EPYC Momentum Grows Claims AMD

Glad to hear that AMD is doing well regarding CPU's, it was about time.

 

Good news we need them back in the fight for the long haul.

The only issue I see for adoption of EPYC is that in many data centers the software licensing is much more expensive than the hardware. Having more cores usually means more licence costs albeit some vendors licence by CPU socket not cores. This is where there will be some push back.

 

They focus on China for a reason.
When AMD would, they could get higher clock-rates, they are just ryzen's glued together.

 

What's different about the 7501 and 7551P?

Also, interesting how they intend to make an 8-core. I think they should, I'm just a little surprised they did. What I don't really understand is why it has such low clocks.

 

"Ryzen's glued together" is a silly, sour-grapes observation that makes not one whit of difference to anyone in any data center anywhere. What such companies care most about is price-performance per watt. China is a huge market, btw--this event was *in* China--which is why they talked about Chinese companies. AMD is a global business.

 

AMD could glue together monkey poo and cheerios and I'd be more then happy if it delivered the performance and pricing that Ryzen, Threadripper, Epyc has/will. I would say the same thing about Intel if the shoes were reversed too.

 

Intel just announced at hot chips conference a couple days ago about there wonderful EMIB. This would be Intel's version of glue. Intel bagged on AMD but they are heading that direction for 10nm with EMIB.

 

Cost savings with big chips are proly too good to miss out.
And it's true, the moment someone suddenly starts talking about the competition, you know that competition has something good going on

 

All the "P" versions are for single socket systems only. Non "P" processors can sit in dual socket systems (that's the highest socket count currently supported on EPYC)

Anandtech confirmed that the 8 core SKU still has all 4 dies, with only 2 cores enabled on each die (one per CCX). AMD need to spec it that way to get full access to the 8 memory controllers and 128 PCIe lanes. I suspect the clock speed is not really relevant because the these SKUs exist solely for platforms with minimal CPU requirements but need masses of memory and/or PCI lanes. A lot of the CPU TDP budget will be going to the cache, memory controllers and pcie lanes, all of which remain active (though half the Lvl3 cache is disabled). No doubt AMD could have bumped the TDP and clock speeds, but that's not the point of this CPU.

They can't easily get higher clock rates at these core counts without completely losing control of TDP. Look at the Threadripper OCing results. Most 1950Xs can be perfectly stable at 4Ghz on all 16 cores, as long as you can handle the 250W-300W+ worth of heat that it starts to spew. You'd need some exotic cooling solutions to push a 32 core EPYC into the high 3Ghz range. In any case, most datacenters prioritise performance per watt over raw performance and you can bet these SKUs are sitting nicely in the efficiently sweet-spot.