AMD Ryzen Discussion Thread.

[cdawall]

So 4-8GB sticks don't work well? That's what would be ideal for my use.

4-8GB sticks are the best actually just make sure it is a single dimm per channel load of single rank memory.

 

[RealNeil]

4-8GB sticks are the best actually just make sure it is a single dimm per channel load of single rank memory.

I have two brands (2-8GB sticks) to try out on the motherboard.
Geil and GSKill. Both are 3200MHz speed, and both are a single sided design.
Whatever works the best, I'll buy two more of.

 

[Aenra]

4-8GB sticks are the best actually

Was thinking that what with Ryzens being dual channel, i'd rather go with 2x16 theoretically; compatibility-wise it's also O.K., as they're single-sided too.
(assuming of course people could push them to 3000ish, hence my enquiring)

That dumb?

 

[notb]

Was thinking that what with Ryzens being dual channel, i'd rather go with 2x16 theoretically; compatibility-wise it's also O.K., as they're single-sided too.
(assuming of course people could push them to 3000ish, hence my enquiring)

That dumb?

Dual-channel means the "dual-channel" thing (whatever it is - lets not get too nerdy ) is applied to pairs of DIMM.
So if your mobo has 4 slots, it actually has 2 pairs of dual-channel slots.

There are a few reasons why you should try to minimize the number of DIMM used (if it can fulfil your RAM size needs):
1) usually lower price ratio ($/GB)
2) less power used, less heat
3) lower stress on the memory controller
4) upgradability

Get 2x8GB.

 

[Aenra]

Let me rephrase it. I meant that it being dual chanel, not quad, i'd have thought i'd be better off fitting all my gigs in two channels, rather than in four. Since only two at a time are interleaved.

Hence my enquiring about frequency differencies, assuming anyone's done it. Hope that's clearer now.

 

[notb]

Let me rephrase it. I meant that it being dual chanel, not quad, i'd have thought i'd be better off fitting all my gigs in two channels, rather than in four. Since only two at a time are interleaved.

OK, I see the issue. Your last sentence is not true.

"Dual-channel" means the memory controller can "join" 2 modules for improved performance. It doesn't mean this is applied to only two at a time. It works in pairs.
If you populate 4 slots, you'll end up with 2 pairs of modules - each working in a dual-channel setup. The "dual-channel effect" will be the same.

 

[Aenra]

Your last sentence is not true

This has confused me from the start.. i am missing simpler, more basic information and i think this is what's causing the confusion
May i ask it differently?

Assume a program requiring 17Gigs of RAM to function. Would there be a difference between a Ryzen rig's having 2x16g @3200 or 4x8g @3200 of RAM?
(because i thought the answer is 'yes', but if i understood you, you're telling me 'no').

 

[Frag_Maniac]

There are a few reasons why you should try to minimize the number of DIMM used (if it can fulfil your RAM size needs):
1) usually lower price ratio ($/GB)
2) less power used, less heat
3) lower stress on the memory controller
4) upgradability

Also less latency with 2 vs 4 sticks.

 

[RealNeil]

Would there be a difference between a Ryzen rig's having 2x16g @3200 or 4x8g @3200 of RAM?

Some people swear that the ~two larger sticks~ solution is best. I always use all of the slots on my boards.
I heard that it's easier to OC with fewer sticks. (something about less stress on the memory controller)

The only system that I've had that was that way was an FX-9590 system. I had to have just two. No more.

 

[Tatty_Two]

Does anyone have any 2x16Gigs to share results with? And if yes, paste their code please

Does this help......

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

 

[cadaveca]

Does this help......

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

Oh look, it's a bunch of G.Skill only.



Ok, well, maybe a few others.

 

[notb]

This has confused me from the start.. i am missing simpler, more basic information and i think this is what's causing the confusion
May i ask it differently?

Assume a program requiring 17Gigs of RAM to function. Would there be a difference between a Ryzen rig's having 2x16g @3200 or 4x8g @3200 of RAM?
(because i thought the answer is 'yes', but if i understood you, you're telling me 'no').

If you think about a performance gain stemming from dual-channel, then NO. In both situations all of your modules will run at a performance granted by dual-channel configuration (so what a module offers + a premium from pairing).

But if you're thinking about general performance, then the 2x16GB will be faster. Less modules mean less work for the memory controller.

