Do you think AMD should offer more than one bin level for its upcoming Zen Opterons?

[cbn]

Should AMD should offer more than one bin level for its upcoming Zen Opteron CPUs (like they do with the Carrizo APUs)?

With Carrizo AMD offers two bin levels (15W and 35W) per SKU. One thing I believe this does is allow OEMs greater versatility with their Carrizo inventory (ie, they can decide to make either 15W or 35W laptop from a single processor). Or in the case of us (as the end users) we should be configure Carrizo Mini-ITX boards as either 15W passive or 35W when running with a fan on the heatsink.

So back to the upcoming Zen Opterons, do you think a strategy like this would be a good idea?

Maybe (as an example) a 120W bin level and 275W bin level for a single SKU?

120W for Open Compute Servers (and the like) plus Workstations from Dell, HP and Lenovo.

275W for custom servers and custom workstations.

P.S. As outlandish as 275W sounds remember that core counts are increasing, so the power consumption per core and heat density could be quite normal even at this elevated cTDP.

 

[cbn]

32C at 275W works out to 8.6W per core.

 

[frozentundra123456]

Yea, worked out real well for Carizzo didnt it? Does anybody even bother to make 35w products for it?

 

[el etro]

They could use and enhance the power management/turbo features of Carrizo, to make Zen single-thread performance rocks(E.G. boosting to 4+GHZ at light threaded loads).

 

[svenge]

There's a fair chance that the only "bin" that Zen Opterons get is the bargain bin.

 

[DrMrLordX]

Yea, worked out real well for Carizzo didnt it? Does anybody even bother to make 35w products for it?

In retrospect, that was probably AMD's biggest mistake: launching Carrizo with OEM-selectable cTDP with the option of user-selectable cTDP (which the OEMs rarely chose to implement).

AMD should have launched Carrizo with fixed TDP per SKU. Variable cTDP could have been an option in a product refresh, if the OEMs were clamoring for it. If they were going to launch Carrizo as-is, they should have insisted on sufficient cooling and VRM support for the max cTDP of the SKU AND user-selectable cTDP in the UEFI.

It will be harder for OEMs to screw them as hard in the desktop scene, but still . . . lesson learned.

 

[Essence_of_War]

How can one sensibly answer this question without knowing how Zen's perf per core, clock and watt actually looks?

 

[cbn]

Yea, worked out real well for Carizzo didnt it? Does anybody even bother to make 35w products for it?

In retrospect, that was probably AMD's biggest mistake: launching Carrizo with OEM-selectable cTDP with the option of user-selectable cTDP (which the OEMs rarely chose to implement).

AMD should have launched Carrizo with fixed TDP per SKU. Variable cTDP could have been an option in a product refresh, if the OEMs were clamoring for it. If they were going to launch Carrizo as-is, they should have insisted on sufficient cooling and VRM support for the max cTDP of the SKU AND user-selectable cTDP in the UEFI.

It will be harder for OEMs to screw them as hard in the desktop scene, but still . . . lesson learned.

I believe you are referring to lack of laptops with 35W Carrizo as a standalone (ie, without dGPU)?

Because haven't OEMs used Carrizo at both 15W and 35W, but the 35W cTDP comes with dGPU (Example here, thickness shown below).

However, even if 35W wasn't used for laptops, I still think the dual bin is a good idea because later on the same chips could be used for 35W AIO desktops or Mini-ITX boards.

P.S. I have been under the impression that since Kaveri, the 35W APUs get paired with dGPU. The lower wattage Construction core APUs are the only ones used as a standalone. For whatever reason, 35W APUs don't get used in laptops as standalone.

 

[cbn]

How can one sensibly answer this question without knowing how Zen's perf per core, clock and watt actually looks?

You don't have to specify the max TDP.

However, I do think it is interesting that the rumored desktop Zen as 8C/16T Summit Ridge is supposed to have a 95W TDP, but if the same CPU core was configured (with same clocks) as 32C/64T the TDP would be 380W.

So turns out 275W for Zen 32C/64T is actually lower power per core than the rumored stock desktop part.

But even at this reduced power of 275W would it be good for AMD to include at least one lower power bin as well? Maybe even one more higher power bin?

Or maybe you don't like TDPs over ~145W (or ~165W) and if AMD were to use multi-binning they should focus at various levels below this?

Of course, 145W/165W is the standard range for Servers (145W for Open Compute Leopard), but it would be nice to see a higher TDP bin option for those that can cool it. For servers maybe watercooling (a nice bonus would be one that can recycle heat) and for custom workstations an AIO cooler should do the job or maybe even a beefy aircooler.)

