser timings) so it's really up to the owner to determine what level of tweaking they want to do with their RAM. I don't agree that bandwidth doesn't matter past a certain point (just go back to that article I linked, 3600 -> 4000 still shows gains) but I do get your point that if you relax timings too much at higher memory speeds it largely negates any potential performance uplift. It's a balancing act, but I wouldn't be so quick to write off DDR4-4000+ memory on CFL as 'useless', especially with high end RAM kits that can achieve decent timings.

My own personal experience with my 8700K is that bandwidth trumps tighter timings, but I am speaking in the context of DDR4-3200 CL14 outperforming DDR4-2666 CL12, I don't have fast enough memory to try extreme memory speeds.

 

[Abwx]

That's incorrect, basically *any* game that is CPU bound will generally run faster on Intel, for example:
https://www.techspot.com/review/1655-core-i7-8700k-vs-ryzen-7-2700x/

You can't say AC:O is mainly bound by single thread performance since older gen 4C/4T i5s like the 6600K/7600K struggle with this game.

You should have specified on an overclocked Intel, and in the chart above the 2700X is actually downclocked, it would certainly perform better at stock than at a fixed 4.2GHz.

 

[epsilon84]

You should have specified on an overclocked Intel, and in the chart above the 2700X is actually downclocked, it would certainly perform better at stock than at a fixed 4.2GHz.

No...

Stock 2700X runs at 4.05GHz with games that stress all (or most) of the cores. The 2700X is not 'downclocked', it is in fact overclocked 150MHz past its max all core turbo. Unfortunately, AMD does fall away when comparing overclocked gaming performance because it simply lacks any meaningful overclocking headroom - AMD is running it very close to the ceiling already.

Please check this link and compare the overclocked 2700X @ 4.2GHz results to a stock 2700X: https://www.techspot.com/review/1613-amd-ryzen-2700x-2600x/page3.html

 

[JoeRambo]

You should have specified on an overclocked Intel, and in the chart above the 2700X is actually downclocked, it would certainly perform better at stock than at a fixed 4.2GHz.

Oh please, that review was already discussed on these forums and was found to run top end DDR4 3400 memory for AMD with Stilt's timing.

And here You are attacking the clock speeds

 

[epsilon84]

Oh please, that review was already discussed on these forums and was found to run top end DDR4 3400 memory for AMD with Stilt's timing.

And here You are attacking the clock speeds

Exactly, that's actually showing Ryzen in the best light, using the best memory timings instead of XMP, and it's not even 'downclocked'... people will just grasp at straws to 'defend' AMD... smh

 

[Paul98]

Other than some outliers that seem to have some bad data somewhere. The 8700k is a little bit faster in games, which seems already known. While the 2700x has more cores/threads, and is faster in things that use those extra cores/threads....

 

[Malogeek]

It's showing that the 2700X is performing really well considering the 800Mhz difference.

I'm not sure why anyone would bother trying to convince that Ryzen could outperform the 8700k in games. It's never going to due to clock speed capabilities of Intel's chips. 2600X and 2700X show however that it's not necessary to always go 8700k for gaming unless you're pushing extremes of refresh rate.

The price difference and # cores per dollar for productivity is where Ryzen shines and for a majority of gamers, it's just as capable as Intels. There's zero reason for me to have anything higher than my 2600X.

 

[epsilon84]

It's showing that the 2700X is performing really well considering the 800Mhz difference.

I'm not sure why anyone would bother trying to convince that Ryzen could outperform the 8700k in games. It's never going to due to clock speed capabilities of Intel's chips. 2600X and 2700X show however that it's not necessary to always go 8700k for gaming unless you're pushing extremes of refresh rate.

The price difference and # cores per dollar for productivity is where Ryzen shines and for a majority of gamers, it's just as capable as Intels. There's zero reason for me to have anything higher than my 2600X.

I don't think anyone is trying to convince Ryzen beats 8700K in games, though I do understand the critique of the OP's link where the 8700K is shown to be ~30% faster than the 2700X - that kind of gap is simply not possible except in extreme outliers, especially at stock speeds.

