Ryzen's Infinity Fabric Clock Speed is Linked to Memory Clock Speed... Might Explain Why Memory OCs Make a Noticeable Impact in Performance on Ryzen?

[Syntaxvgm]

3 hours ago, Droidbot said:

I can make shit up and I'd more reputable than wccftech's sources

CNN is more reputable the WCCFtech somehow 

 

[Syntaxvgm]

3 hours ago, RadiatingLight said:

This is basically the only picture I've seen stating that performance increses this much with faster RAM.

 

I overclocked my sticks from 1600Mhz to 2000Mhz, (and tightened timings) but only got an 8% performance boost in benchmarks like unigine heaven or metro last light.

So for a 25% increase in memory speed you got a 8% performance increase? that's actually impressive, as far as memory overclocking goes imo. If this is true this leads me to believe memory may have been overlooked in terms of importance for benchmarks. 

[Tam3n]

3 hours ago, Humbug said:

It looks weird how the fps gains increase more at the higher memory speeds. One would guess that you would first see bigger gains as you alleviate the bottleneck and then diminishing returns, but now it looks exponential instead... Anyhow Ryzen tweakers probably want to go loose timings high memory speed route.

[Ashaira]

So now i have a reason to watercool and OC my ram? PERFECT

[Citadelen]

6 hours ago, themctipers said:

-snip-

Because it's cheaper than a single monolithic 8 core die.

--------------------------------------------------------------Also, AMD made a consioucs decision to do this, Jim Keller is one of the best CPU architects in the world, if there was something better he could realistically do it would've been done. Seeing all these comments I can't help think this is relevant here...

[MageTank]

6 hours ago, RadiatingLight said:

Well it's a comparatively small price to pay for the 50% lower cost over intel 8 cores.

 

Depending on how the Ryzen quad cores are built, they could either suck or be really good.

if they are a single CCX, then it would be freaking awesome. on the other hand, if they are 2 CCXs with only 2 cores enabled on each, then that would really fucking suck.

^ although they would still be good CPUs don't get me wrong.

You are making the assumption that Intel CPU's cost 50% more simply because they cost 50% more to make, and not because Intel overpriced them due to the fact that they could get away with it.

 

Just because Intel sells a $1000 CPU, doesn't automatically make it worth $1000. Nor does it make any other similarly specced CPU worth $1000. 

 

7 hours ago, DocSwag said:

What are your thoughts?

My thoughts are very dumb. In fact, I'll link my thoughts from another thread. 

8 hours ago, MageTank said:

Don't mind me, three dudes arguing about the CCX. I am just gonna sit here by myself and spew some random thoughts in my head until someone chimes in to correct me.

 

What if this Disney™ Infinity Fabric isn't just connected to the clock speed of the ram itself. What if it's actually connected to the IMC's RTL? Huge stretch, I know, but bear with me. If the RTL itself has an impact on the Infinity Fabric (since RTL is dictated not only by your frequency, but by CAS, Command Rate, Ranks, and nearly every tertiary timing); this would make sense as to why AMD locked down the modification of any of these timings, as it could potentially break absolutely everything connected to them. I have zero evidence for this, and zero knowledge of the CCX's and this Infinity Fabric, but at this point, I am grasping for straws. 

 

There has to be a reason beyond laziness or oversight for them to not let us touch tertiary timings. We've had access to them for as long as I could remember. Long before most of us even knew what they were for. Why fail to include them now, unless there was a serious reason for doing so? They included them on Carrizo, and that wasn't long ago.

 

Man. This forum needs a shock collar that zaps me every single time I mention Ryzen's IMC. I know people have to be tired of hearing about it by now, lol. 

[Bananasplit_00]

6 hours ago, DocSwag said:

Because it's easier. Cache and stuff gets really complicated the larger it is and it would probably add at least a month or two in development to make a single CCX eight core die. Maybe it's something we'll see with Zen2 though probably not.

Don't want to ruin your hopes but the hexa-cores are 3+3 and quad cores are 2+2. And yes, this is confirmed straight from AMD. Honestly it makes sense from a yields standpoint since if it was 4+0 if there were two cores on separate CCXs that were bad then you couldn't do sh*t with that die but now you could make that into a six core CPU. With 2+2 there's 36 ways they could arrange that but with 4+0 it's 2 ways.

 

EDIT: But yeah it sucks from a performance standpoint though less cores will be utilizing the fabric so maybe its bottleneck will be less noticeable. And maybe AMD will eventually have 4 core dies.

that makes me wish for an extreme Ryzen quadcore model that IS 4+0 tbh because that would be awesome 

[N3v3r3nding_N3wb]

8 hours ago, themctipers said:

I didn't expect RYZEN to be two 4 core things connected by infinity fabric (I'll call it micro-pcie, ok?)

