The Kaveri Pre-Launch Thread (A10-7800 and A10-6800k @3,5 Ghz)

[Fjodor2001]

You compare apples and oranges. GPU part is more dense by nature. Much more. As example before, Pitcairn 212mm2 2.8B. Or 13.2M/mm2 by your math.

Well the 4770K has a quite large iGPU too. So since you still make such a statement, how many percent of the 4770K die area is occupied by iGPU? And similarly for the A10-7850K die?

 

[inf64]

L3 cache is also taking a solid part of Haswell's die.

 

[Abwx]

Remember the APU on 32nm GloFo? 1.3B/246mm2. Or 5.2M/mm2.

Yes , we remember that its GPU was 42% of the die , about
the same as Kaveri s 47%....

Well the 4770K has a quite large iGPU too. So since you still make such a statement, how many percent of the 4770K die area is occupied by iGPU

31%.

 

[naukkis]

But with native 256bit pipelines in Carizo it will just blast SR core in AVX/FMA stuff so workloads that were running better on intel before will see major shrinkage of the gap with Excavator

There's no point of widening CPU's SIMD pipelines if there is possibility to use gpu SIMD for highly parallelizable workloads. Much better perf/watt with gpu. Even Haswell shouldn't have go for wider SIMD-parts resulting higher power use and lower clock speeds hampering all but niche workloads performance, which still would be better suited for hsa and GPU offloading. I hope AMD finds it's own way and won't just follow Intel's moves.

 

[ShintaiDK]

I actually wonder if the 2.41B number is true. AMD also first claimed BD was 2B, then revised to 1.2B. Looking on the Hawaii numbers for GCN1.1, the GPU in Kaveri should be around 1.1B transistors roughly. I dont see Streamroller using 1.3B, the same as a 4M/8T with L3 BD/PD chip. Or what the entire Richland chip uses.

 

[monstercameron]

I actually wonder if the 2.41B number is true. AMD also first claimed BD was 2B, then revised to 1.2B. Looking on the Hawaii numbers for GCN1.1, the GPU in Kaveri should be around 1.1B transistors roughly. I dont see Streamroller using 1.3B, the same as a BD/PD chip. Or what the entire Richland chip uses.

what about the extra decoder stuff and hq features and trueaudio?
[note; speaking mile high level here]

 

[Fjodor2001]

Yes , we remember that its GPU was 42% of the die , about
the same as Kaveri s 47%....



31%.

Ok, so a 47-31=16% difference in percent of die area occupied by iGPU when comparing A10-7850K to 4770K. Not that much. And it certainly cannot explain the density difference alone.

Also, note that Intel's process technology is supposed to be 22 nm vs GloFo/AMD's which is 28 nm. So you'd expect the 22 nm process to be denser than the 28 nm process, even taking into account the 16% difference in iGPU die area percentage.

And yet the A10-7850K has 9.80M transistors/mm^2 vs 4770K at 7.91M transistors/mm^2. I would have expected about the opposite density advantage in Intel's favor instead. Interesting to see that is not the case! 😎

 

[ShintaiDK]

what about the extra decoder stuff and hq features and trueaudio?
[note; speaking mile high level here]

Still doesnt add up. Its all tiny parts.

 

[ShintaiDK]

Ok, so a 47-31=16% difference in percent of die area occupied by iGPU when comparing A10-7850K to 4770K. Not that much. And it certainly cannot explain the density difference alone.

Also, note that Intel's process technology is supposed to be 22 nm vs GloFo/AMD's which is 28 nm. So you'd expect the 22 nm process to be denser than the 28 nm process, even taking into account the 16% difference in iGPU die area percentage.

The A10-7850K has 9.80M transistors/mm^2 vs 4770K at 7.91M transistors/mm^2. I would have expected about the opposite density advantage in Intel's favor instead. Interesting to see that is not the case! 😎

Still apple and oranges. Plus the raised question if Kaveri actually got 2.41B transistors.

Remember the APU on 32nm GloFo? 1.3B/246mm2. Or 5.2M/mm2.
Or BD on 32nm GloFo. 1.2B/315mm2. Or 3.81M/mm2. Same process as the APU above.

