Question Zen 6 Speculation Thread

[HurleyBird]

Nehalem went tri-channel in 2008. No followups since then (IIRC), right?

I wouldn't bet on it, but I also wouldn't be surprised to see a future socket go triple channel again.

1) Core counts are going up

2) If Zen6 leaks are any indication, even desktop is moving towards non-crap tier integrated graphics.

3) The emerging AI prosumer market means +50% capacity matters almost as much as much as +50% bandwidth now.

If it does happen, and here again I wouldn't bet on it, but I could see it being limited to high end boards and processors. Basically socket AM6 or 7 or whatever Intel equivalent pulling double duty as mainstream and HEDT. Besides the socket being a little bigger and/or the pins being a little smaller, mainstream would look about the same and the added cost would be a rounding error. It's just that the form factor be able to scale up and facilitate triple channel at the same time.

 

[Kaffeekenan]

It seems that people don't learn from history.

If you take the best case for every possible outcome and then add them cumulatively for Zen 6 to form expectations for Zen 6 performance vs Zen 5, I give you guys a 100% chance of being bitterly disappointed with the Zen 6 release.

First, I am not convinced AMD will use N2X for anything other than server chips where the margins will be big and they will need the density improvement in order to get improvements over the N3E Zen 5 Turin D currently shipping. I think it much more likely that desktop Zen 6 will use N3P. I might even bet a coke on it .

Second, since transistor density is not going up greatly between shrinks any longer, the transistor budget between generations isn't either. this means that IPC increases will be smaller. I would guess that desktop Zen 6 is only 15-20% higher IPC than Zen 5 ..... and most of that will likely come in the form of a higher bandwidth, lower latency IOD.

In MT, I expect that Zen 6 will easily improve over Zen 5 by 50% simply due to having 50% more cores. The issue that AMD is going to run into with Zen 6 MT is that Intel will have even MORE cores (54 if the rumors are correct). Those "mont" cores are not nearly the computational power houses that Zen c cores rather on full Zen cores are, but they do crunch through low computational throughput stuff (like Cinebench) very effectively. I suspect on more computationally challenging MT tasks, Zen 6 will rule.... but lots of people seem to take those Cinebench scores pretty seriously.

N2X in Server does not make sense, because your high end EPYCs have way too many cores to clock them high. And N2X us used if you want to clock your cores really high. So I would say they use N2X for the high end desktop chips, to smash the 6 GHz barrier.

My prediction for the best desktopchip is:

6,1 GHz
13% combined IPC (more in games, less in applications)

 

[eek2121]

I wouldn't bet on it, but I also wouldn't be surprised to see a future socket go triple channel again.

1) Core counts are going up

2) If Zen6 leaks are any indication, even desktop is moving towards non-crap tier integrated graphics.

3) The emerging AI prosumer market means +50% capacity matters almost as much as much as +50% bandwidth now.

If it does happen, and here again I wouldn't bet on it, but I could see it being limited to high end boards and processors. Basically socket AM6 or 7 or whatever Intel equivalent pulling double duty as mainstream and HEDT. Besides the socket being a little bigger and/or the pins being a little smaller, mainstream would look about the same and the added cost would be a rounding error. It's just that the form factor be able to scale up and facilitate triple channel at the same time.

I am actually surprised we don’t have support for higher channels now. It isn’t like you have to populate all the channels. My Threadripper has no issue becoming tri/dual/single channel as I pull dimms. Motherboards have added enough layers that the “upgrade” should cost very little.

@adroc_thurston is somewhat right that it doesn’t add much performance, but it did add enough to make it worth running 4 dimms. Something like 10-30% based on the workload.

I need to get that system back up and running so I can do some proper benchmarks.

 

[Glo.]

I am actually surprised we don’t have support for higher channels now. It isn’t like you have to populate all the channels. My Threadripper has no issue becoming tri/dual/single channel as I pull dimms. Motherboards have added enough layers that the “upgrade” should cost very little.

@adroc_thurston is somewhat right that it doesn’t add much performance, but it did add enough to make it worth running 4 dimms. Something like 10-30% based on the workload.

I need to get that system back up and running so I can do some proper benchmarks.

Considering how Zen 5 without the x3D cache is memory starved I would say that we DO NEED higher memory bandwdith already on mainstream platforms.

