FRONTPAGE Computex 2018: Extreme Memory Demos from G.Skill

[Overclockers.com]

G.Skill, well known in the enthusiast segment for their high-speed memory, has been at it again and are displaying their highest speed DDR4 memory kits at Computex 2018

Click here to view the article.

 

[Alaric]

I wonder how do-able the Ryzen rig is without full coverage waterblocks. That seems like a huge potential for Ryzen chips.

 

[Woomack]

Most Ryzen motherboards won't even boot at 4000. Many won't work past 3466. What you see is one of maybe 2-3 motherboards on the market that can boot at 4000 and depends on many factors will work stable or not.
Water cooling is not helping Ryzen to run memory at any higher clock. It's only for marketing purposes as it simply looks better. Also 2k Ryzens have the same IMC as 1k series. On the other hand AMD memory bandwidth is not any worse than Intel's what I find a huge improvement. There is higher latency but with fast cache it doesn't really matter in most operations. If only AMD was working better with motherboard manufacturers then we could see improvements everywhere.

 

[Alaric]

So do any of those 2 or 3 motherboards cost less than $270? That CH VII isn't really priced too high, but it's high enough.

 

[Woomack]

I know that Johan was able to run at 4000 on MSI which was used for Ryzen 2600/2700 reviews. I guess that not stable but could boot and make some tests. I think that if one MSI runs then they made other one run at 4000 too. I haven't seen any Gigabyte run at even 3866. Some ASRocks run at 3866 but I haven't seen any 4000 result. From ASUS only CHVII supposed to work at 4000 but all we know is that G.Skill was showing it on the Computex. Some of these stands are with selected CPUs and unofficial BIOS like you simply won't run any memory so high on an average CPU/IMC as they are showing there. My 7900X's IMC ends on 4100/SODIMM, G.Skill is showing 4200, most users can't pass 4000 on ASUS Hero, G.Skill is showing 4600, on 4 memory slots, I can't even boot at more than 4700 on exactly the same MSI Z370 motherboard as G.Skill is showing their 5066 kit, which supposed to run stable (and I have unofficial OC BIOS for this board). I guess you get my point.

I'm not saying it's not possible but there is a low chance you will see it in stores anytime soon or under the same product numbers.

 

[Johan45]

Womack's right I was able to test at 4000 with tight timings, never tried for anything stable as I knew that was a waste of time. Personally, I don't believe a Ryzen rig will run STABLE at that speed just because of IMC/IF gets unstable. Anything over 3600 is near impossible for 90% of the users I have seen. Most who are pushing and tweaking the memory stop around 3466-3533.
That was the MSI board, haven't even tried on Giga but new BIOS shows most timings now which may give me a nudge to do so.

 

[Woomack]

thats 3999, not 4000

 

[Johan45]

thats 3999, not 4000

 

[Alaric]

I was starting to roll around the idea of a Ryzen build in the future (in case I have some money that wasn't spent watercooling AM3+ rigs ), and RAM speed seems to be a big part of getting them to perform up to snuff. 3600-3733 would be nice, so I was wondering whose boards I should be looking at. Interesting turn of fortune with MSI boards. They made the worst AM3+ boards and seem to be rocking the Ryzens.

- - - Updated - - -

thats 3999, not 4000

 

[Johan45]

The MSI have been good to me, I haven't played so much with the Giga so no comment. Still haven't got the CHVII yet so........ I want to hit the mem on Giga maybe tomorrow if time allows and I can tell you more after that but if it's what I think I'll be sticking to the MSI for now.

 

[Johan45]

I can say I don't like trying to tune the memory on the Gigabyte, still doesn't show current timings only stick XMP so I was wrong about that. 3733-3800 barely makes it to Windows on easy timings and still not stable enough to run 3D11 physics test

 

[Woomack]

I was starting to roll around the idea of a Ryzen build in the future (in case I have some money that wasn't spent watercooling AM3+ rigs ), and RAM speed seems to be a big part of getting them to perform up to snuff. 3600-3733 would be nice, so I was wondering whose boards I should be looking at. Interesting turn of fortune with MSI boards. They made the worst AM3+ boards and seem to be rocking the Ryzens.

