AMD - Piledriver chitchat

[kompukare]

Well throwing die space at a bandwidth contained iGPU does seem wasteful. Maybe if OpenCL actually had any useful programs but even then. 384 shaders and an aggressive turbo for the CPU would have made a lot more sense.

Or 384 shaders would save ~15% of the overall die (say, the iGPU is 60% and 384/512 = 45% + 40% = 85%), so using that space for extra iGPU cache or something probably would possibly have been a better use of space. Of course, GCN uses a lot more transistors than VLWI5/6.

So 8 years after buying ATI, AMD still hasn't actually released any really compelling Fusion chip. Or at least nothing worth $6 billion. I don't think the Fusion vision was necessarily wrong, it's just that the purchase left AMD far too cash strapped to execute it.

[CAT-THE-FIFTH]

Well throwing die space at a bandwidth contained iGPU does seem wasteful. Maybe if OpenCL actually had any useful programs but even then. 384 shaders and an aggressive turbo for the CPU would have made a lot more sense.

Or 384 shaders would save ~15% of the overall die (say, the iGPU is 60% and 384/512 = 45% + 40% = 85%), so using that space for extra iGPU cache or something probably would possibly have been a better use of space. Of course, GCN uses a lot more transistors than VLWI5/6.

So 8 years after buying ATI, AMD still hasn't actually released any really compelling Fusion chip. Or at least nothing worth $6 billion. I don't think the Fusion vision was necessarily wrong, it's just that the purchase left AMD far too cash strapped to execute it.

The problem also is their marketing.

Lower TDP parts like the A8 7600 seems to be a decent boost in both CPU and IGP performance over the A10 6700T. I suspect mobile parts will show this improvement.

However,by showing the higher TDP desktop parts,it will look a sidegrade at best(whilst paying more money),and this is the first impression which will stick with people.

They should have just reviewed the mobiles parts first.

Now all we will have is reviews going on about the decrease in single thread performance,which is probably not going to be seen in mobile and lower TDP parts.

It is like anti-marketing.

Edit!!



They even state that higher TDP parts have somewhat gimped clockspeeds.

So what does AMD do??

Send out the higher TDP parts first!!

LOLWTF??

[HalloweenJack]

what speed DDR3 did they use btw?

oh and does the 5800K support 1866 or can it go to 2133?


and

lol

how does it compare against an i5 4570R? - cost differences aside

[The Hand]

You will be better served getting an FX6300 or FX8320 TBH. Someone at AMD must be smoking skunk to think they can charge £130 to £140 for an A10 7850K. Supposedly there is noise BF4 will be included with the A10 7850K and even then it makes it only slightly better value than an A10 6800K.

I was just looking at the FX-8320 actually lol. I could get an FX-6300 and a £100 graphics card for roughly £50 more than the 7850K if it does come out at the £130 mark.

[CAT-THE-FIFTH]

what speed DDR3 did they use btw?

oh and does the 5800K support 1866 or can it go to 2133?


and

lol

how does it compare against an i5 4570R? - cost differences aside

The Core i5 4570R is probably faster. OTH,it is quite gimped as it only has 4MB of L3 cache and runs at 2.7GHZ to 3.2GHZ,meaning its slow as Core i5 parts go. It would not surprise me if a Haswell Core i3 is faster in many cases.

[watercooled]

TBH we should probably wait for some more reviews before jumping to any conclusions. IIRC AMD's main claims about IPC uplift were more related to integer rather than FP performace; Cinebench is almost entirely FP, is somewhat 'synthetic' in nature, and as we know is Intel-biased/lacks AMD-specific optimisations. All we have for integer is part of a Geekbench test.

Power/performance is another point, and I suspect the lower TDP parts will show more of an improvement here, with the flagship parts possibly pushing up against the clock/efficiency wall, hence lower clock vs Richland's process.

[CAT-THE-FIFTH]

TBH we should probably wait for some more reviews before jumping to any conclusions. IIRC AMD's main claims about IPC uplift were more related to integer rather than FP performace; Cinebench is almost entirely FP, is somewhat 'synthetic' in nature, and as we know is Intel-biased/lacks AMD-specific optimisations. All we have for integer is part of a Geekbench test.