 

[RTX]

N2X in Server does not make sense, because your high end EPYCs have way too many cores to clock them high. And N2X us used if you want to clock your cores really high. So I would say they use N2X for the high end desktop chips, to smash the 6 GHz barrier.

My prediction for the best desktopchip is:

6,1 GHz
13% combined IPC (more in games, less in applications)

If it's the same as N3P/N3X, N2X should consume slightly less power than N2P.

Page 88 - Question - Zen 6 Speculation Thread

Page 88 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

 

[StefanR5R]

"Mainstream" client platforms are not meant to be used for Big Data analytics. Dual channel memory attachment for CPUs¹ is alright² on these platforms. Memory latency remains the big, looming problem.

¹) not for GPUs, nor for big iGPUs
²) as are eight cores (less actually, almost all the time)

 

[Hitman928]

If it's the same as N3P/N3X, N2X should consume slightly less power than N2P.

Page 88 - Question - Zen 6 Speculation Thread

Page 88 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

Less dynamic power to reach higher frequencies but more leakage power consumption which hurts idle and lower frequency operation.

 

[igor_kavinski]

Separate memory channels require more traces on the motherboard. That in turn may require more PCB layers which increases costs further. Mobo OEMs have experienced similar challenges with implementing PCIe5 on consumer motherboards.

Would it require more than the 10 PCB layers used on $500 to $600 consumer mobos? I think I've seen server mobos using 8 PCB layers with hexachannel RAM support.

 

[igor_kavinski]

more money on boards and DRAM for limited to nonexistent gains on msdt sockets.
worthless.

Seriously dude. Not helping. You or anyone else has no right to dictate what we can or cannot have. Join the cause and make the change happen. Let's worry about the outcome later.

 

[soresu]

2) If Zen6 leaks are any indication, even desktop is moving towards non-crap tier integrated graphics.

Which it must be noted doesn't make any sense.

IMHO the only real use for DT iGPU is purely for system testing to isolate dGPU related issues from the equation.

In this case only a bare minimum 1 WGP + the GPU basics (video output, HW video ASICs, HDMI multichannel audio etc) are necessary.

99+% of DT AMx system buyers use a dGPU for everything a GPU is needed for.

Sure the esports contingent would make use of such a beefier iGPU for "lossless scaling", or whatever their "must haz uber FPS" addled brains think might boost their gaming fu just a little bit more - but they are hardly a majority revenue driver of desktop sales.

That extra silicon would be better spent on more cache, IO - or just plain saving area, as it's not like silicon area is even remotely 'free' on any process node below 28nm.

 

[LightningZ71]

Unless that same IOD is also used for their H/HX class laptops. Look at the Arrow Lake H series laptops. Those come with a notably beefy iGPU that's currently besting the 880/890m in many tests. There's something to be said for saving the development costs from another IOD with limited scope.

 

[adroc_thurston]

Seriously dude. Not helping. You or anyone else has no right to dictate what we can or cannot have. Join the cause and make the change happen. Let's worry about the outcome later.

Sorry we ain't doing communist revolutions in semis.

 

[StefanR5R]

IMHO the only real use for DT iGPU is purely for system testing to isolate dGPU related issues from the equation.

There is another real use: As display output in desktop computers which are not used as toys. However, for those a small iGPU is often sufficient too.

 

[GTracing]

Which it must be noted doesn't make any sense.

IMHO the only real use for DT iGPU is purely for system testing to isolate dGPU related issues from the equation.

In this case only a bare minimum 1 WGP + the GPU basics (video output, HW video ASICs, HDMI multichannel audio etc) are necessary.

99+% of DT AMx system buyers use a dGPU for everything a GPU is needed for.

Sure the esports contingent would make use of such a beefier iGPU for "lossless scaling", or whatever their "must haz uber FPS" addled brains think might boost their gaming fu just a little bit more - but they are hardly a majority revenue driver of desktop sales.

That extra silicon would be better spent on more cache, IO - or just plain saving area, as it's not like silicon area is even remotely 'free' on any process node below 28nm.

I for one would be happy to see them bump it up to 4 or 8 compute units. The 2 compute units they currently have are a tiny portion of the chip. Quadrupling the performance of the iGPU would cost only a few dollars.

AMD's Zen 4 I/O Die Shot Reveals A Fascinating Ryzen CCD Detail

Analysis of a die shot for the Zen 4 cIOD confirms a lot of what we new and reveals some interesting surprises.

hothardware.com