 

[Abwx]

Ok, so a 47-31=16% difference in percent of die area occupied by iGPU when comparing A10-7850K to 4770K. Not that much. And it certainly cannot explain the density difference alone.

Also, note that Intel's process technology is supposed to be 22 nm vs GloFo/AMD's which is 28 nm. So you'd expect the 22 nm process to be denser than the 28 nm process, even taking into account the 16% difference in iGPU die area percentage.

The A10-7850K has 9.8M/mm^2 vs 4770K at 7.91M/mm^2. I would have expected about the opposite density advantage in Intel's favor instead. Interesting to see that is not the case! 😎

Hawai GPU has about 14M/mm2 so Kaveri s 28nm is still well below TSMCs 28nm density , as for Intel they surely choosed to increase speed at the expense of density , slightly increasing the distances will reduce notably the parasistic strain capacitances and reduce accordingly dynamic lossses,
i guess that a given process is firstly a compromise.

 

[Fjodor2001]

Still apple and oranges.

Not apples and oranges enough to explain the density advantage that GloFo's 28 nm has over Intel's 22 nm. Only 16% difference in die area occupied by iGPU cannot explain that alone.

Plus the raised question if Kaveri actually got 2.41B transistors.

If you have some more accurate info than what can be found in AMD's public slides, please share it.

 

[Haserath]

You compare apples and oranges. GPU part is more dense by nature. Much more. As example before, Pitcairn 212mm2 2.8B. Or 13.2M/mm2 by your math.

Remember the APU on 32nm GloFo? 1.3B/246mm2. Or 5.2M/mm2.
Or BD on 32nm GloFo. 1.2B/315mm2. Or 3.81M/mm2. Same process as the APU above.

Intels 32nm with SB. 1.16B/216mm2. Or 5.3M/mm2. Surprisingly better than the APUs. But again, cache is also dense.

Intel's IGP is actually less dense than their CPU/cache logic.

Quad GT2: 1.4B-177mm
Dual GT3: 1.3B-181mm

So design choice and available process both play big roles.

 

[inf64]

Not apples and oranges enough to explain the density advantage that GloFo's 28 nm has over Intel's 22 nm. Only 16% difference in die area occupied by iGPU cannot explain that alone.

If you have some more accurate info than what can be found in AMD's public slides, please share it.

He is comparing apples and oranges. TSMC and GloFo's process nodes are not comparable in that regard.

 

[ShintaiDK]

He is comparing apples and oranges. TSMC and GloFo's process nodes are not comparable in that regard.

And GloFo to GLoFo? 😉

 

[monstercameron]

speaking about density...check this out, kaveri in a briefcase!
http://apusilicon.com/the-cats-out-of-the-bad-but-amd-still-plays-coy/

 

[littlebluesmurf]

The chart shows desktop shipments. Not APUs specificly. So you need to look at the entire desktop product mix. I was simply being generous only using the FM2 socket to compare.

So the chart showing just desktop A10 APU's is now not showing APU's specifically. 🙄

 

[inf64]

And GloFo to GLoFo? 😉

Not necessarily the same as AMD might have used high density libraries on different parts of different APU designs.

 

[mrmt]

Well the 4770K has a quite large iGPU too. So since you still make such a statement, how many percent of the 4770K die area is occupied by iGPU? And similarly for the A10-7850K die?

GT2 is around 33% of the die area, or roughly 65mm^2, A10 7850 has 47% of the area devoted to iGPU, or 115mm^2.

 

[Abwx]

Not necessarily the same as AMD might have used high density libraries on different parts of different APU designs.

They got 89% higher density than on 32nm, almost a full node.

The die has 85% more transistors than Trinity/Richland ,
the theorical throughput gain is 36% so they did very well
assuming the meager early infos are accurate.

 

[frozentundra123456]

So the chart showing just desktop A10 APU's is now not showing APU's specifically. 🙄

That graph is actually meaningless without a numerical y axis. It is impossible to tell if there is a large increase, or if it just appears that way because the "y" axis only shows a small portion of the total scale, which is what I suspect.

 

[Homeles]

That's even higher density than Intel's 22 nm process! :awe:

Intel 4770K is 1.4B transistors on 177mm^2. So it's:
Intel 22 nm: 1.4B/177mm^2=7.91M/mm^2
GloFo/AMD 28 nm: 2.4B/245mm^2=9.80M/mm^2

And yet it's 22 nm vs 28 nm. Do those nm numbers actually mean anything anymore? 😕

They don't mean anything. I don't mean to be rude, but have you been living under a rock?