So far I had luck with ASRock X370 Gaming ITX/ac which runs at 3733 using nothing but XMP profile on Patriot Viper4 memory. Also overclocks to 3866 using couple of Samsung based kits. I guess that X470 Gaming ITX/ac will be the same if not better. Maybe ASRock will send it for tests so I will let you know.
I could recommend X399M Taichi but I guess you won't get TR anytime soon

MSI has some really nice Z170/270/370 and X299 boards. I had no chance to test any X370/X470 board but B350M Mortar could run at 3600 without issues.
Right now I run stability test on MSI Z370I Carbon and memory at 4500 CL20-20-20 1.40V

 

[Alaric]

TR would be nice, but it would be a waste in my hands. Cinebench is all I'd do with it. LOL. A Ryzen w/fast memory would be more useful for me. Honestly, my current rig does the things I do pretty well. The occasional game and web browsing doesn't need a new build, and an 8700k would probably suit my uses better, but AMD is back and I'd like to do my bit to support/encourage that sort of thing.

 

[wingman99]

DDR4-5066MHz is some nice speed.

 

[Woomack]

DDR4-5066MHz is some nice speed.

Now try to run at 4600+ on any retail setup. That's my point. I asked G.Skill how they set "stable" 5GHz+ when I can't even boot at more than 4.7GHz (single stick) using their memory on the same IC, the same motherboard with OC/Xtreme BIOS used for WR stage on the Computex and 2 different CPUs. The only answer was that probably my CPU's IMC is too weak... exactly the same is on new 8086K so I doubt. Somehow I don't believe in 400MHz difference between IMC.
Now think that every motherboard that I know and can run at 4.5GHz+ memory will set 1.45V IO/SA and most of them won't set more automatically. So how high it has to be to set 5GHz+? ... stable ... in dual channel.
On my board it looks like this ... 4500 20-20-20 1.40V 100% stable (runs like that for 2 weeks), 4533+ has problems with booting but I made it to run at 4666 in dual channel in some benchmarks and was able to boot at 4700 using single stick in 2nd memory slot. It was available on unofficial BIOS while latest official release could run stable at max 4400.

 

[wingman99]

Now try to run at 4600+ on any retail setup. That's my point. I asked G.Skill how they set "stable" 5GHz+ when I can't even boot at more than 4.7GHz (single stick) using their memory on the same IC, the same motherboard with OC/Xtreme BIOS used for WR stage on the Computex and 2 different CPUs. The only answer was that probably my CPU's IMC is too weak... exactly the same is on new 8086K so I doubt. Somehow I don't believe in 400MHz difference between IMC.
Now think that every motherboard that I know and can run at 4.5GHz+ memory will set 1.45V IO/SA and most of them won't set more automatically. So how high it has to be to set 5GHz+? ... stable ... in dual channel.
On my board it looks like this ... 4500 20-20-20 1.40V 100% stable (runs like that for 2 weeks), 4533+ has problems with booting but I made it to run at 4666 in dual channel in some benchmarks and was able to boot at 4700 using single stick in 2nd memory slot. It was available on unofficial BIOS while latest official release could run stable at max 4400.

Maybe they ran loos timings with DDR4-5066MHz?

 

[Woomack]

I was trying as loose as my mobo lets me with really relaxed RTL/IOL but it didn't help. It's like a wall past some frequency. I know that Johan tested other GSkill 4266 kit on ASUS APEX with similar experience. Maybe it works only on some really cherry picked kits like one per 100k ... or there are really weird settings in SPD/XMP. I have no idea.
I'm not saying it's not possible but there are many questions, including why in mass sales there is no memory kit above 4600 and why there are only 2 motherboards on which were officially tested all 4500/4600 kits. There is quite big gap between the 4600 and the 5066.

Other thing is how 4x8GB kit works on ASUS Hero at 4600 when I don't know anyone who used this motherboard and was able to set more than 4133 on 2 memory modules stable (some other users had the same experience and are wondering the same thing).

On X299 ITX I can already tell you it's possible to set SODIMM at 4200 as long as IMC is strong. My 7900X runs up to 4100 using Micron based SODIMM. Somewhere there IMC gives up. Stable is 4000, 4100 is benchable ... 4150 can't even boot.