apu "Raven Ridge" APU more details leaked

[Jito463]

@Phate.exe @Dackzy

Perhaps we can all agree that anytime you abuse something, it's likely to fail?  Take care of your equipment, and it will last a whole lot longer.  Beat it up, toss it around and all around don't care for it, and any component is bound to eventually break, no matter how well built they are.

 

Frankly, I find the HDD is the first thing most likely to fail in a heavily used laptop, that's why I far prefer an SSD for them.  The case is hardly my biggest concern (though I concede that it is a concern).

[Drak3]

14 minutes ago, Jito463 said:

Perhaps we can all agree that anytime you abuse something, it's likely to fail?

I can't.

 

Agreeing with people is just so, agreeable.

 

 

*shudders*

[sazrocks]

6 hours ago, MageTank said:

All of that is done over the QPI or "QuickPath Interconnect". It allows your CPU to communicate with the uncore subsystem, which includes your IMC, CPU PCIe lanes, and iGPU. 

 

As far as bandwidth goes, QPI is worlds slower than even AMD's older HyperTransport 3.1 (from 2008). QPI's max bandwidth is 38.4GB/s, while HT3.1 had a bandwidth of 51.2GB/s (roughly a 33% difference). To put this into perspective, Ryzen's InfinityFabric at 2666 has a peak theoretical bandwidth of 341.2GB/s using 2666mhz ram at a 512bit width. If they improve upon the IMC to allow for DDR4's 4266 spec to be achieved, then Infinity Fabric's max theoretical bandwidth is 512GB/s. 

 

NVLink 2.0 is "only" 150GB/s, and NVLink 1.0 was 80GB/s. If AMD is good for anything, it's making extremely fast interconnects. They've always been ahead in this regard. 

 

EDIT: I should also clarify that QPI is mostly used on Xeons. A lot of your desktop platforms handle iGPU transactions on the PCH (Platform Controller Hub). It speaks directly to your FDI (Flexible Display Interface). It's current interconnect bandwidth is 2.7G/bits per second, or 0.33GB/s. Sorry for not stating this earlier, mind was in another place. 

That's insanely fast. I wonder how much it will help the GPU performance on these APUs.

[MageTank]

Just now, sazrocks said:

That's insanely fast. I wonder how much it will help the GPU performance on these APUs.

Would depend on whether or not the GPU is bottlenecked by the interconnect. I highly doubt it would be, as that was not the case on previous APU's. I am more interested in seeing the CCX design on these CPU's. If they are only a single CCX, memory latency might be better on them. Can anyone confirm the CCX configuration on these APU's?

[Dackzy]

7 hours ago, Phate.exe said:

Snip

We are having totally different experiences with these machines. I see MBPs and ZenBooks being killed fast, while a T430s and such are still running with no problem at all. 

Also repairing one of these machines is easy AF,  sure there might be a screw or two extra,  but you can basically always get parts for one of them and it is easier than a lot of your new ultrabooks, but yes over the gens they have made it a bit harder,  which is really a shame.

I think my T420s is one of the last models where it is super easy to swap hard-drive, ram, CD drive and SD card reader. Though I must say Dell and their new precision models are very interesting to me,  way more than this gem of thinkpads (yes I really don't care about brand, as long as the build quality is there)

 

Back to topic. I it would be very interesting to see one of these APUs in a machine that has good cooling and then Ofc with fast ram,  I don't think it would be a machine that would make that much sense,  but I think that we can count on HP to try and make one,  at least if you trust their track record then we should see laptops from then with these APUs.