AMD/Intel Processors

[GoldenLag]

5 hours ago, KarathKasun said:

The ring will be going away if they are forced above 10 cores in the mainstream.

nope, they can use ringbus to 12-14 cores. intel is node and modularity limited, not buss limited. 

5 hours ago, KarathKasun said:

Technically different companies, so there is promise.

RTG and AMD CPU devision are the same company. they just operate relativly seperatly with more close cooperation since IF being a thing

 

5 hours ago, TahoeDust said:

Sorry, I think everyone knew what I meant.  But damn, you got me good.  

gotta be picky. 

5 hours ago, Plutosaurus said:

12 and 16 core chips are overkill for 99% of users. 

same with 8 cores, but it doesnt stop people when we finally have rapid progression in performance. 

[KarathKasun]

Just now, GoldenLag said:

WE DONT KNOW ANYTHING FOR CERTAIN. if you hype up a product you will be dissapointed, we dont know what Zen 2 is actually capable of.

We do know the basic IF architecture is getting a major overhaul due to the divorced setup.  To determine weather or not this is a net positive and by how much, we will have to wait and see to find out.  I do not imagine AMD taking a step backwards though.

[Mister Woof]

2 minutes ago, GoldenLag said:

gotta be picky. 

same with 8 cores, but it doesnt stop people when we finally have rapid progression in performance. 

I don't really feel we are getting rapid progression in performance, just more parallelization

 

But I agree, which is why I'm perfectly happy with my 6/12s

[GoldenLag]

1 minute ago, KarathKasun said:

We do know the basic IF architecture is getting a major overhaul due to the divorced setup.  To determine weather or not this is a net positive and by how much, we will have to wait and see to find out.  I do not imagine AMD taking a step backwards though.

all we know its a net loss in latency.  

 

and as such we should take everything with a spoon of salt untill actual launch

[KarathKasun]

1 minute ago, GoldenLag said:

nope, they can use ringbus to 12-14 cores. intel is node and modularity limited, not buss limited. 

RTG and AMD CPU devision are the same company. they just operate relativly seperatly with more close cooperation since IF being a thing

The way it is structured financially, RTG is treated as an independent entity last I checked.  They do operate more closely than, say, intel and Qualcomm for example due to both being under the AMD umbrella though.

[GoldenLag]

Just now, Plutosaurus said:

We're not really getting rapid progression in performance, just more parallelization

i mean, sort of what i mean.  

 

atm AMD is forcing the game of paralization. something they physically cant loose at. we have a doubling of core over 2 years essentially. which is neat

[Mister Woof]

1 minute ago, GoldenLag said:

i mean, sort of what i mean.  

 

atm AMD is forcing the game of paralization. something they physically cant loose at. we have a doubling of core over 2 years essentially. which is neat

The hurdle is waiting for software to catch up

 

4/8 has been the high end standard for over a decade, and it's only now that it's starting to become the edge of insufficient

[TahoeDust]

4 minutes ago, Plutosaurus said:

I don't really feel we are getting rapid progression in performance, just more parallelization

 

But I agree, which is why I'm perfectly happy with my 6/12s

 

The way AMD convinced people that more "slow" cores that communicate slowly with eachother was better for the average user/gamer is a marketing masterpiece.

 

[GoldenLag]

2 minutes ago, TahoeDust said:

 

The way AMD convinced people that more "slow" cores that communicate slowly with eachother was better for the average user/gamer is a marketing masterpiece.

 

um... your forgot your /s

 

its kinda important, 

[KarathKasun]

1 minute ago, TahoeDust said:

 

The way AMD convinced people that more "slow" cores that communicate slowly with eachother was better for the average user/gamer is a marketing masterpiece.

 

Specs are unimportant, resulting performance and how much it costs is important.

This is why FX was a failure.  Resulting performance was a significant hindrance in most cases.  With Ryzen, performance is good enough in games and outstanding in embarrassingly parallel scenarios.  MT scaling is in its favor compared to equal core/thread parts from Intel.

[Stefan Payne]

5 hours ago, KarathKasun said:

IF = mesh if ya didnt know.

No, as AMD does something in between, based on the Console designs they have and are proven.

They put 2 4 Core Dies together and interfaced that with Infinity Fabric.

And IIRC the per CCX latency is on par or lower than Intel, where it hurts is CCX to CCX, but that might also change with the upcoming Zen2 design...

