Do memory prices actually make Coffee Lake much cheaper than Ryzen?

[Bhav]

As most people already know, Ryzen's performance is very dependant on the memory. To get the best possible performance out of these chips, you need to get ram that's specified at 3200 MHz CL14, or 3600 MHz CL16.

 

On the other hand, Coffee Lake cares very little as long as you manage to get 2666 MHz CL14 ram:

 

 

Currently, 3200 MHz CL14 ram based on Samsung b die chips is extremely expensive - For the G.Skill RGB kit, it costs £450 for 32 GB, and around £250 for 16 GB on both OCUK and Ebuyer.

 

On the other hand, Crucial Ballistix Sport 2666 MHz can be bought directly from Crucial for £327 for 32 GB, or £160 for 16 GB. If you shop around, you can find it for even less (I paid £270 for 32 GB, saving £180 over G.Skill RGB 3200 CL14). Both kits of this ram that I have easily manage 2666-2700 MHz CL14 when increased to 1.35v (alternatively 3000+ at CL16), which is all that the 8700k needs to reach its optimal performance as per this review:

 

https://www.techpowerup.com/reviews/Intel/Core_i7_8700K_Coffee_Lake_Memory_Performance_Benchmark_Analysis/10.html

 

Of course while you can still buy 3200 CL14 for Coffee Lake, it hardly seems worth it for around £150+ more for a few % more performance. On the other hand if you wanted to save money and buy the 2666 MHz ram for Ryzen, you would end up severely reducing your performance in comparison to getting 3200 CL14.

[ARikozuM]

Honestly, you should opt for 3000MHz for either platform. The good thing about Intel is the strong IMC which allows heavy manual overclocking while Ryzen seems to do better with picked kits. If you're looking to buy a system RIGHT NOW, I would say get server-marketed, unbuffered DDR4 as it's generally cheaper than anything with a flashy heatspreader. 

[KarathKasun]

Some games are more memory speed dependent than others.  3000+ helps both platforms significantly in those games.

 

Ryzen tops out at ~3200 right now anyway, which does not cost significantly more than 2400/2666.  Making ram price irrelevant when choosing to buy either platform.  The difference will be less than 10% overall.

 

Also, in my experience, Ryzen seems to OC memory really well.  I was able to get a 2133 (micron) kit running at 2933 simply by changing the memory speed, adding ~3 to all other timings, and changing memory voltage to 1.35v.

[MageTank]

29 minutes ago, Bhav said:

snip

None of this is helpful, considering they only tested average framerates. Memory speed has a significant impact on minimum framerates, which is arguably the most important framerate for ones gaming experience. If they would include a frametime test, or at the very least, monitor the lowest dips on average at each speed, we would at least see a more substantial difference in performance. 

 

I'd say, memory pricing on Intel matters less not because you don't NEED faster ram, but because cheaper ram is easier to overclock for free (if you have a Z series board). Ryzen is far pickier with the types of IC's it needs for memory overclocking, while even the worst Micron kits are easy to OC on Intel. Both still benefit greatly from higher memory clocks, and lower memory latency. 

[Bhav]

1 hour ago, ARikozuM said:

Honestly, you should opt for 3000MHz for either platform. The good thing about Intel is the strong IMC which allows heavy manual overclocking while Ryzen seems to do better with picked kits. If you're looking to buy a system RIGHT NOW, I would say get server-marketed, unbuffered DDR4 as it's generally cheaper than anything with a flashy heatspreader. 

Problem with my 2666 kit is I have to choose between 1.35v CL14 2700 or CL16 3000 (or 2800 / 3200 if I go up to 1.4v which doesn't seem worth it). I can try testing the differences myself, but it would be difficult to measure minimum frames in games.

 

So its more a case of save £150ish and get Crucial 2666, or get ripped off with Samsung b die.

 

For Intel at least, you don't lose as much performance with 2666 MHz ram as you would on Ryzen.

 

I can always sell my current kit and get Samsung b die or better in the future if prices come down, and the difference between the cost and what I'd get back for my crucial ram would likely also be a lot less than the current £180 at that point, but the only reason I wanted the G skill ram was for the RGB, my crucial ballistix kits look amazing and perform well enough for Intel.