Power/performance is another point, and I suspect the lower TDP parts will show more of an improvement here, with the flagship parts possibly pushing up against the clock/efficiency wall, hence lower clock vs Richland's process.

Still look at those IGP results too.

They should be pushing out the lower TDP parts first just so the first reviews people read are more positive. First impressions count.

They did the same with the A10 5700 and A10 6700,with review sites having to go out of the way to obtain them.

These parts were not any harder to buy at retail than the A10 5800K or A10 6800K.

Yet they were significantly better in terms of performance per watt. However,Intel gets a pass on the efficiency of the Core i3 and AMD looked like it could not compete in anyway,but it could:

http://hexus.net/tech/reviews/cpu/47...0-5700/?page=8

How many people were really aware of the A10 5700 and A10 6700??

As for why we haven't been slapped with an NDA notice yet, we received explicit permission from AMD that we could post this review since we are one of their select system builders that was given permission to do pre-sales.

How the heck can AMD marketing vet such a review?? The first sanctioned review of the A10 7850K makes it look like a flop.

The thing is the CB R15 performance is not better than a far cheaper A10 5800K or Athlon II X4 760K,and I suspect these will overclock far more too.

The same goes with those Geekbench scores.

AMD really needs to be more careful about reviews,otherwise they are doing their competitors advertising for them free.

[watercooled]

Oh yeah you're definitely right about the marketing fails, and under-marketed performance/efficiency of the lower TDP parts, hence my earlier comment.

[CAT-THE-FIFTH]

Oh yeah you're definitely right about the marketing fails, and under-marketed performance/efficiency of the lower TDP parts, hence my earlier comment.

The A8 7600 with a configurable 45W/65W TDP is supposedly priced at around $119,which is around £85.

It seems the best part in the entire desktop range,but I suspect AMD will not bother sending it out for review.

Edit!!

It looks like the review has been taken down now.

Second Edit!!

It seems Hans de Vries has done more analysis of his Kaveri die picture breakdown:

http://semiaccurate.com/forums/showp...postcount=2925

It looks like there is a quad channel memory controller.

[watercooled]

TBH I was thinking it looked like a quad-channel controller - two beefed up controllers being that big just didn't sit right with me, even without the smaller node. It basically looks like two stacked Richland IMCs, and there seems to be too much separation between the two 'stacks' for each side to be part of the same channel.

I wonder if this has something to do with the server side, or perhaps we'll see some quad-channel consumer platform? It could help *massively* with IGP performance, and I see no reason to not use it if it's there!

[CAT-THE-FIFTH]

If that is the case I really hope they use it!

That would make the IGP really shine and probably would beat even a GT3e IGP.

[CAT-THE-FIFTH]

On SA forums they are speculating that there is one memory controller each for the CPU and IGP.

[watercooled]

How would that work though? A separate dual-channel interface for each GPU and CPU? Because that sounds like it goes against the whole unified memory concept introduced with Kaveri.

[CAT-THE-FIFTH]

How would that work though? A separate dual-channel interface for each GPU and CPU? Because that sounds like it goes against the whole unified memory concept introduced with Kaveri.

It does sound kind of a weird concept! But hopefully we will soon find out what it is(hopefully).

BTW,more rumours of AMD doing a phone SOC:

http://basicsmedia.com/advanced-micr...g-formula-8058
http://gulfnews.com/business/technol...ntel-1.1156937

Another one from SA forums.

[watercooled]

Something's just occurred to me - remember those 'leaked' Kaveri benchmarks we saw a few months back? IIRC they showed a drop in FP performance (which would fit in with Cinebench results) but a large increase in integer. Besides one synthetic benchmark, we've seen no integer results yet...

[CAT-THE-FIFTH]

Something's just occurred to me - remember those 'leaked' Kaveri benchmarks we saw a few months back? IIRC they showed a drop in FP performance (which would fit in with Cinebench results) but a large increase in integer. Besides one synthetic benchmark, we've seen no integer results yet...

Good point!! Maybe,AMD is trolling us. Well I hope they are!