AMD Ryzen has issues with high frequency RAM! - Fix will come in 1-2 months

[Starelementpoke]

9 minutes ago, Enderman said:

I said almost

True

[Doobeedoo]

Was just thinking which RAM should I go for for next build heh. 

[Colonel_Gerdauf]

1 hour ago, That Norwegian Guy said:

2133Mhz RAM doesn't have to be JEDEC compliant though, there are CL10 and CL11 DIMMs out there.

 

I'd choose 2133Mhz at CAS 10 over 2400 and CAS 15 in a heartbeat.

 

Though it's not a necessary choice, as there are 2400Mhz CL11 options.

Just to make it clear, what you are referring to is DDR3 (where the JEDEC standard is 1333MHz at CL9, if I am not mistaken), whereas I was talking about DDR4 which are "slower" in terms of the clock-to-latency ratio.

[That Norwegian Guy]

Just now, Colonel_Gerdauf said:

Just to make it clear, what you are referring to is DDR3 (where the JEDEC standard is 1333MHz at CL9, if I am not mistaken), whereas I was talking about DDR4 which are "slower" in terms of the clock-to-latency ratio.

no, I'm talking about DDR4.

 

Looking at price aggregator for Norway there's 8 options of 2133Mhz DDR4 RAM at CL11-12

[Trixanity]

I hope there is a proper fix in the pipeline and not just a 'we'll fix X but it will also break Y in the process' deal.

 

If a microcode update fixes it, I don't see the problem. Remember Sandy Bridge? Motherboards were recalled and they also had to release the Z series motherboards to replace the P series.

[zMeul]

here we go 

and those people who pre-ordered the darn things without knowing all the details .. serves them right

[Dabombinable]

AMD always has had a rather poor IMC.

[RS2007GOD]

No one can afford anything faster than 2133 anyway since ram prices skyrocketed.......

[suchamoneypit]

4 hours ago, bgibbz said:

RIP. AMDs long awaited launch has a critical flaw, sounds like every single game company out there. 

What critical flaw? Very few users have a setup that could encounter this issue.

[Space Reptile]

[LAwLz]

1 hour ago, MageTank said:

I can actually demonstrate this if need be.

Demonstrate what? The part about how a higher CL such as CL16 can still be lower latency than CL14?

You don't even need to demonstrate it. As soon as people realize that the number is not a measurement of time, but clock cycles they should realize why higher frequency RAM often has a higher number (but not necessarily a higher latency).

Like how 3200MHz CL18 has the same latency as 1600MHz CL9 (11.25 nanoseconds).

 

Do you have any good reads about sub timings? I only really know about CL.

 

 

On topic: Good thing I was planning on getting 2x16GB sticks, if Ryzen turns out to be good.

[sof006]

BIOS updates are nothing new and fixes like this are nothing new.

[corsairian]

AMD is playing the long term. We really need to see reviews and Quarterly reports on sales and adoption etc.

[Dabombinable]

6 minutes ago, LAwLz said:

Demonstrate what? The part about how a higher CL such as CL16 can still be lower latency than CL14?

You don't even need to demonstrate it. As soon as people realize that the number is not a measurement of time, but clock cycles they should realize why higher frequency RAM often has a higher number (but not necessarily a higher latency).

Like how 3200MHz CL18 has the same latency as 1600MHz CL9 (11.25 nanoseconds).

 

Do you have any good reads about sub timings? I only really know about CL.

 

 

On topic: Good thing I was planning on getting 2x16GB sticks, if Ryzen turns out to be good.

I love how my DDR2 1066 @ 5-6-6-18 had far lower latency then when I set it to 800 @ 4-4-4-12.

[MageTank]

12 minutes ago, zMeul said:

here we go 

and those people who pre-ordered the darn things without knowing all the details .. serves them right

Serves them right for what? The ram will still work, it just won't work at their XMP speeds. To be fair, I didn't expect any Intel XMP certified kit to run perfectly out of the box, simply because those kits were not tested with AMD's IMC in mind. Anyone that expected their XMP's to load on an AMD CPU should perhaps read up on how XMP works in relation to DOCP or whatever AMD is calling it these days.

 

That being said, your anti-AMD crusade is never going to end, is it? 

26 minutes ago, That Norwegian Guy said:

no, I'm talking about DDR4.

 

Looking at price aggregator for Norway there's 8 options of 2133Mhz DDR4 RAM at CL11-12

Can I get a link to those kits? I've certainly never heard of any DDR4 2133 C11 kits yet. I know GeiL was supposedly slated to release some, but I've yet to see those hit the market. As for 2133 C12, those are very common in Mushkin's lineup, but they are by no means cheap in the US. We are talking $150 for 16gb of 2133 C12. That's a garbage price if i've ever seen one. 

 

Just now, LAwLz said:

Demonstrate what? The part about how a higher CL such as CL16 can still be lower latency than CL14?