[thorhammerz]

3 minutes ago, Bhav said:

Problem with my 2666 kit is I have to choose between 1.35v CL14 2700 or CL16 3000 (or 2800 / 3200 if I go up to 1.4v which doesn't seem worth it). I can try testing the differences myself, but it would be difficult to measure minimum frames in games.

1.4v on the the DRAM voltage isn't going to harm your RAM sticks any time soon.

[Bhav]

1 minute ago, thorhammerz said:

1.4v on the the DRAM voltage isn't going to harm your RAM sticks any time soon.

Right, I came to that conclusion too and will try it out next. 

[Bhav]

4 hours ago, thorhammerz said:

1.4v on the the DRAM voltage isn't going to harm your RAM sticks any time soon.

I hate u now u broke my winblows QQ.

 

1.415v couldn't get 2800 MHz CL14 stable and corrupted my windows to the point that I have to reinstall it. Its all your fault.

 

Gonna leave it at 2700 and never go higher now.

[TitanXp]

5 hours ago, MageTank said:

one of this is helpful, considering they only tested average framerates. Memory speed has a significant impact on minimum framerates, which is arguably the most important framerate for ones gaming experience. If they would include a frametime test, or at the very least, monitor the lowest dips on average at each speed, we would at least see a more substantial difference in performance. 

Hi,

I'm sorry the article in in german. But I hope the graphs speak for themselfs. In essence you get a good bump on both platforms by opting for fastermemory.

Timings are also test but they have a negligible impackt.

 

 

https://www.computerbase.de/2017-07/core-i-ryzen-ddr4-ram-benchmark/

[SupaKomputa]

WOOOOW REALLY SIGNIFICANT!

5 extra FPS for $30 extra dollar!

B3$T B4NG FOR THE BUXXX!

[porina]

IMO for gaming, ram performance is less of a factor than other things. On either platform, I'd aim for higher than base ram speeds where it doesn't come at a disproportionate cost, and think 3000-ish on either system isn't difficult.

[Bhav]

3 hours ago, TitanXp said:

Hi,

I'm sorry the article in in german. But I hope the graphs speak for themselfs. In essence you get a good bump on both platforms by opting for fastermemory.

Timings are also test but they have a negligible impackt.

 

 

https://www.computerbase.de/2017-07/core-i-ryzen-ddr4-ram-benchmark/

Well maybe you should learn to read English better then?

 

Because none of the CPUs on that list are Coffee Lake.

 

Every individual platform responds differently to ram specifications.

[Bhav]

51 minutes ago, SupaKomputa said:

WOOOOW REALLY SIGNIFICANT!

5 extra FPS for $30 extra dollar!

B3$T B4NG FOR THE BUXXX!

Actually £180, not $30. And for maybe 1.5% extra performance on 8700k.

[MageTank]

24 minutes ago, Bhav said:

Well maybe you should learn to read English better then?

 

Because none of the CPUs on that list are Coffee Lake.

 

Every individual platform responds differently to ram specifications.

He was demonstrating the difference in frametime performance on both AMD and Intel platforms. He replied to me as the context of my post was about minimum framerates and frametime, his post makes perfect sense in that regard. The graph you linked is worthless without those metrics.  Spend less time insulting people, and more time reading yourself.

 

3 hours ago, TitanXp said:

Hi,

I'm sorry the article in in german. But I hope the graphs speak for themselfs. In essence you get a good bump on both platforms by opting for fastermemory.

Timings are also test but they have a negligible impackt.

 

 

https://www.computerbase.de/2017-07/core-i-ryzen-ddr4-ram-benchmark/

This is pretty much inline with what I've seen in my personal testing, though it does depend on the title. Most titles see a 10% boost in minimum framerates, while some (Fallout 4 for example) can see upwards of 20% boosts. It depends entirely on the amount of CPU overhead that is present at any given time in the game. Seeing as CoffeeLake is the exact same as Skylake and Kaby Lake from an architecture standpoint, it's safe to assume it scales in the exact same way that they do. 

 

Overall, I'd say most consumers are better off buying the cheapest ram they can if they are on Intel platforms, and just overclock manually. I have tested dozens of kits from every IC on the market, and have yet to find a kit that couldn't do 3000mhz on even the worst IMC's on Intel. It takes extra work, but the end result is satisfying, and can save you a decent amount of money that is better spent elsewhere in the build. For AMD users, it's difficult to make Micron or Hynix kits comply most of the time, and the IMC really prefers Samsung B-Die for easier overclocking. That's not to say you can't overclock Micron or Hynix IC's on Ryzen, you certainly can, it's just going to be far more difficult, and you will likely never achieve the performance that a B-Die kit would have under the same effort.

