AMD - Piledriver chitchat

[watercooled]

It could be tricky by the sounds of it, in the same way it's hard to know if lower clock 'notches' will be stable when using offset undervolting. I wonder if the throttling is detectable by software e.g. a drop in reported clock speed?

[HalloweenJack]

whats the wattage tdp of that slot powered 7850?

could there *maybe* be an ondie version soon?

[CAT-THE-FIFTH]

The PCI-E slot can supply upto 75W.

[HalloweenJack]

6670 uses 58w peak according to tpu.... hmmmm could be close to an on die 7850 maybe?

would totally blow away anything intel can make this or next year

[watercooled]

Power aside, I suspect die size could start to become a problem at that point, unless we're talking the next node? Also, bandwidth is an obvious concern, but DDR4 can't be that far off, and we have those GDDR5 rumours. I wonder how that would work though, you can't just buy GDDR5 modules! On-board memory isn't a great solution for desktop IMO, but maybe as a stopgap if necessary?

Edit: Just realised I've skipped over a page or two of the thread, and that was already brought up.

[DanceswithUnix]

GDDR5 would work lovely for sideport ram.

Edit to add: Just leave me some DDR3/4 sockets to fill/upgrade as well.

[CAT-THE-FIFTH]

Where are all the Richland reviews??

Come on AMD,the chips are already available.

I really hope they send out the A10-6700. It was utter fail they did not advertise the A10-5700 more.

[kompukare]

GDDR5 would work lovely for sideport ram.

Edit to add: Just leave me some DDR3/4 sockets to fill/upgrade as well.

Sideport RAM could be a return to something like the Amiga chipram / fastram divide. Which would be few steps backwards and the total opposite of HSA. Mind you, back when there was 'unequal' RAM (and I'm including the early Mac's 24-bit addressing and Atari's TT and Falcon here), operating systems were a lot more primitive and MNU were either not used or poorly supported so a modern OS could make this transparent. Or a modern OS could handle this with a bit (lot?) or kernel patching.

[DanceswithUnix]

Modern OSes already understand NUMA, and I have a couple of machines around here with AM2+ motherboards with sideport ram on them which work fine with Win7 and Linux.

It helped with just 40 old shaders on the chipset to have a 32 bit ram chip attached (I think on my board it is 128MB of DDR3), I imagine the idea would benefit modern machines a lot more.

[HalloweenJack]

apparently theres an NDA on richland?? so anyone buying it can actually review It before the websites!

[CAT-THE-FIFTH]

A10-6800K review:

http://translate.google.ro/translate...ew%2Fpage%2F23

[inf64]

I would give you "thanks" if I could Cat . Great find!

[HalloweenJack]

the OC 3dmark score is better than my kids pc

http://www.3dmark.com/fs/240666


but


if the scores from anand are right its LOWER than iris pro

http://www.anandtech.com/show/6993/i...50hq-tested/16

[inf64]

the OC 3dmark score is better than my kids pc

if the scores from anand are right its LOWER than iris pro

Of course it's lower. Iris pro is a big iGPU with large L4. A brute force attack in engineering and pricing terms . It's not directly comparable to 6800K as it will be much cheaper than any GT3e enabled desktop part (if intel decides to launch such a thing).

[HalloweenJack]

http://www.anandtech.com/show/6993/i...950hq-tested/4


what is that on package??

[watercooled]

I wonder how the kerfuffle about Haswell needing uber-low power idle support on the PSU is playing out? Platform idle power looks lower than 1155 but it's not something I can see causing problems for any remotely recent/decent PSU.

Oh and about the 'on-board VRM', I'm still seeing caps, inductors and heatsink'd FETs on 1150 motherboards. Maybe it's just the PWM controller that's been moved on-board, as I predicted?

Edit: The smaller die on the CPU package is the L4/Crystalwell memory.