 

but why..?

It probably has lots to do with modularity. They're using this one architecture to scale all the way from the very low end (mobile) to the very high end (servers, etc.).  They also have their semi-custom business.  This modularity should decrease costs of development and production.

[N3v3r3nding_N3wb]

8 hours ago, themctipers said:

*sighs*

Why the sigh?

[That Norwegian Guy]

Man, imagine if they made 6 cores the upper limit and designed such a SKU on a single core complex. That would be so much better for everyone.

 

Still going to get a 1600X mind you.

[jmababa]

Yeah right but you cannot run windows 7 or 8 in Rhyzen 7 as said here http://www.pcworld.com/article/3181814/windows/microsoft-says-its-blocking-windows-7-8-patches-on-latest-amd-intel-chips.html so I'm sticking with my AMD Rhyzen 5 1600x

[The Benjamins]

8 hours ago, DocSwag said:

Because it's easier. Cache and stuff gets really complicated the larger it is and it would probably add at least a month or two in development to make a single CCX eight core die. Maybe it's something we'll see with Zen2 though probably not.

Don't want to ruin your hopes but the hexa-cores are 3+3 and quad cores are 2+2. And yes, this is confirmed straight from AMD. Honestly it makes sense from a yields standpoint since if it was 4+0 if there were two cores on separate CCXs that were bad then you couldn't do sh*t with that die but now you could make that into a six core CPU. With 2+2 there's 36 ways they could arrange that but with 4+0 it's 2 ways.

 

EDIT: But yeah it sucks from a performance standpoint though less cores will be utilizing the fabric so maybe its bottleneck will be less noticeable. And maybe AMD will eventually have 4 core dies.

The 1500x can't be a 4 core die due to having 16MB cache the 1400 could be a 4 core die based on specs. I think we may see them in the R3 release in 2H

[DocSwag]

8 hours ago, Hunter259 said:

The first one is fine but then you screwed up by using the memory speed instead of the IMC/Fabric speed @DocSwag. The 2933 DDR4 would make the fabric = 256*(2933/2 as 2133/1066 ~= 2 thus 1467Mhz)*1/8 = 47GB/s

Na I did it right hence the "/2" after the 2933 MHz in my post . And I ended up with 47 GB/s so...

1 hour ago, The Benjamins said:

The 1500x can't be a 4 core die due to having 16MB cache the 1400 could be a 4 core die based on specs. I think we may see them in the R3 release in 2H

Yeah ik I was just talking about the 4 core CPUs in general.

 

EDIT: wouldn't it be ironic if in the end the quad core r5 CPUs were worse than the r3s for gaming?  

[Morgan MLGman]

10 hours ago, Belgarathian said:

They're a 2+2 arrangement, and hex-core are 3+3. 

 

Sorry to disappoint. 

That's actually not confirmed anywhere IIRC. Though it's very likely because from what I know, that one CCX supports one memory channel so you need two to allow RAM to go in a dual-channel mode.

 

10 hours ago, themctipers said:

I didn't expect RYZEN to be two 4 core things connected by infinity fabric (I'll call it micro-pcie, ok?)

 

but why..?

Infinity Fabric is not micro-pcie, it's more of an equivalent to Intel's QPI link which is used to connect CPUs in dual-socket PC's. Though Infinity Fabric is quite a bit faster, especially with a memory OC. It's faster than QPI even with RAM at 2133MHz and that bandwidth increases hugely when a memory OC is applied.

[DocSwag]

25 minutes ago, Morgan MLGman said:

That's actually not confirmed anywhere. Though it's very likely because it's possible that one CCX supports one memory channel, so you need two to allow RAM to go in dual.

 

Infinity Fabric is not micro-pcie, it's more of an equivalent to Intel's QPI link which is used to connect CPUs in dual-socket PC's. Though Infinity Fabric is quite a bit faster, especially with a memory OC. It's faster than QPI even with RAM at 2133MHz and that bandwidth increases hugely when a memory OC is applied.

Nope, it's confirmed. Straight from the Anandtech article about Ryzen 5:

Quote

We have confirmation from AMD that there are no silly games going to be played with Ryzen 5. The six-core parts will be a strict 3+3 combination, while the four-core parts will use 2+2. This will be true across all CPUs, ensuring a consistent performance throughout.

