Intel to launch 6 core Coffee Lake-S CPUs & Z370 chipset 5 October 2017

AndreiD · 15 Oct 2017 at 18:00

@Doobedoo According to their testing that's not the case even for Ryzen: https://www.techpowerup.com/reviews/AMD/Ryzen_Memory_Analysis/

Issue with their testing is that they're using only a GTX 1080 FE, and a lot of those games are GPU limited so the performance delta isn't as high.

Doobedoo · 15 Oct 2017 at 18:03

Lol. But their are bigger gains to be had. Never mind it just sheds a different light on those claiming you need fast ram to get the best out of them. So pretty much any ram will give you great performance without the need to spend stupid money above what is already a massively overpriced component.

CAT-THE-FIFTH · 15 Oct 2017 at 18:03

Doobedoo said:
So bar Hitman 2666C14 performs within spitting distance of everything else.

So it's only really Ryzen that benefits from faster ram then.

You did see their Ryzen results too??

Core i7 8700K

Ryzen 7 1800X

Core i7 8700k

Ryzen 7 1800x

TPU Ryzen RAM scaling conclusions said:

So should you make do with slower memory? Memory prices are hot, but the price differences among the various speeds isn't huge. Looking at these results, you could buy a 2x 8 GB DDR4-2400 kit for $100 or a DDR4-2666 kit that's about $15 costlier. DDR4-3200 could be about $40 costlier, which results in a 40 percent increase in price for a 3-5 percent performance uplift. If you're alright with that, then maybe spend another $50 to pick up a 32 GB DDR4-2133 kit for future-proofing (16 GB is sufficient for high-end gaming today).

Click to expand...

https://www.techpowerup.com/reviews/AMD/Ryzen_Memory_Analysis/13.html

If you ignore the 720p results,according to both their articles you only see a 5% to 6% increase in FPS going from 2133MHZ DDR4 to 3200MHZ DDR4 at 1080p for both Coffeelake and Ryzen.

Whether you believe the Ryzen results is one thing though,as other reviews showed RAM speed was helpful.

Looking at the DF review,taxing games like The Witcher 3 saw a 5GHZ Core i7 8700k was slower with 2133MHZ DDR4 than a stock clocked chip running with 3000MHZ DDR4,but with some other games it was only a few percent difference.

So its going to be game dependent,but I think we might need to look at more articles as time progresses.

AndreiD · 15 Oct 2017 at 18:06

DF also use a Titan X for testing I recall, while TPU only use a GTX 1080, so the performance delta will be higher on the DF testing since they'll be less likely to be GPU bottlenecked.
Either way, with current RAM prices, getting the cheapest 3000~3200 kit seems to be the way to go.

CAT-THE-FIFTH · 15 Oct 2017 at 18:11

AndreiD said:
DF also use a Titan X for testing I recall, while TPU only use a GTX 1080, so the performance delta will be higher on the DF testing since they'll be less likely to be GPU bottlenecked.
Either way, with current RAM prices, getting the cheapest 3000~3200 kit seems to be the way to go.

I would say so too,as you can get 3GHZ/3.2GHZ sets for not much more than the 2400MHZ/2667MHZ ones if you shop around.

Doobedoo · 15 Oct 2017 at 18:13

AndreiD said:
DF also use a Titan X for testing I recall, while TPU only use a GTX 1080, so the performance delta will be higher on the DF testing since they'll be less likely to be GPU bottlenecked.
Either way, with current RAM prices, getting the cheapest 3000~3200 kit seems to be the way to go.

You would have to say a 1080 is a more realistic portrayal of the average gaming rig, very few people will be using a £1200 graphics card. So realistically anything below 3200mhz will be just fine.

CAT-THE-FIFTH · 15 Oct 2017 at 18:20

Doobedoo said:
You would have to say a 1080 is a more realistic portrayal of the average gaming rig, very few people will be using a £1200 graphics card. So realistically anything below 3200mhz will be just fine.

TBH,even with Ryzen I would stick to 3GHZ/3.2GHZ RAM too,as the cost increases stupidly and its not in line with the performance jump you get. The only thing is you need to be more weary of the motherboard you get if you want to run faster RAM,as one which has a decently long QVL will mean you can avoid the double ranked stuff which can be a bit hit and miss it seems. I suppose Intel will be a bit less finicky in that regard!

Doobedoo · 15 Oct 2017 at 18:20

CAT-THE-FIFTH said:
You did see their Ryzen results too??

After I posted

.

Journey · 15 Oct 2017 at 18:23

Doobedoo said:
You would have to say a 1080 is a more realistic portrayal of the average gaming rig, very few people will be using a £1200 graphics card. So realistically anything below 3200mhz will be just fine.

The average gaming rig is far below the GTX 1080, more like GTX 1060/RX480 and GTX 1070 at a push. At 3200MHz LL is where it seems to be excellent, but I'm going to do some more testing at 2933MHz to see what the LL timings offer, not that is makes much difference price wise, about £10-15 from 3000 to 3200. Blooming RAM prices, should be back to normal in 2019 if Korea haven't gone to war.

Doobedoo · 15 Oct 2017 at 18:26

CAT-THE-FIFTH said:
TBH,even with Ryzen I would stick to 3GHZ/3.2GHZ RAM too,as the cost increases stupidly and its not in line with the performance jump you get. The only thing is you need to be more weary of the motherboard you get if you want to run faster RAM,as one which has a decently long QVL will mean you can avoid the double ranked stuff which can be a bit hit and miss it seems. I suppose Intel will be a bit less finicky in that regard!