But no, GloFo's 28nm process is not denser than Intel's 22nm process. They have minimum SRAM cell sizes of 0.120 µm2 and 0.092 μm2 respectively.

 

[Spawne32]

I wonder why AMD just doesn't use that 45W model as a base for mobile chip? It would sell for sure and command higher price than most of their mobile lineup. Pair it with some 7750 ddr3 version for HCF and you get a nice laptop for games.

I think AMD is kinda out of the game when it comes to mobile processors if you discount mobile gaming laptops, or anything performance oriented for graphics. Haswell's TDP ranges anywhere from 11.5w to 57w across the entire mobile range from the 2961Y Celeron notebook processor to the 4930MX beast with the HD4600 graphics. In a time where 9 hour+ battery lifes are a deciding factor in netbook or small form factor laptop purchases, AMD just cant compete still. The deciding factor behind my purchase of my dell inspiron 11 was exclusively the fact that it used the haswell 15w 2955U, which made the battery life absolutely amazing for a 3 cell lithium ion.

Like i said though, alot of this is just speculation until we get some solid reviews from good sites about what its capable of. I for one am very excited to see what the results are and my hope is that kaveri puts AMD back within some range of competition with intel again. We are once again in a situation where Intel having zero competition has allowed them to price their parts exceptionally high. I have never in my life paid more then 250 dollars for a processor for any of my PC's including my gaming PC's.

 

[ShintaiDK]

I have never in my life paid more then 250 dollars for a processor for any of my PC's including my gaming PC's.

Today 250$ gives you about the fastest desktop CPU money can buy for consumer usage. 10 years ago when X2 ruled, 250$ wouldnt even buy you the cheapest X2. Looking back further, 250$ would give you even less. And thats even without accounting for inflation.

We are once again in a situation where Intel having zero competition has allowed them to price their parts exceptionally high.

Funny since CPUs have never been cheaper since the "no competition" began in 2006. And if you think on why, you would know competition got nothing to do with it anymore. Its a simple volume/margin ratio. And its obvious that Intel think we are at the sweet spot and been so for years.

 

[Abwx]

I think AMD is kinda out of the game when it comes to mobile processors if you discount mobile gaming laptops, or anything performance oriented for graphics. Haswell's TDP ranges anywhere from 11.5w to 57w across the entire mobile range from the 2961Y Celeron notebook processor to the 4930MX beast with the HD4600 graphics. In a time where 9 hour+ battery lifes are a deciding factor in netbook or small form factor laptop purchases, AMD just cant compete still. The deciding factor behind my purchase of my dell inspiron 11 was exclusively the fact that it used the haswell 15w 2955U, which made the battery life absolutely amazing for a 3 cell lithium ion.

Like i said though, alot of this is just speculation until we get some solid reviews from good sites about what its capable of. I for one am very excited to see what the results are and my hope is that kaveri puts AMD back within some range of competition with intel again. We are once again in a situation where Intel having zero competition has allowed them to price their parts exceptionally high. I have never in my life paid more then 250 dollars for a processor for any of my PC's including my gaming PC's.

Kaveri will help a lot on the mobile side wich is currently
not that bad at all...

http://www.pcper.com/reviews/Graphi...rinity-and-Richland-Integrated-Graphics/Perfo

 

[Spawne32]

Today 250$ gives you about the fastest CPU money can buy. 10 years ago when X2 ruled, 250$ wouldnt even buy you the cheapest X2. Looking back further, 250$ would give you even less. And thats even without accounting for inflation.



Funny since CPUs have never been cheaper since the "no competition" began in 2006. And if you think on why, you would know competition got nothing to do with it anymore. Its a simple volume/margin ratio. And its obvious that Intel think we are at the sweet spot and been so for years.

The fastest "mainstream" processor right now is the 4770k i7 which is 339.99 on newegg while the i7-3960X Extreme Edition is $1,069.99. Not much has changed about intels lineup on pricing in 10 years. There were always high end mainstreams priced between 300-500 dollar's and "extreme edition" processors priced between 650-1000 dollars.