 

[Morgan MLGman]

6 minutes ago, DocSwag said:

Nope, it's confirmed. Straight from the Anandtech article about Ryzen 5:

 

Okay, then my prediction seems accurate Good to know

[RadiatingLight]

11 hours ago, Humbug said:

isn't 8% considered a good gain from a memory overclock? In the pic I posted they had gone all the way up to 3200Mhz. Obviously not every game will gain big but what needs to be tested is whether the outliers where ryzen is underperforming vs intel do gain...

 

even AMD is prioritizing this. 5:20

https://m.youtube.com/watch?v=yWdhLXl5a5s

but the image showed actual improvements between 3000Mhz and 3200Mhz, which if they do exist, are miniscule (not 10fps like it showed)

[pas008]

maybe this can drive 4xxxmhz ram prices down?

[DocSwag]

13 minutes ago, pas008 said:

maybe this can drive 4xxxmhz ram prices down?

If anything it would drive it up since more people would be buying so they would raise prices. Anyhow, Ryzen is having issues going much higher than 3200 MHz or so and I haven't seen any go above 3600 MHz so it's not even like people can use it.

[pas008]

13 minutes ago, DocSwag said:

If anything it would drive it up since more people would be buying so they would raise prices. Anyhow, Ryzen is having issues going much higher than 3200 MHz or so and I haven't seen any go above 3600 MHz so it's not even like people can use it.

supply and demand

if they fix the ram speed issues with bios updates or etc

and it increases performance, many companies might only try to produce more higher frequency ram which will result in lower prices

we seen ddr3 shortages few times which resulted in increased price but after many companies flooded the market it dropped and was really cheap again

[DocSwag]

1 minute ago, pas008 said:

supply and demand

if they fix the ram speed issues with bios updates or etc

and it increases performance, many companies might only try to produce more higher frequency ram which will result in lower prices

we seen ddr3 shortages few times which resulted in increased price but after many companies flooded the market it dropped and was really cheap again

We haven't seen any Ryzen CPUs reach over 4 GHz memory. While a bios update could help, you can't change the IMC and chances are it is what is really limiting in the end, so I doubt that we'll see it happen.

[NumLock21]

Leak clip showing how AMD assigns the number of cores per CCX for their Naples 32 core cpu!

 

Each CCX contains 4 cores. They decided to disable 3 of them and just use 1 per CCX. So for Naples the configuration comes out to 1+1+1+1+1+1+1...

 

 

[WMGroomAK]

15 hours ago, DocSwag said:

Source:

Or alternatively, the picture from AMD:

So basically, it seems that the clock speed at which the infinity fabric used to connect the CCXs in Ryzen is governed by the memory clock speed. People have noticed quite noticeable performance improvements from overclocking memory on Ryzen, and this could be part of the reason. I'm not too sure, but it could also potentially partially explain why Ryzen is having a hard time overclock past around 3600 MHz, though I'd guess that it's more of an IMC thing rather than the Infinity Fabric not being able to run at those clock speeds.

 

If we do a little math, we find that the transfer rate of the infinity fabric between CCXs is 256bits*1066 MHz*1byte/8bits = 34112 MB/s or about 34 GB/s. However, if you just overclock the memory to 2933 MHz (which seems to be a popular overclock for Ryzen, not sure why), the transfer rate becomes 256 bits*2933 MHz/2*1 byte/8 bits = 46928 MB/s or about 47 GB/s. This is quite a substantial increase in bandwidth and I'm guessing that this explains some of the increase in performance from memory overclocking, and tells us that Ryzen may not be as memory bandwidth starved as we initially thought. Honestly, though, I wish they had made the two have been separate so that the infinity fabric could be OCed separately. What are your thoughts?

I had been thinking for a little while now that AMDs Infinity Fabric may be both their strength and weakness with the new platform...  It's giving them flexibility in the way they can layout and design their chips and I can see this benefit coming into play when they begin releasing the Zen based APUs, however, the cross CCX L3 cache and thread handoff is definitely an issue...  Hopefully there can be some software and firmware updates that help to mitigate some of this in the current Zen generation as well as adding in support for faster RAM clocks.  I'm also hopeful that as they further implement the Infinity Fabric, they can increase the transfer speeds in the Zen2 or Zen3 processors.

[Spenser1337]

Amd fix this stuff please thanks bye 

[Ekin]

15 hours ago, Humbug said:

snip

 

15 hours ago, RadiatingLight said:

This is basically the only picture I've seen stating that performance increses this much with faster RAM.

 

11 hours ago, Tam3n said:

It looks weird how the fps gains increase more at the higher memory speeds. 

 

The RAM speed picture is just marketing bullshit...