 

[DrMrLordX]

I believe you are referring to lack of laptops with 35W Carrizo as a standalone (ie, without dGPU)?

No. I refer to the shortage of laptops with 35W FX-8800P period.

Standalone or not, OEMs don't seem to want to do it. Even the y700 has problems . . . the VRMs are underbuilt for the application.

 

[cbn]

No. I refer to the shortage of laptops with 35W FX-8800P period.

Standalone or not, OEMs don't seem to want to do it.

Found some good info on these pages:

http://www.overclock.net/t/1573937/carrizo-fx-8800p-a10-8700p-info-and-optimizations/10

http://www.overclock.net/t/1573937/carrizo-fx-8800p-a10-8700p-info-and-optimizations/20

http://www.overclock.net/t/1573937/carrizo-fx-8800p-a10-8700p-info-and-optimizations/30

Interesting the Stilt's findings on the Y700 (even though it is accompanied by a powerful 75W Bonaire GPU).

Also very interesting he mentions here that many 15W Carrizo are accompanied by a 15W dGPU.

The trouble is that many of the 15W designed systems have a discrete GPU present, and the dGPU dumps it´s dissipated power in the same cooling element as the SoC does.

Most of the lower-end discrete GPUs paired with Carrizo have TDP of 15W.

(Boggles the mind, when I think about how Carrizo can be configured as 35W standalone...but OEMs are using 15W Carrizo + 15W dGPU instead)

But then when I think about 35W Carrizo standalone, I can possibly understand why 35W isn't used: Since Carrizo can shift the whole TDP budget to either CPU or iGPU each power plane needs to be rated at the full 35W. Is that just too expensive to implement?

 

[DrMrLordX]

Interesting the Stilt's findings on the Y700

The y700, unfortunately, has underbuilt VRMs, so it struggles with full power usage at least on the APU.

But then when I think about 35W Carrizo standalone, I can possibly understand why 35W isn't used: Since Carrizo can shift the whole TDP budget to either CPU or iGPU each power plane needs to be rated at the full 35W. Is that just too expensive to implement?

As The Stilt mentioned, every major power domain on Carrizo can use the full power budget. I'm not sure the power management of the chip will allow every major domain to pull that much power at once, otherwise you're looking at a 70W+ power draw. Of course, that's Prime95 + Furmark territory.

The whole chip is set up like a tablet SoC: it's designed for workload bursts followed by downtime. Apparently it does an okay-ish job of that with proper VRM support and 35-42W cTDP limits. Anything below that and the FX-8800P has problems.

 

[Vesku]

In retrospect, that was probably AMD's biggest mistake: launching Carrizo with OEM-selectable cTDP with the option of user-selectable cTDP (which the OEMs rarely chose to implement).

AMD should have launched Carrizo with fixed TDP per SKU. Variable cTDP could have been an option in a product refresh, if the OEMs were clamoring for it. If they were going to launch Carrizo as-is, they should have insisted on sufficient cooling and VRM support for the max cTDP of the SKU AND user-selectable cTDP in the UEFI.

It will be harder for OEMs to screw them as hard in the desktop scene, but still . . . lesson learned.

Even when AMD had the better CPUs they couldn't do much dictating to OEMs. There is just no way they would be able to dictate whether a particular SKU received adequate cooling for the top tdp config. If the tdp configuration isn't a cost savings move to reduce the number of total SKUs AMD has to deal with they are better off ditching it for optimally balanced SKUs.

 

[JDG1980]

AMD really should have made a reference Carrizo laptop. The OEM offerings are all crap. Why would you stick a low-tier dGPU in a laptop with a chip that has one of the best iGPUs on the market? It makes no sense.

 

[DrMrLordX]

Even when AMD had the better CPUs they couldn't do much dictating to OEMs. There is just no way they would be able to dictate whether a particular SKU received adequate cooling for the top tdp config. If the tdp configuration isn't a cost savings move to reduce the number of total SKUs AMD has to deal with they are better off ditching it for optimally balanced SKUs.

Overall I agree with you, which is why they should have fixed the cTDP per sku. If OEMs want to use the 8800p, they'd best treat it as a 35w chip. Of course, as the y700 has shown us, even that wouldn't be a panacea for all the problems Carrizo has faced from OEMs, but it would have been an improvement.

 

[DrMrLordX]

AMD really should have made a reference Carrizo laptop. The OEM offerings are all crap. Why would you stick a low-tier dGPU in a laptop with a chip that has one of the best iGPUs on the market? It makes no sense.

I think they did. The OEMs seem intent on not building off of that.