 

Right now you can say that Single Thread is slightly better for Intel and the biggest advantage is Latency and Frequency. If those things get fixed, Intel is in a world of hurt...

And What people also tend to forget:
Ryzen is still a new Architecture and even if it performs rather well, there are some performance hits in the design because its not a "proven" design based on a 10 year old thing that got incremental improvements.

So there are some things AMD can do - and already did with Zen+ - that will increase the performance dramatically...

 

5 hours ago, TahoeDust said:

The way AMD convinced people that more "slow" cores that communicate slowly with eachother was better for the average user/gamer is a marketing masterpiece.

This is FUD and makes no sense.

What the heck are you talking about?!
 

Could you pls stop the flames and start with facts?

[Jurrunio]

1 hour ago, Plutosaurus said:

I've got big hopes.

 

I, too, want a 5ghz 8/16 for $200.

They did a 5ghz 8/8 before and sold some at $200....

[Mister Woof]

Just now, Jurrunio said:

They did a 8/8 before and sold some at $200....

Lol well

 

You know what I mean

[Jurrunio]

1 hour ago, TahoeDust said:

 

The way AMD convinced people that more "slow" cores that communicate slowly with eachother was better for the average user/gamer is a marketing masterpiece.

 

What is a masterpiece is selling inferior stuff at a price low enough to be attractive, something neither Nvidia nor Intel get the hang of.

 

23 minutes ago, Stefan Payne said:

And IIRC the per CCX latency is on par or lower than Intel,

Interesting, because 2200G still gets beaten by the i3-8100 when they run at similar clocks from the factory

[Mister Woof]

7 minutes ago, Jurrunio said:

What is a masterpiece is selling inferior stuff at a price low enough to be attractive, something neither Nvidia nor Intel get the hang of.

 

Interesting, because 2200G still gets beaten by the i3-8100 when they run at similar clocks from the factory

Is the 2200g one ccx? Or is it two lower binned CCXs combined?

 

IIRC even the 1500x was still 2 CCXs with failed cores on each ccx 

 

I could be totally wrong

 

[Jurrunio]

6 minutes ago, Plutosaurus said:

Is the 2200g one ccx? Or is it two lower binned CCXs combined?

 

IIRC even the 1500x was still 2 CCXs with failed cores on each ccx 

 

I could be totally wrong

 

2200G has one CCX with all 4 cores enabled and a GPU. That's why I didnt use 1200 and 1300X as my example

[Mister Woof]

1 minute ago, Jurrunio said:

2200G has one CCX with all 4 cores enabled and a GPU. That's why I didnt use 1200 and 1300X as my example

Well there goes that theory. Guess it's not just the inter-CCX latency but inter core in same ccx latency that is slower

[KarathKasun]

11 hours ago, Plutosaurus said:

Well there goes that theory. Guess it's not just the inter-CCX latency but inter core in same ccx latency that is slower

In-CCX latency is not out of the realm on normal for modern CPU design.  There are latency maps that illustrate this if you know how to google.

 

12 hours ago, Stefan Payne said:

No, as AMD does something in between, based on the Console designs they have and are proven.

The IF design/implementation being an evolution of console designs is questionable.  The on chip network looks more like a traditional crossbar interconnect using HyperTransport protocols.  This design is significantly different than Garlic/Onion interconnects/protocols used in previous APU designs.

[Stefan Payne]

9 hours ago, KarathKasun said:

The IF design/implementation being an evolution of console designs is questionable.

No its not!

Both consoles use two 4 Core Modules "glued together", why is it so hard to believe that its an evolution of that??

 

9 hours ago, KarathKasun said:

  The on chip network looks more like a traditional crossbar interconnect using HyperTransport protocols.  This design is significantly different than Garlic/Onion interconnects/protocols used in previous APU designs.

Zen also uses Crossbars inside the CCIX and IF to connect two CCIX together - similar to the Consoles, though its not known how the 2 4 Core Parts are connected.

[KarathKasun]

3 hours ago, Stefan Payne said:

No its not!

Both consoles use two 4 Core Modules "glued together", why is it so hard to believe that its an evolution of that??

 

Zen also uses Crossbars inside the CCIX and IF to connect two CCIX together - similar to the Consoles, though its not known how the 2 4 Core Parts are connected.

AMD bought much of their on die and off die bus tech from a company that specifically was in the bus development game for cluster computing.  This happened after console development was already significantly underway.

 

It is, there are diagrams of the chips if you know where to look.