AMD Ryzen Discussion Thread.

[R-T-B]

Just use 2 sticks. 16GB should be fine for most workloads anyways.

I'm more curious how much I'll lose so I can make that decision myself. I know 32GBs isn't commonly used but I actually have some workloads that can benefit.

 

[Rehmanpa]

I'm more curious how much I'll lose so I can make that decision myself. I know 32GBs isn't commonly used but I actually have some workloads that can benefit.

I got 32gb in my current PC, and with what I'm planning on doing (hosting game servers and playing at the same time) I'll benefit. Plus it's nice when playing planetary annihilation, that game recommends at least 16gb of ram.

 

[R-T-B]

I admit I don't know much about ram, but what's wrong with using 2x16gb ddr4 sticks and then getting 2 more later?

Ryzen basically doesn't like to use more than 2 "ranks" of ram. The 16gb sticks tend to be "dual sided/rank" (basically two single sided sticks on one stick) and thus are like 4 "ranks" of ram. The IMC speed suffers under this set of circumstances, at least right now.

My question is on AGESA 1.0.0.6, how bad this "penalty" for using 4 ranks will be.

 

[Rehmanpa]

Ryzen basically doesn't like to use more than 2 "ranks" of ram. The 16gb sticks tend to be "dual sided/rank" (basically two single sided sticks on one stick) and thus are like 4 "ranks" of ram. The IMC speed suffers under this set of circumstances, at least right now.

By the time I'd get 64gb of ram I'd hope they'd have fixed it lol, that's hecka far in the future.

 

[R-T-B]

By the time I'd get 64gb of ram I'd hope they'd have fixed it lol, that's hecka far in the future.

Yeah, why I just plan to probably use all my ram. They'll fix it sooner or later I figure.

 

[FR@NK]

My question is on AGESA 1.0.0.6, how bad this "penalty" for using 4 ranks will be.

I thought the agesa update just enabled more ram dividers and sub timings? I dont think it will help much when you are pushing dual rank or 4x modules.

Some guys are claiming at even at slower speeds, dual rank does increase performance. Since you already have 4 sticks of samsung B, maybe you can do some tests? Will be interesting to see what kind of difference in performance and clock speeds your chip is capable of running.

 

[R-T-B]

Some guys are claiming at even at slower speeds, dual rank does increase performance. Since you already have 4 sticks of samsung B, maybe you can do some tests? Will be interesting to see what kind of difference in performance and clock speeds your chip is capable of running.

I'll be glad to report back as soon as I know.

 

[Rehmanpa]

I thought the agesa update just enabled more ram dividers and sub timings? I dont think it will help much when you are pushing dual rank or 4x modules.

Some guys are claiming at even at slower speeds, dual rank does increase performance. Since you already have 4 sticks of samsung B, maybe you can do some tests? Will be interesting to see what kind of difference in performance and clock speeds your chip is capable of running.

What's the difference between dual rank and dual channel ram/quad channel ram? Sorry I really don't know all that much about ram.

 

[FR@NK]

Dual rank modules put a higher load on the memory controller. With ryzen it can greatly effect your memory speeds. And since memory clocks on ryzen are directly tied the infinity fabric speeds it can increase cross CCX access latencies which harms most desktop workloads. But really, if you are just wanting the best bang for your buck with a cheap 8 core system then none of this really matters. If you want ryzen to be close to what intel offers performance wise, then you will need to use single rank samsung B die sticks that are running at least 3200.

 

[Rehmanpa]

Dual rank modules put a higher load on the memory controller. With ryzen it can greatly effect your memory speeds. And since memory clocks on ryzen are directly tied the infinity fabric speeds it can increase cross CCX access latencies which harms most desktop workloads. But really, if you are just wanting the best bang for your buck with a cheap 8 core system then none of this really matters. If you want ryzen to be close to what intel offers performance wise, then you will need to use single rank samsung B die sticks that are running at least 3200.

What kind of ram would you recommend then?

 

[R-T-B]

This list may help you. You generally want single-sided samsung b-die:

https://www.reddit.com/r/Amd/comments/62vp2g/clearing_up_any_samsung_bdie_confusion_eg_on/

 

[toilet pepper]

The K5 is actually a slightly higher SKU than mine, but it seems to lag behind in their bios releases. Not to say it's not an excellent board, just be prepared to get the last update out of the series of boards for whatever reason. One reason I chose the gaming 5 was it gets really early AGESA releases.

As sort of the "TPU Ryzen bios update-guy" around here, I know a thing or two about the software release schedules they tend to follow.

K5 is actually a lower tier than gaming 5 which is odd. It has bclk but fewer phases. I've seen people in giga forums able to hit 2933 with the new agesa.

I will warn you that voltage does not go down when overclock and there are no pstate options yet.

 

[EarthDog]

First off not all games show much of an fps difference, and the games that do are usually about or less than 10 percent. 10 percent, that's like 10 fps vs 11 fps. Who needs 160fps when you have a 145fps

Check out this forum here:
https://linustechtips.com/main/topi...e-boost-does-an-overclock-give-anyway/?page=2

CPU Scaling does not scale linearly with overclocking. An example from this post is, if you overclock your cpu by 20%, you will not gain a 20% performance increase.

10%... although its not a game changer, it solidifies my point there is a difference in most titles... again, that some people dont want. 10 vs 11 fps doesnt matter... but 60 to 66 or 60 to 54 can. Point is... glass ceiling...like i said.