I would agree but something like 8packs 3200 kit costs £200, whilst you can find other 3200 kits between £160-170. £30 for very little difference I've even seen 3000 for £135-150.

CAT-THE-FIFTH · 15 Oct 2017 at 18:29

I think at the beginning it was more of an issue,but with later AGESA versions and a good motherboard with a solid BIOS,you can probably have a good chance of getting around 3GHZ maybe even a bit higher with normal stuff(but you still need to do your research). The B-die stuff is still the best for pushing 3.2GHZ and above consistently with Ryzen or if you want super low latencies,but not sure if its worth the cost. Intel is probably more plug and play.

Doobedoo said:
I would agree but something like 8packs 3200 kit costs £200, whilst you can find other 3200 kits between £160-170. £30 for very little difference.

I suppose that is not too bad,but for instance I saw some "cheapo" 3200MHZ stuff for £143 when helping someone with a build,and with a few tweaks it was running at 3066MHZ in a MSI Tomahawk. OK,not quite 3200MHZ(might need a bit of messing around to do that),but still nearly £60 than a better set for example.

Doobedoo · 15 Oct 2017 at 18:38

CAT-THE-FIFTH said:
I think at the beginning it was more of an issue,but with later AGESA versions and a good motherboard with a solid BIOS,you can probably have a good chance of getting around 3GHZ maybe even a bit higher with normal stuff(but you still need to do your research). The B-die stuff is still the best for pushing 3.2GHZ and above consistently with Ryzen or if you want super low latencies,but not sure if its worth the cost. Intel is probably more plug and play.

I suppose that is not too bad,but for instance I saw some "cheapo" 3200MHZ stuff for £143 when helping someone with a build,and with a few tweaks it was running at 3066MHZ in a MSI Tomahawk. OK,not quite 3200MHZ(might need a bit of messing around to do that),but still nearly £60 than a better set for example.

It's quite a difference though for a lot of people. That's potentially £60 towards the cost of a graphics card.

Samsung b is obviously better for ryzen but not necessarily worth it for the majority and Intel users can use anything 2666 and above and get perfectly good results. It seems their stronger IMC has practically no real world benefits.

Silent_Scone · 15 Oct 2017 at 19:03

Doobedoo said:
It's quite a difference though for a lot of people. That's potentially £60 towards the cost of a graphics card.

Samsung b is obviously better for ryzen but not necessarily worth it for the majority and Intel users can use anything 2666 and above and get perfectly good results. It seems their stronger IMC has practically no real world benefits.

Depends what speed you intend to run, so that's an invalid statement. Bdie is still the best choice on Intel platforms. Most bdie kits have considerable guardband, so they're a worthy investment if looking to overclock a fairly low bin further.

Doobedoo · 15 Oct 2017 at 19:07

Silent_Scone said:
Depends what speed you intend to run, so that's an invalid statement. Bdie is still the best choice on Intel platforms. Most bdie kits have considerable guardband, so they're a worthy investment if looking to overclock a fairly low bin further.

Going by the techpowerup gaming benchmark that was posted in the last page. It seems that from 2666 and above their is only 1-2% in it. So while b die is better, is it worth the premium?

Doobedoo · 15 Oct 2017 at 19:10

Doobedoo said:
Going by the techpowerup gaming benchmark that was posted in the last page. It seems that from 2666 and above their is only 1-2% in it. So while b die is better, is it worth the premium?

Silent_Scone · 15 Oct 2017 at 19:38

Not exactly a wealth of information there...

Doobedoo · 15 Oct 2017 at 19:41

Silent_Scone said:
Not exactly a wealth of information there...

The link to the article has already been posted, that is what we have been discussing.

Silent_Scone · 15 Oct 2017 at 20:00

Doobedoo said:
The link to the article has already been posted, that is what we have been discussing.

Your point was that you can buy any memory regardless of IC on Intel platforms and expect good results, that as a statement is grossly incorrect which is what I was responding to. Gains are far more linear on AMD platforms than on Intel in terms of gaming performance, however, bdie kits generally overclock with good margin. Far more margin than what's possible on AMD platforms. Your comments about needed expense depend on the users budget. After all, it's not all about gaming performance. Even if it was, the metric for those results isn't always the best one. Minimums and frame times are also affected by overclocking memory.

Here is GSKILL CAS14 3200Mhz 4x8GB running at CAS16 4000 @ 1.4v

Doobedoo · 15 Oct 2017 at 20:10

I'm going by what the benchmarks states. Their testing shows that from 2666mhz to 4000mhz the performance difference across the games they tested only differs by 1.5%. They are testing coffeelake which is a mainstream platform targeting gamers so that is why were discussing gaming results.

If you have results that show techpowerup to be wrong then by all means post it up, otherwise that is what we have to go by. Anandtech have also tested memory scaling but are due to follow up with a tighter timings article.

Silent_Scone · 15 Oct 2017 at 20:17

Doobedoo said:
I'm going by what the benchmarks states. Their testing shows that from 2666mhz to 4000mhz the performance difference across the games they tested only differs by 1.5%. They are testing coffeelake which is a mainstream platform targeting gamers so that is why were discussing gaming results.

If you have results that show techpowerup to be wrong then by all means post it up, otherwise that is what we have to go by. Anandtech have also tested memory scaling but are due to follow up with a tighter timings article.

What's the metric for the 1.5%?