 

I do believe DigitalFoundry did a test with the 8700k and various memory speeds, and included a frametime graph in the video:

 

Notice the orange line is far more stable than the rest. While it still dips eventually, it still has far less dips than what the others do, which makes for a better gaming experience overall. Nobody really notices when their framerate is steady, but almost everyone notices when their framerate starts to dip. Minimizing these dips in minimum framerates is key to a better gaming experience overall. The fact that it sees a decent boost in average framerates is simply icing on the cake:

[Bhav]

5 minutes ago, MageTank said:

Notice the orange line is far more stable than the rest. While it still dips eventually, it still has far less dips than what the others do, which makes for a better gaming experience overall. Nobody really notices when their framerate is steady, but almost everyone notices when their framerate starts to dip. Minimizing these dips in minimum framerates is key to a better gaming experience overall. The fact that it sees a decent boost in average framerates is simply icing on the cake:

Depends on which one you want to believe then:

 

 

 

[MageTank]

2 minutes ago, Bhav said:

Depends on which one you want to believe then:

 

 

 

I'd have to take DigitalFoundry over TechPowerUp, seeing as DF actually provides a testing methodology to test against. TPU just gives graphs with no context as to how they obtained the numbers, where in the game it was tested, and (the most important part) what the minimum framerates/frametime looked like. 

 

If need be, I'll go pickup an 8700k and Z370 board to test this, though I am pretty sure the result will look identical to what I've already tested on Skylake and Kaby Lake. 

[Bhav]

15 minutes ago, MageTank said:

Overall, I'd say most consumers are better off buying the cheapest ram they can if they are on Intel platforms, and just overclock manually. I have tested dozens of kits from every IC on the market, and have yet to find a kit that couldn't do 3000mhz on even the worst IMC's on Intel. It takes extra work, but the end result is satisfying, and can save you a decent amount of money that is better spent elsewhere in the build. For AMD users, it's difficult to make Micron or Hynix kits comply most of the time, and the IMC really prefers Samsung B-Die for easier overclocking. That's not to say you can't overclock Micron or Hynix IC's on Ryzen, you certainly can, it's just going to be far more difficult, and you will likely never achieve the performance that a B-Die kit would have under the same effort.

 

The problem here is what gives better performance / minimum frame rates on Intel then, 2666 CL14 or 3000 CL16?

 

2 minutes ago, MageTank said:

I'd have to take DigitalFoundry over TechPowerUp, seeing as DF actually provides a testing methodology to test against. TPU just gives graphs with no context as to how they obtained the numbers, where in the game it was tested, and (the most important part) what the minimum framerates/frametime looked like. 

 

If need be, I'll go pickup an 8700k and Z370 board to test this, though I am pretty sure the result will look identical to what I've already tested on Skylake and Kaby Lake. 

 

Do the digital foundry results exist outside of a video, and did they test latencies as well as frequencies? Personally I wont trust any results that have only been posted in a video.

[MageTank]

Just now, Bhav said:

 

The problem here is what gives better performance / minimum frame rates on Intel then, 2666 CL14 or 3000 CL16?

Depends on the amount of overhead, and the title. Tomb Raider for example, scaled better with lower latency, while Metro LL scaled almost linearly with better bandwidth. I would say 2800 C14 was the point where I hit diminishing returns, having gotten the best minimum framerates with the least amount of effort, and anything higher was basically negligible. Granted, I only had a GTX 770 at the time, so I was unable to see differences in average framerates (GPU bottleneck), so I can also go back and test with this 1080 Ti to demonstrate any differences in that regard.

 

Memory overclocking isn't as simple as changing frequency and primary timings. Both of these values impact arguably the most important "timings", which are RTL/IO-L. RTL (Round Trip Latency) is dependent on virtually every timing, from primaries all the way down to tertiary timings, and is also heavily dependent on frequency. If you increase frequency, but loosen your timings, your RTL will either remain the same, or worse, raise in value, which results in either no performance boost at all, or a negative impact on performance. 

 

It's why I recommend people overclock all of their timings manually, in order to get their performance benefits without risking any of it to terrible auto-training from the motherboard. Especially if you have an ASRock board, lol. 