Some people swear that the ~two larger sticks~ solution is best. I always use all of the slots on my boards.
I heard that it's easier to OC with fewer sticks. (something about less stress on the memory controller)

Well... people swear different things all the time. Listening to them doesn't really make one smarter.

I know we're all very busy being "computer enthusiasts" - writing posts, doing benchmarks and thinking about hardware upgrades.
Still, why not find a few hours a week in our busy schedules and actually learn something about the computers that are so dear to us?

Once one understands how RAM works (it's enough to know the idea and mechanisms - becoming an electronic engineer is not necessary), it's fairly obvious why less RAM modules are preferable.

 

[RealNeil]

Still, why not find a few hours a week in our busy schedules and actually learn something about the computers that are so dear to us?

Smart-ass,.......

I can't tell for sure if you're a PAB or a BAP

 

[Kanan]

I always started with half-populated Ram slots on my new systems/new mainboards (eg. Socket A, Socket 939, Socket AM2+) and later populated the other half (= full population) and never had any problems or performance decrease. Ryzen is a special thing atm though, for Ryzen I would go with 2x8 or 2x16 (single sided Samsung B) at about 3200 MHz speed. What's so special about it? Well it's maybe the first time in history you have to actually really know which Ram you're buying, meaning OEM and it's exact specifications so you don't buy the wrong ones. Nobody wants Hynix for Ryzen atm. and Samsung E is inferior as well. It's the first time there's so much fuzz about Ram, that I hear of, in my life.

 

[HTC]

I always started with half-populated Ram slots on my new systems/new mainboards (eg. Socket A, Socket 939, Socket AM2+) and later populated the other half (= full population) and never had any problems or performance decrease. Ryzen is a special thing atm though, for Ryzen I would go with 2x8 or 2x16 (single sided Samsung B) at about 3200 MHz speed. What's so special about it? Well it's maybe the first time in history you have to actually really know which Ram you're buying, meaning OEM and it's exact specifications so you don't buy the wrong ones. Nobody wants Hynix for Ryzen atm. and Samsung E is inferior as well. It's the first time there's so much fuzz about Ram, that I hear of, in my life.

It isn't that simple: @ least not @ the moment, which may ofc change as BIOSes mature.

Atm (using Samsung B-die RAM), 2*16 GB 3200 work well with the AsRock Taichi but you'll be hardpressed to do the same with 4*8 GB 3200. OTOH, 4*8 GB 3200 work well with the ASUS CH6 but the same can't be said of 2*16 GB 3200.

Read this post and the following 3 replies.

 

[Kanan]

It isn't that simple: @ least not @ the moment, which may ofc change as BIOSes mature.

Atm (using Samsung B-die RAM), 2*16 GB 3200 work well with the AsRock Taichi but you'll be hardpressed to do the same with 4*8 GB 3200. OTOH, 4*8 GB 3200 work well with the ASUS CH6 but the same can't be said of 2*16 GB 3200.

Read this post and the following 3 replies.

I didn't said it's simple, read the whole post and don't stop after I said "what's so special about it?". I actually went ahead and described further.

 

[HTC]

I didn't said it's simple, read the whole post and don't stop after I said "what's so special about it?". I actually went ahead and described further.

Either i miss understood you or you miss understood me, i think.

The approach you mentioned works for the Taichi but not for the CH6 (not with 2*16 3200 atm), which was my point.

Suggest you read the link i posted.

 

[Kanan]

Either i miss understood you or you miss understood me, i think.

The approach you mentioned works for the Taichi but not for the CH6 (not with 2*16 3200 atm), which was my point.

Suggest you read the link i posted.

I read stuff about Ryzen all the time and I saw the CH6 already do way higher clocks than just 3200 - it does 3200 with pretty much any Ram that is capable. I didn't mention any "approach" in regards to Ryzen aside from buying 2x8 or 2x16 3200 Ram which isn't something special - this should run on multiple boards by now.

In the end, yeah I think you misunderstood me. I just talked about something in general.

 

[Aenra]

Thank you all for your patience and help
[it's why i occasionally ask the same thing in different ways, i know it can seem like stubborness, but things do have a way of coming up]

@HTC thanks for that link, will make an account over there so i can /sub to that thread.

If anyone else has working numbers (freq + model serial) on a 2x16g configuration, do please post them.