You don't even need to demonstrate it. As soon as people realize that the number is not a measurement of time, but clock cycles they should realize why higher frequency RAM often has a higher number (but not necessarily a higher latency).

Like how 3200MHz CL18 has the same latency as 1600MHz CL9 (11.25 nanoseconds).

 

Do you have any good reads about sub timings? I only really know about CL.

 

 

On topic: Good thing I was planning on getting 2x16GB sticks, if Ryzen turns out to be good.

Reads? Not really... Most of what I know is through trial and error, but I do use both Samsung and Hynix's whitesheets, combined with some reference images to better help me understand individual timings and their relation to each other. 

 

As for the whitesheets for your individual IC's, you can find those simply by googling DDR4 (IC Manufacturer) and you will find all of the info you need. It really helps you understand how each IC acts when using tighter or looser tertiary timings. Let's use tRDRD_dr and tRDRD_dd as an example. Hynix prefers both of these timings to be 0, or as close to 0 as physically possible. Samsung on the other hand, prefers 5-6. Your bandwidth and latency will actually be negatively impacted if you run your Samsung kit at different values. The same can be said of Hynix.  

 

Now, for the topic, I do want to say this. After re-reading the information, I have no idea if ASUS is referring to XMP (DOCP or whatever it's called for AMD) being unable to access tertiary timings to properly adjust them for specific kits (or more specifically, load the values from the memory themselves) or if they are unable to adjust them manually. If it's an issue with XMP, then it won't be that big of a deal. Plenty of memory overclockers around to help people around that issue. If it's an issue with the microcode of the CPU completely blocking access to all tertiary adjustments, then it's a bigger issue that we simply cannot solve ourselves. Hopefully it's just an XMP issue, and that we can test Ryzen to it's full potential. 

[ace_cheaply]

4 hours ago, cluelessgenius said:

its not even out yet and theres already problems? man im soo not gonna pre order.

You're clueless, genius.

[jaggysnake57]

meh 2400 is gonna be fine for now its not like you can only run 2 sticks.......i doubt its gonna be an issue for most people

[Colonel_Gerdauf]

42 minutes ago, That Norwegian Guy said:

no, I'm talking about DDR4.

 

Looking at price aggregator for Norway there's 8 options of 2133Mhz DDR4 RAM at CL11-12

I need a link for reference, because I have never hear of that XMP configuration, but there are CL10 and CL12 for 2400MHz from Corsair. Not only do they not exist in 16GB DIMMs, but they are quite a bit slower in practice and more expensive than the 3200MHz-CL14 kits from G.Skill (which has 16GB per stick as an option). Not exactly a good deal, all things considered.

[Curious Pineapple]

A team of engineers spent 4 years putting over 2 billion transistors into a square inch of silicon, and there is a small, correctable bug that will affect a small percentage of users. Oh noes it's a failure and not worth buying, get an i5 instead so you can use 4 sticks of high speed RAM without a microcode update. AMD suck and can't make CPU's.

 

Let's see if the LPC clock keeps going after 16 months and doesn't cost billions in replacing faulty hardware.

 

[WereCat]

46 minutes ago, zMeul said:

here we go 

and those people who pre-ordered the darn things without knowing all the details .. serves them right

 

How many people will actually use more than 2 sticks of RAM?

Besides, you can always cancel pre-order if you find out there is issue you do not want to deal with.

[zMeul]

1 minute ago, WereCat said:

How many people will actually use more than 2 sticks of RAM?

Besides, you can always cancel pre-order if you find out there is issue you do not want to deal with.

most people will not know about this

and the amount of people that buy 4x kitts, even for Intel, is quite large

[| Rembo |]

4 hours ago, huilun02 said:

But of course anyone making a Ryzen build this late would demand no less than 32GB of memory, thus the necessity to use all 4 slots.

 

Who makes new builds with only two slots occupied anyway? Its unacceptable even if it allows 3200+Mhz. This alone justifies the extra $500+ for the i7-6900K. 

if you are making a build right know and you want 32 u can just run it at lower speeds for a few weeks or you can just buy 2 sticks of 16gb and leave 2 dimms open for future upgradability

[WereCat]

Just now, zMeul said:

most people will not know about this

and the amount of people that buy 4x kitts, even for Intel, is quite large

 

I know that many people use more than 2 sticks of RAM, however it usually is because they are on old platforms and have like 4x 4GB RAM sticks or 4X 2GB RAM sticks.

Now, most people just tend to pop in 2x8GB and call it a day. I am sure that since most people that are looking at 8c/16t CPU do some workloads which may require more than 16GB RAM, may get affected by this.

 

It is not like Intel never had any launch issues with CPU too.

[zMeul]

Just now, WereCat said:

It is not like Intel never had any launch issues with CPU too.

so what, does that makes it right? 

[| Rembo |]

That feel when all the AMD haters had all their hate bottled up the last few weeks because everything ryzen related was good news and now something that won't change anything in the long run and that affect a very small percentage of people in a very small way has come up they are blowing up.