New Zen microarchitecture details

[lopri]

On a second thought, What might happen is AMD will leave the multipliers "unlocked" up to a point where a given SKU is allowed to turbo. So a CPU with base/turbo of 3.0/3.4 GHz might have multiplier "unlocked" up to x34 in the BIOS/EFI.

 

[AtenRa]

FX 8300, FX 8320, FX 8350 and FX 8370 all are unlocked. Same can happen with RYZEN.

 

[Lepton87]

On a second thought, What might happen is AMD will leave the multipliers "unlocked" up to a point where a given SKU is allowed to turbo. So a CPU with base/turbo of 3.0/3.4 GHz might have multiplier "unlocked" up to x34 in the BIOS/EFI.

This is actually how turbo works at default on my board. It sets the multiplier at 36x under any load which is the all core turbo multiplier of that CPU. It's called multi-core enhancement and all high-end boards have this and some set this as the default setting like mine does.

 

[The Stilt]

So . . . are you allowed to tell us what is the VRM configuration on the C6H?

I cannot disclose any technical details.
However you can tell from the pictures in those slides that it is "sufficient"

 

[Atari2600]

I cannot disclose any technical details.
However you can tell from the pictures in those slides that it is "sufficient"

Have you had the good fortune to be involved in the certification? If so, did the result meet your expectations?

 

[krumme]

It's still too early to say but my guess would be that Snowy Owl would be at least $2500 to start, including the board.

Ok. And how much do you think it can sell a year?

 

[lopri]

FX 8300, FX 8320, FX 8350 and FX 8370 all are unlocked. Same can happen with RYZEN.

Because that was the only way for those chips to compete. For AMD's sake I hope Zen is more competitive than FX.

 

[The Stilt]

On a second thought, What might happen is AMD will leave the multipliers "unlocked" up to a point where a given SKU is allowed to turbo. So a CPU with base/turbo of 3.0/3.4 GHz might have multiplier "unlocked" up to x34 in the BIOS/EFI.

FX 8300, FX 8320, FX 8350 and FX 8370 all are unlocked. Same can happen with RYZEN.

AMD has already stated that all Ryzen SKUs are fully unlocked.
While the CPUs themselves are unlocked, you will need a motherboard with a chipset variant (B350, X300 or X370) which has the overclocking feature unlocked.
No different to Intel's chipset segmentation, where only Z-series consumer chipsets support overclocking regardless of the installed CPU SKU (K).

On Ryzen base and boost states are completely isolated and the latter are completely invisible outside the SMU. So don't expect that you can trick the CPU to maintain the maximum boost speeds on boards which don't support overclocking.

 

[guskline]

Nice avatar the Stilt!

 

[AtenRa]

Because that was the only way for those chips to compete. For AMD's sake I hope Zen is more competitive than FX.

Intel sells two unlocked 6C 12T CPUs, one has more PCIe lanes but majority of buyers get the cheaper one.
Perhaps AMD will do something similar.

 

[zinfamous]

Marketing bullsheit.

yeah well if you don't have VR optimized USBs, then your USBs will run out of RAMs for VR. You're going to have to create a GUI to hack in more RAMs.

 

[LTC8K6]

AMD has already stated that all Ryzen SKUs are fully unlocked.
While the CPUs themselves are unlocked, you will need a motherboard with a chipset variant (B350, X300 or X370) which has the overclocking feature unlocked.
No different to Intel's chipset segmentation, where only Z-series consumer chipsets support overclocking regardless of the installed CPU SKU (K).

On Ryzen base and boost states are completely isolated and the latter are completely invisible outside the SMU. So don't expect that you can trick the CPU to maintain the maximum boost speeds on boards which don't support overclocking.

Okay, so if we have two 8c/16t variants, and I have a top end board, what will be the reason I might buy one chip over the other, if both are fully unlocked?
It has to just be binning that would justify different pricing. One happens to auto overclock higher than the other with the same cooling.
But, AMD has no idea what cooling I might use, and therefore should not know what auto clock speeds I will get for my money. So how will they price them?

With Intel if I want to overclock a desktop system, I basically have three chip choices today. 7350K/7600K/7700K which is really only a single top chip, with 8 threads.
If I want to overclock a HEDT, I have 4 choices. 6800K/6850K/6900K/6950X with only one top chip, with 20 threads. Or if we eliminate the extreme edition, then one chip with 16 threads.
Intel does not offer two versions of the top chips. There's only one top chip with the most threads.
If Intel did offer such, they would have different base/turbo clocks and that would be the main reason to pick one. So that's Intel's version of binning, I guess. But that's how Intel would set the pricing.

 

[CatMerc]

I mentioned the possibility of AMD having built up stock of an earlier stepping that doesn't overclock as high as the newer ones. Fottemberg claimed that newer steppings are clocking higher than expected. The newer ones being limited would fetch a higher price tag, while the rest will be made into the lower 8C/16T SKU.

Over time the newer stepping will replace every chip as there's no point in manufacturing the old stepping, but for now there would be a clear benefit in overclocking potential to get the higher end SKU.