 

[notb]

I always started with half-populated Ram slots on my new systems/new mainboards (eg. Socket A, Socket 939, Socket AM2+) and later populated the other half (= full population) and never had any problems or performance decrease.

No one said this is a significant effect. It should be just a few %, so you might have not noticed.
And keep in mind dual-channel also gives fairly small performance gain in real-world scenarios.

Thing is though: these small % add up. So if someone is really going for a highly optimized build, he should also consider such things. I'm not saying this is sensible, but most hobbies aren't.
Let's be honest: people spend hundreds of $ on overclocking, but the actual gain in many tasks (including gaming) is often fairly marginal. One shouldn't waste a lot of this gain on something as silly as wrong RAM choice.

 

[trparky]

This may seem like a really stupid series of questions but I'm going to ask them anyways. Sorry if I come off like a total n00b.

I've heard of the term IPC being thrown about, I know what it means (Instructions Per Clock) but what are the numbers for Ryzen? I've not seen an actual number. Everyone talks about gaming or encoding benchmarks but I've not seen any raw IPC numbers being thrown out. What are they? How do they compare to that of Intel? AMD said that the new Ryzen chip is 50% faster in IPC's than previous chips but 50% faster compared to what number? People have said that Ryzen has less IPC's than Intel but without actual raw numbers I can't even begin to personally compare them. Do these raw numbers mean anything?

Yeah... there's my n00b set of questions for the day. *walks away sheepishly*

 

[bpgt64]

This may seem like a really stupid question but I'm going to ask it anyways. Sorry if I come off like a total n00b.

I've heard of the term IPC being thrown about, I know what it means (Instructions Per Clock) but what are the numbers for Ryzen? I've not seen an actual number. Everyone talks about gaming or encoding benchmarks but I've not seen any raw IPC numbers being thrown out. What are they? How do they compare to that of Intel? AMD said that the new Ryzen chip is 50% faster in IPC's than previous chips but 50% faster compared to what number? People have said that Ryzen has less IPC's than Intel but without actual raw numbers I can't even begin to personally compare them. Do these raw numbers mean anything?

Yeah... there's my n00b question for the day. *walks away sheepishly*

Read the site; https://www.techpowerup.com/212315/amd-zen-offers-a-40-ipc-increase-over-excavator

Been on a 1700x for about a month, getting higher minimum frames and less drop outs than my old 6600k.

 

[trparky]

OK I read that article you ( @bpgt64 ) linked to but I fail to see numbers. I see graphs, sure, but no numbers.

 

[hellrazor]

No one said this is a significant effect. It should be just a few %, so you might have not noticed.
And keep in mind dual-channel also gives fairly small performance gain in real-world scenarios.

Thing is though: these small % add up. So if someone is really going for a highly optimized build, he should also consider such things. I'm not saying this is sensible, but most hobbies aren't.
Let's be honest: people spend hundreds of $ on overclocking, but the actual gain in many tasks (including gaming) is often fairly marginal. One shouldn't waste a lot of this gain on something as silly as wrong RAM choice.

You're smoking too much. Ryzen is kinda fucky, for sure, but outside of Ryzen the time it takes to get data from main RAM is determined solely by the memory controller, operating frequency, and timings - none of which change by simply adding more identical sticks.

 

[r9]

This may seem like a really stupid series of questions but I'm going to ask them anyways. Sorry if I come off like a total n00b.

I've heard of the term IPC being thrown about, I know what it means (Instructions Per Clock) but what are the numbers for Ryzen? I've not seen an actual number. Everyone talks about gaming or encoding benchmarks but I've not seen any raw IPC numbers being thrown out. What are they? How do they compare to that of Intel? AMD said that the new Ryzen chip is 50% faster in IPC's than previous chips but 50% faster compared to what number? People have said that Ryzen has less IPC's than Intel but without actual raw numbers I can't even begin to personally compare them. Do these raw numbers mean anything?

Yeah... there's my n00b set of questions for the day. *walks away sheepishly*

IPC is relative performance Core vs Core at same frequency and its not a fixed number.
On one application Intel might be 20% faster on other Ryzen might be 20% faster the average of all of them would be the IPC but that will always be relative to what you're comparing it to.
In office application rendering and stuff Ryzen seems to be on par in games lacking maybe 10% and this depends what games with what graphics card the test was done etc.