I agree that the 2600X (and 2600) are very good value and I'm actually unsure as to what to recommend my cousin for his next gaming build. The 8700K is too expensive so the next tier down is 8600K or 2600/2600X. I think the 8600K is a bit overpriced considering it's a 6C/6T chip and you can get double the threads for cheaper on Ryzen, but in terms of actual gaming performance a 5GHz 8600K is *almost* as good as a 5GHz 8700K, at least in todays games.

 

[arandomguy]

My own personal experience with my 8700K is that bandwidth trumps tighter timings, but I am speaking in the context of DDR4-3200 CL14 outperforming DDR4-2666 CL12, I don't have fast enough memory to try extreme memory speeds.

3200CL14 is lower latency than 2666CL12 (slightly) but with higher bandwidth.

In general regarding memory issue and "normalizing" or making it "fair" between Coffeelake and Ryzen is a bit tricky in my opinion.

Ryzen's "stock" speed is technically faster even though in practice that isn't meaningful for most builders. Coffeelake can achieve higher speeds at a given tier of memory/ease and has a higher peak but then how do you normalize for that in a comparison? So I'm not sure if there is a truly best way to approach this beyond having a lot of data points with various different configurations.

 

[epsilon84]

3200CL14 is lower latency than 2666CL12 (slightly) but with higher bandwidth.

In general regarding memory issue and "normalizing" or making it "fair" between Coffeelake and Ryzen is a bit tricky in my opinion.

Ryzen's "stock" speed is technically faster even though in practice that isn't meaningful for most builders. Coffeelake can achieve higher speeds at a given tier of memory/ease and has a higher peak but then how do you normalize for that in a comparison? So I'm not sure if there is a truly best way to approach this beyond having a lot of data points with various different configurations.

Good points, I don't have any issue with reviewers using the same memory speeds / timings for simplicity sake. Personally I think the way most reviewers do it (via XMP settings) is the 'fairest' way and is how the vast majority of people will set up their systems.

There will always be the argument that you are 'handicapping' AMD by not using custom timing presets like the ones from Stilt. Then there is the counter-argument that Intel can achieve significantly higher (overclocked) memory speeds, but you will never find a review where Intel is running faster memory than AMD - they will be equal at best, or in fact Intel might sometimes be running slower memory if we use 'stock' memory settings - as you pointed out, Intel in fact runs slower memory at stock - 2666 vs 2933.

 

[Elfear]

I believe this is fast switching by the scheduler giving the impression that the cpu is running at an all-core turbo of 4.35Ghz. It is not. The switching is happening faster than the software polling. I see this phenomenon on my Lenovo Y700. Are FXR2 turbo clocks not fixed?

That could be the case although most games will probably get ~4.35Ghz due to not being multi-threaded very well.

 

[Hitman928]

That could be the case although most games will probably get ~4.35Ghz due to not being multi-threaded very well.

With Zen+ and x470 motherboards there is the possibility that you get max XFR2 frequencies on more than 2 cores. Basically AMD made it possible for motherboard makers to let the CPUs boost on more cores and to a higher frequency than "normal" precision boost or even XFR2 settings if they have the VRMs and cooling to allow it. Not sure if that's what's happening with yours or not, but it's possible.

 

[PeterScott]

That could be the case although most games will probably get ~4.35Ghz due to not being multi-threaded very well.

By 5 threads (3 cores) you are down to about 4.1GHz and these days most games will make use of 3 cores/5 threads.

 

[Abwx]

Well, ACO in this obscure and seemingly paid review :

https://www.computerbase.de/2018-09...e-rtx-2080-ti/#diagramm-performancerating-fps


And the review of the new sequel, also at 1080p :

https://www.computerbase.de/2018-10...sorbenchmarks_auf_einer_rtx_2080_ti_fe_update

 

[Elfear]

Well, ACO in this obscure and seemingly paid review :

https://www.computerbase.de/2018-09...e-rtx-2080-ti/#diagramm-performancerating-fps


And the review of the new sequel, also at 1080p :

https://www.computerbase.de/2018-10...sorbenchmarks_auf_einer_rtx_2080_ti_fe_update

Odd results. Can ACO really utilize so many threads that a TR 2950X would show a noticeable improvement over a 2700X?

 

[Abwx]

It s perhaps bandwith related if we are to look at the EHDT perfs from both brands.