I wasnt talking gaming performance with cpu scaling either. You are correct it doesnt in games. We weretalking cpu benchmarks. Where this all started,
Trparky in post 928, was talking about IPC, which scales pretty linearly with intel. Yoy brought something unrelated into it.

 

[Johan45]

IPC scales very well on Ryzen as well right up till it runs out of gas

This list may help you. You generally want single-sided samsung b-die:

https://www.reddit.com/r/Amd/comments/62vp2g/clearing_up_any_samsung_bdie_confusion_eg_on/

I can't speak for Gigabyte but 4x8GB samsung on CHVI will run as rated at 3200 without issue. Even before the new AGESA I could get 2 mixed samsung B to run without issue.

 

[Rehmanpa]

10%... although its not a game changer, it solidifies my point there is a difference in most titles... again, that some people dont want. 10 vs 11 fps doesnt matter... but 60 to 66 or 60 to 54 can. Point is... glass ceiling...like i said.

I wasnt talking gaming performance with cpu scaling either. You are correct it doesnt in games. We weretalking cpu benchmarks. Where this all started,
Trparky in post 928, was talking about IPC, which scales pretty linearly with intel. Yoy brought something unrelated into it.

Yes but if you notice where that 10% difference is, is at lower resolutions. And there are still quite a few games where they're within like 1fps of each other. The games where the difference is bigger is generally at the point where you have plenty of FPS anyway and it doesn't matter all that much, plus more and more games are getting multithreaded optimizations

 

[r9]

Yes but if you notice where that 10% difference is, is at lower resolutions. And there are still quite a few games where they're within like 1fps of each other. The games where the difference is bigger is generally at the point where you have plenty of FPS anyway and it doesn't matter all that much, plus more and more games are getting multithreaded optimizations

The games that got optimized for Ryzen were optimized for the architecture not for more threads if that is what you are referring to.
If the optimization was towards multithreading the same patch would have increased the fps for 6+ core Intel's not just Ryzen.

 

[purecain]

BTW I keep reading about a wall at 4ghz or around... the only wall people will find, is at 80c when the chip internally throttles. avoid that and you will find your own chips particular limit under your cooling.

realistically I cant see many of us getting beyond 4.15ghz on ambient water due to the voltage needed causing too much heat.

my chip is performing really well and sits at 29c idle and 49c after an hours load. so not warm enough to cause heat creep across the socket either.

I'm pretty much guttered about the memory corruption situation. ive spent a while on the g.skill forums. i'll update here when I hear anything. they say they are working on a solution...

in the meantime I'm left thinking if I should reflash my memory. I'm worried my system wont be able to apply the correct timings one day.

what to do...?

 

[Rehmanpa]

BTW I keep reading about a wall at 4ghz or around... the only wall people will find, is at 80c when the chip internally throttles. avoid that and you will find your own chips particular limit under your cooling.

realistically I cant see many of us getting beyond 4.15ghz on ambient water due to the voltage needed causing too much heat.

my chip is performing really well and sits at 29c idle and 49c after an hours load. so not warm enough to cause heat creep across the socket either.

I'm pretty much guttered about the memory corruption situation. ive spent a while on the g.skill forums. i'll update here when I hear anything. they say they are working on a solution...

in the meantime I'm left thinking if I should reflash my memory. I'm worried my system wont be able to apply the correct timings one day.

what to do...?

What's your 1800x running at? 4.1gh/z is plenty for me to be honest. Hell, I've left my 4790k to stock turbo at 4.36gh/z and it's fine enough for me (not worth the effort for overclocking), so I imagine that an overclock to 4gh/z would be fine.

 

[R-T-B]

If I could get 4.1Ghz with good temps I'd be thrilled to be frank.

Almost done with this Windows setup. OC land is calling me shortly.

 

[GoldenX]

First batch of Phenom IIs could barely reach 3,8GHz in the best of cases, last ones easily hopped over 4,1 with cheap cooling. Future revisions of even this first gen. of Ryzen should reach better clocks.

 

[Chicken Patty]

Looks like I can finally join the club!

 

[purecain]

welcome to the dark side....

ps, if you havnt already. download typhoon burner and browse your memory spd.

the aura programs slowly corrupt the ram, whats not known is if this effects non rgb ram also... although I'm thinking probably not.

ive just ended up having to re-flash mine...

well stock boost is 4.1, only its with 2 cores. ive found 4.07Ghz at 1.395v with droop down to 1.34v to be perfect. no llc(auto). for all cores.
lets see how long it takes for the board to corrupt my memory this time.

 

[Rehmanpa]

welcome to the dark side....

ps, if you havnt already. download typhoon burner and browse your memory spd.

the aura programs slowly corrupt the ram, whats not known is if this effects non rgb ram also... although I'm thinking probably not.

ive just ended up having to re-flash mine...

well stock boost is 4.1, only its with 2 cores. ive found 4.07 at 1.395 with droop down to 1.34 to be perfect. no llc(auto). for all cores.
lets see how long it takes for the board to corrupt my memory this time.

Wait, what corrupts the ram????

 

[HTC]

Wait, what corrupts the ram????

https://www.gskill.us/forum/showthread.php?t=14357

 

[Rehmanpa]

https://www.gskill.us/forum/showthread.php?t=14357

So it looks like RGB ram can get corrupted by the RGB feature? I guess that means that non-RGB ram is not at risk of corruption due to that then.