Mobile kaveri

[strata8]

Looking at the Xbox One die...

What about approximately half of that?

4 Puma cores
6 GCN CUs
16MB eSRAM (a la Crystalwell)
170-180mm2

I have no idea how 16MB of SRAM would fare though.

 

[Enigmoid]

If your definition of "a lot of the time" is running a power virus, then yes.

Under extreme situations clocks cannot be maintained.

This is for IVB but even in a gaming situation, CPU clocks drop.

http://www.anandtech.com/show/5878/...ation-realtime-igpu-clocks-on-ulv-vs-quadcore

 

[coercitiv]

Under extreme situations clocks cannot be maintained.

This is for IVB but even in a gaming situation, CPU clocks drop.

I see CPU base clocks being maintained, just as I stated in my own example.

The main point I was making is we should not dismiss turbo speeds as long as they can be maintained under CPU only stress. There are plenty of situations where GPU will see little or no stress and the system will greatly benefit from the increased CPU performance.

 

[NTMBK]

Looking at the Xbox One die...

What about approximately half of that?

4 Puma cores
6 GCN CUs
16MB eSRAM (a la Crystalwell)
170-180mm2

I have no idea how 16MB of SRAM would fare though.

That SRAM takes up a lot of die area, and without software explicitly written for it (like on a console) will be tricky to use properly. Just look at Intel's Crystalwell, even with 128MB it can still not perform well if the game is not a good candidate. Just going up to a dual-channel memory architecture would make more difference, compared to Beema/Mullins.

 

[sm625]

Mobile kaveri will be at least 40% slower than haswell in single thread operations. I'm basing that on previous gen comparisons:

AMD A8-5545M APU : Single Thread Rating: 814
Intel Core i5-3337U : Single Thread Rating: 1284

Here's a cat core for comparison:
AMD Athlon 5350 APU : Single Thread Rating: 818

The sad part about these numbers is that the A8-5545M is supposed to be boosting to 2.7GHz during single thread operations (same as the i5), yet it scores the same as a kabini athlon at 2.1GHz. I cannot even find a javascript benchmark of the A8-5545M or any <20W mobile richland. Clearly the big cores are dead. AMD just could never get them working correctly in a < 20W environment. My 6 year old desktop dual core has a passmark single thread rating of over 1100. And even that still has trouble on certain websites like ebay. There's just no way an 800 single thread score can be acceptable in anything costing more than $199.

 

[strata8]

The sad part about these numbers is that the A8-5545M is supposed to be boosting to 2.7GHz during single thread operations (same as the i5), yet it scores the same as a kabini athlon at 2.1GHz. I cannot even find a javascript benchmark of the A8-5545M or any <20W mobile richland. Clearly the big cores are dead. AMD just could never get them working correctly in a < 20W environment. My 6 year old desktop dual core has a passmark single thread rating of over 1100. And even that still has trouble on certain websites like ebay. There's just no way an 800 single thread score can be acceptable in anything costing more than $199.

The Geekbench 3 scores are similar...

A8-5545M: Single thread - 1052, Multi thread - 2826
A6-5200: Single thread - 1026, Multi thread - 3296

The small cores have been encroaching on the big cores for a while now:

Mobile Kaveri will have to improve significantly to pull away.

 

[pw257008]

http://wccftech.com/amd-mobile-kaveri-crystal-series-leaked/

 

[NostaSeronx]

http://wccftech.com/amd-mobile-kaveri-crystal-series-leaked/

Jet = Sun/Hainan
Topaz = Mars/Oland
Opal = Mars/Oland

It's is just rebranded Solar System GPUs. Not the Volcanic Island 2.0 GPUs promised with Kaveri 2.0.

---
It's interesting that they started with low numbers so Kaveri 2.0 would fit in.

FX-7600P
FX-7500
A10-7400P
A10-7300
A8-7200P
A8-7100
A6-7000