[thorhammerz]

4 hours ago, Bhav said:

I hate u now u broke my winblows QQ.

 

1.415v couldn't get 2800 MHz CL14 stable and corrupted my windows to the point that I have to reinstall it. Its all your fault.

 

Gonna leave it at 2700 and never go higher now.

That's simply your RAM remaining unstable at the given frequency & voltage (and unfortunately for you, corrupting some of your windows files along the way) .

[DarkSmith2]

both techpowerup and digitalfoundry are right and have "accurate" results with there testing methology. (Meaning DF gone overkill in the GPU department)

 

Which also match my personal experience with any plattform since a decade. If you cheap out on the graphics card you will most likely always be bottlenecked by it and will only get measurements within margin of error when testing different RAM speeds.

 

Basically saying:

- if you aim for "beautiful" gaming then ram speeds doesnt matter as much since your game on "Ultra" settings does only increase your minimum FPS in certain demanding areas of a game for a very short amount of time which most likely wont impact your gaming experience by much.  

- if you aim for "performance" gaming then ram speeds do matter alot, since most of the games where FPS really matter are competitive/online multiplayer games and those are being played on low graphic settings or even sometimes lower resolutions to achieve the highest FPS and therefore lowest frametimes.

SO its all about, which games do you play with which hardware and settings, like its always been. Since im almost playing no "single player" games and most of the Multiplayer Onlinegames are in a CPU-bound kind of state, for me high RAM speed is necessary to build a well performing Gaming-PC. 

 

If i would have to build a gaming rig as of Today there is no way around something like this:

https://www.gskill.com/en/product/f4-4133c17q-32gtzr

 

Just because you cant get high bandwith with low latencies work in dual rank without having 32GB Samsung B-Die.

 

 

 

 

[Bhav]

2 minutes ago, DarkSmith2 said:

If i would have to build a gaming rig as of Today there is no way around something like this:

https://www.gskill.com/en/product/f4-4133c17q-32gtzr

 

Just because you cant get high bandwith with low latencies work in dual rank without having 32GB Samsung B-Die.

But its £450+ for 32 Gb. What a waste.

[DarkSmith2]

1 minute ago, Bhav said:

But its £450+ for 32 Gb. What a waste.

Ofc it is from multiple points of view. But its also an investigation. I mean 2011 i also wouldnt thought i would ever need more then 8GB. Imean even if i would be perfectly fine with 16GB right now, i cant see into the future of the next 5years and as i said you only get single rank working with 16GB Samsung B-Die.

[Bhav]

1 minute ago, DarkSmith2 said:

Ofc it is from multiple points of view. But its also an investigation. I mean 2011 i also wouldnt thought i would ever need more then 8GB. Imean even if i would be perfectly fine with 16GB right now, i cant see into the future of the next 5years and as i said you only get single rank working with 16GB Samsung B-Die.

No no.

 

The ram you posted is £450 for 32 GB.

 

The ram I got is £270 for 32 GB.

 

See the difference? £180? 

 

Are you really going to get £180 worth of performance increase with that ram?

[DarkSmith2]

1 minute ago, Bhav said:

No no.

 

The ram you posted is £450 for 32 GB.

 

The ram I got is £270 for 32 GB.

 

See the difference? £180? 

 

Are you really going to get £180 worth of performance increase with that ram?

Yea thats the point, if i play games that are that hard CPU bottlenecked then riding with that more expansive ram gives me performance i cant get elsewhere, GPU upgrades wont help and overclocking the CPU wont help. So yeah from that very specific point of view the performance difference is worth it. I mean it can make a 20%+ FPS difference in some games. 

[Bhav]

2 minutes ago, DarkSmith2 said:

Yea thats the point, if i play games that are that hard CPU bottlenecked then riding with that more expansive ram gives me performance i cant get elsewhere, GPU upgrades wont help and overclocking the CPU wont help. So yeah from that very specific point of view the performance difference is worth it. I mean it can make a 20%+ FPS difference in some games. 

Well thats the thing, I've not seen any information that supports that big a performance gain on any Intel setup, and Coffee Lake seems to have the smallest gains.

 

Also that G Skill ram would be £100 more than the cost of my 8700k. I really dont think Im getting that big a bottleneck with CL14 2700 to need to have spent that much more on the G Skill / samsung b die.