 

[realibrad]

Okay, so if we have two 8c/16t variants, and I have a top end board, what will be the reason I might buy one chip over the other, if both are fully unlocked?
It has to just be binning that would justify different pricing. One happens to auto overclock higher than the other with the same cooling.
But, AMD has no idea what cooling I might use, and therefore should not know what auto clock speeds I will get for my money. So how will they price them?

With Intel if I want to overclock a desktop system, I basically have three chip choices today. 7350K/7600K/7700K which is really only a single top chip, with 8 threads.
If I want to overclock a HEDT, I have 4 choices. 6800K/6850K/6900K/6950X with only one top chip, with 20 threads. Or if we eliminate the extreme edition, then one chip with 16 threads.
Intel does not offer two versions of the top chips. There's only one top chip with the most threads.
If Intel did offer such, they would have different base/turbo clocks and that would be the main reason to pick one. So that's Intel's version of binning, I guess. But that's how Intel would set the pricing.

Binning and base clocks I would imagine. So if your cpu cooler can dissipate X amount of watts then the higher binned chip will offer more perf vs the lower.

 

[pj-]

Marketing bullsheit.

I dunno, have you been to the Oculus subreddit lately?

If they have to put up stuff like this, then maybe there is something mobo makers can do to help. "plug in VR here" is a lot simpler than "balance your sensor and HMD connections over multiple host controllers and make sure you only have 2 sensors working in usb 3.0 mode, and maybe buy a PCIe USB card, that could help"

If they can actually do something to ensure good VR tracking performance, which could be as simple as a dedicated host controller for those ports, ( or giving 'priority' to those ports, if that's possible) people would pay for it.

 

[dacostafilipe]

Canard PC Hardware

AMD currently validating Ryzen 6C/12T@3.3GHz base but no 8C/8T. The goal is clearly to recycle the failed 8C dies.

Source:https://twitter.com/cpchardware/status/826829044402552833

 

[Atari2600]

As long as validation only means a representative CPU that is checking for errors and not the final performance ceiling.

3.3G on a 6C12T wouldn't really cut it at the top end.

 

[lixlax]

As long as validation only means a representative CPU that is checking for errors and not the final performance ceiling.

3.3G on a 6C12T wouldn't really cut it at the top end.

If that's achieved at 65W(as rumoured) TDP and has unlocked multiplier (as promised) then I don't see the problem for enthusiasts. Plus system integrators will love it because of low TDP.

But if they are validating it right now does it mean that it's probably a H2 2017 product or how long does it usually take?

 

[LTC8K6]

So we've heard that there are no 6c/12t chips and that they are validating 6c/12t chips.

 

[The Stilt]

Okay, so if we have two 8c/16t variants, and I have a top end board, what will be the reason I might buy one chip over the other, if both are fully unlocked?
It has to just be binning that would justify different pricing. One happens to auto overclock higher than the other with the same cooling.
But, AMD has no idea what cooling I might use, and therefore should not know what auto clock speeds I will get for my money. So how will they price them?

With Intel if I want to overclock a desktop system, I basically have three chip choices today. 7350K/7600K/7700K which is really only a single top chip, with 8 threads.
If I want to overclock a HEDT, I have 4 choices. 6800K/6850K/6900K/6950X with only one top chip, with 20 threads. Or if we eliminate the extreme edition, then one chip with 16 threads.
Intel does not offer two versions of the top chips. There's only one top chip with the most threads.
If Intel did offer such, they would have different base/turbo clocks and that would be the main reason to pick one. So that's Intel's version of binning, I guess. But that's how Intel would set the pricing.

Honestly I have no idea what the final lineup will be. Zeppelin is rather flexible, so there a plenty of potential configurations.

 

[CatMerc]

Honestly I have no idea what the final lineup will be. Zeppelin is rather flexible, so there a plenty of potential configurations.

Nobody appears to know much as far as SKU's go. AMD managed to keep the lid on that quite well.
I'm in general impressed with just how good AMD's control of information has been, looks like they took the lessons from the Eyefinity situation to heart.

 

[realibrad]

Honestly I have no idea what the final lineup will be. Zeppelin is rather flexible, so there a plenty of potential configurations.

Is it possible to have 6C/6T that is cut down from 8C/16T cpus?

 

[The Stilt]

Is it possible to have 6C/6T that is cut down from 8C/16T cpus?

Naturally.
The design dictates that both CCXs are enabled (three cores each), however besides that there are no limitations.

 

[realibrad]

Naturally.
The design dictates that both CCXs are enabled (three cores each), however besides that there are no limitations.

That is what I thought. So it should be quite possible to have a 8C chip that gets cut into 6C chips much like the x3 720 was cut down from 4C correct?

If so, interesting...

 

[PPB]

Which should actually just be binning alone in the case of RyZen, since the processor itself determines it's clocks.

There should not be any AMD locks or settings on the clocks. Each one should be determining it's own clocks and overclocks.

Unless AMD is releasing locked chips.

AMD is releasing locked platforms. The chip determine it's clock with a predefined set of parameters. XFR won't make 2 different bins of 8c/16t clock the same because they have the same cooling.