AMD might announce their new CPU lineup on Tuesday

[GoldenLag]

Just now, cj09beira said:

ya nvidea would probably prefer a amd vs nvidea market than a nvidea vs intel as with them now working on gpus they might try to remove pcie to destroy nvidea overnight 

providing that Intel and AMD soon both have GPUs means they can virtually kill off Nvidia without too many law suits due to there still being multiple GPU providers

[Mira Yurizaki]

2 minutes ago, cj09beira said:

 try to remove pcie

Good luck with that.

[Deus Voltage]

2 minutes ago, cj09beira said:

ya nvidea would probably prefer a amd vs nvidea market than a nvidea vs intel as with them now working on gpus they might try to remove pcie to destroy nvidea overnight 

Can you imagine that, Intel higher ups sitting in a dark room saying to eachother:

 

First, we'll steal AMD's GPU engineers

Next, we'll work on a new non-PCIE GPU that has far less latency than anything on the market

Finally, after Nvidia is gone, we'll buy AMD

 

I'm kidding of course, but it's fun to think about this stuff

[cj09beira]

Just now, GoldenLag said:

providing that Intel and AMD soon both have GPUs means they can virtually kill off Nvidia without too many law suits due to there still being multiple GPU providers

true but unlikely, then intel would focus 100% on amd which would not be good for them, keeping intel busy is important 

 

 

[GoldenLag]

Just now, M.Yurizaki said:

Good luck with that.

i mean. 2 out of 3 x86 CPU providers just swapping standards could really just be the end of PCIe

[Zagna]

EPYCs I/O die seems like 128 PCIe 4.0 lanes, 8 channel memory and 8 Infinity Fabric links?

Threadripper I/O could be 64 PCIe 3.0 lanes, 4 channel memory and 4 Infinity Fabric links?

Ryzen I/O, if Ryzen somehow does also go with chiplets, 24 PCIe 3.0 lanes, 2 channel memory and 2 Infinity Fabric links?

APU I/O, crazy thought, 16 PCIe 3.0 lanes, 2 channel memory, 1 Infinity Fabric link and 12 CU Vega?

I can dream right?

[Mira Yurizaki]

Just now, GoldenLag said:

i mean. 2 out of 3 x86 CPU providers just swapping standards could really just be the end of PCIe

And replace it with what? A proprietary interface?

 

I'm sure all the peripheral manufacturers would love to jump on board to a new interface.

[Taf the Ghost]

Just as a practical note, this is a really, really serious presentation. As in, they spent a lot of time & effort on this. (These things aren't that expensive, relatively.) They clearly know it's really important, though why no official live stream?

[cj09beira]

Just now, Zagna said:

EPYCs I/O die seems like 128 PCIe 4.0 lanes, 8 channel memory and 8 Infinity Fabric links?

Threadripper I/O could be 64 PCIe 3.0 lanes, 4 channel memory and 4 Infinity Fabric links?

Ryzen I/O, if Ryzen somehow does also go with chiplets, 24 PCIe 3.0 lanes, 2 channel memory and 2 Infinity Fabric links?

APU I/O, crazy thought, 16 PCIe 3.0 lanes, 2 channel memory, 1 Infinity Fabric link and 12-16 CU Vega?

I can dream right?

those infinity fabric links are on the gpu only i think, and they talked about up to 4 pcie 4.0 gpus, which seems to point to less pcie lanes 

[Deus Voltage]

Just now, GoldenLag said:

i mean. 2 out of 3 x86 CPU providers just swapping standards could really just be the end of PCIe

Exactly my thought process. Motherboards are made primarily ( I think) with CPU's in mind. Intel can literally obliterate Nvidia in an alternate universe where they say fuck it and demand PCIe to be removed from the manufacturing process

[NumLock21]

Just now, Taf the Ghost said:

Just as a practical note, this is a really, really serious presentation. As in, they spent a lot of time & effort on this. (These things aren't that expensive, relatively.) They clearly know it's really important, though why no official live stream?

AMD decided to let Alex on Tech do the stream instead of them doing it themselves. AMD gave him full permission to do the live stream.

[GoldenLag]

Just now, M.Yurizaki said:

And replace it with what? A proprietary interface?

 

I'm sure all the peripheral manufacturers would love to jump on board to a new interface.

well, its not like they have a lot of choice. 

 

i mean, it would be horrible for most parties, but it could be done. 

[Zagna]

1 minute ago, cj09beira said:

those infinity fabric links are on the gpu only i think, and they talked about up to 4 pcie 4.0 gpus, which seems to point to less pcie lanes 

There are still IF links between the chiplets and the I/O die.

The GPU IF links are something even faster?

[Taf the Ghost]

2 minutes ago, NumLock21 said:

AMD decided to let Alex on Tech do the stream instead of them doing it themselves. AMD gave him full permission to do the live stream.

I guess the people they really wanted to inform are all there, but, still, it seems like a missed opportunity.

[cj09beira]

1 minute ago, Zagna said:

There are still IF links between the chiplets and the I/O die.

The GPU IF links are something even faster?

oh thats what you mean, nevermind 

[Mira Yurizaki]

2 minutes ago, GoldenLag said:

well, its not like they have a lot of choice. 

 

i mean, it would be horrible for most parties, but it could be done. 

Are we talking about a situation where AMD and Intel, if Intel ever releases a video card, that they'll both create a proprietary interface to shutout NVIDIA?

 

Great, now we're just locked in to either AMD xor Intel for systems. That's exactly what I wanted.

[GoldenLag]

1 minute ago, M.Yurizaki said:

Are we talking about a situation where AMD and Intel, if Intel ever releases a video card, that they'll both create a proprietary interface to shutout NVIDIA?

 

Great, now we're just locked in to either AMD xor Intel for systems. That's exactly what I wanted.

pretty much. it would ruin so many things on so many levels. 

[Taf the Ghost]

Cray's CEO. That's pretty important.

[NunoLava1998]

Just now, GoldenLag said:

pretty much. it would ruin so many things on so many levels. 

Couldn't NVIDIA just adjust GPUs to support that proprietary interface? Or would they not license it

[cj09beira]

ahahha, amd sponsors Ferrari then helps their competitors be faster

[GoldenLag]

Just now, NunoLava1998 said:

Couldn't NVIDIA just adjust GPUs to support that proprietary interface? Or would they not license it

i mean AMD and Intel can technically make it impossible for them to do so. or just "accidentaly" gimp it due to "incompatible" interfaces.

 

its highly unlikely that it would ever happen

[GoldenLag]

*shaky cam for action scenes

[Mira Yurizaki]

Just now, NunoLava1998 said:

Couldn't NVIDIA just adjust GPUs to support that proprietary interface? Or would they not license it

They could, but that doesn't mean the maker of the interface can give every secret sauce and make it hard for NVIDIA to attain any real competitive edge. On top of fees that could be unfavorable.

[cj09beira]

we are just getting started v2.0 get hyped

[Zagna]

Infinity Fabric