AMD on track for launch of Kaveri in February 2014

[NTMBK]

No we don't. Maximum official for Trinity is 2133MHz for the 6800K only, everything else is 1866MHz (not counting OC obviously). I might add that my own 6800K has serious trouble with 2400MHz, it requires EXTREMELY conservative timings (15-15-15-37 CR2) to even post. Weather that is just mine or its general for 6800K's, I don't know.

Ah, my bad, I thought Trinity could handle 2400. Thanks for the info.

 

[ShintaiDK]

I doubt you can run DDR3 and DDR4 on the same controller. DDR4 DIMMs are also 288pins and uses point to point. And when Kaveri is replaced and maybe the FM2+ socket too in favour of FM3. Then DDR4 should be cheap enough.

 

[Insert_Nickname]

Ah, my bad, I thought Trinity could handle 2400. Thanks for the info.

I'm not saying it can't handle 2400, but those timings do somewhat negate the benefit of the extra 133MHz(266MHz DDR) on the memory, which makes running 2400 kind of pointless for me...

Oh, yeah. That kit I was using when I tested was rated at 2666MHz, so its not the memory.

 

[Homeles]

Correct me if I'm wrong here but there isn't gonna be a quad or even dual channel schema for DDR4 right? I was under the impression that point-to-point topology doesn't allow for such a thing, I could be wrong though

There will still be channels. The difference is that each socket corresponds to a specific bus. Say you have 2 ram slots, and have a dual channel-capable chip, like a theoretical i5 2500K with DDR4 support -- slot 1 goes to controller 1, slot 2 goes to controller 2.

You cannot have more than 1 DIMM per controller. This isn't an issue, since DDR4 has higher bandwidth and much higher density. This also reduces a lot of complexity; from my understanding, many of the clock speed bumps we've seen since the original DDR have been due to changes made every generation to reduce complexity. DDR3 well exceeded its frequency expectations; it'll be interesting to see how DDR4 fares as time goes on.

 

[AtenRa]

No we don't. Maximum official for Trinity is 2133MHz for the 6800K only, everything else is 1866MHz (not counting OC obviously). I might add that my own 6800K has serious trouble with 2400MHz, it requires EXTREMELY conservative timings (15-15-15-37 CR2) to even post. Weather that is just mine or its general for 6800K's, I don't know.

You have to raise the NB voltage to run at 2400MHz and above with better timings. I could run at 2400MHz with 11-13-13 on the A10-6800K in my review (link in sig bellow).

 

[Insert_Nickname]

You have to raise the NB voltage to run at 2400MHz and above with better timings. I could run at 2400MHz with 11-13-13 on the A10-6800K in my review (link in sig bellow).

So there is a "trick" involved. I'll see if I can borrow that kit again, to try it out. :sneaky:

Any recommendations on voltage and do I need to run the northbridge faster too?

 

[AtenRa]

So there is a "trick" involved. I'll see if I can borrow that kit again, to try it out. :sneaky:

Any recommendations on voltage and do I need to run the northbridge faster too?

As you can see i didnt go crazy on the vcore, the CPU easily got at 4.8GHz with WC (Havent tryed OC on Air to tell you the truth). NB frequency helps with the the RAM, so raise it a little will help but dont expect a huge increase in performance.

 

[Insert_Nickname]

As you can see i didnt go crazy on the vcore, the CPU easily got at 4.8GHz with WC (Havent tryed OC on Air to tell you the truth). NB frequency helps with the the RAM, so raise it a little will help but dont expect a huge increase in performance.

Thanks. Its just for trying to see weather or not it'll work.

 

[NostaSeronx]

Kaveri Mobile^

20% faster than A10-5750M(CPU)
30% faster than 8650G(GPU)

via pre-silicon projection

 

[frozentundra123456]

Kaveri Mobile^

20% faster than A10-5750M(CPU)
30% faster than 8650G(GPU)

via pre-silicon projection

Ahh... the infamous "up to" increases qualified by being not based on actual silicon. Doesnt tell us much really.

 

[Vesku]

Ahh... the infamous "up to" increases qualified by being not based on actual silicon. Doesnt tell us much really.

Especially when they were showing off "actual Kaveri chips" during the Summer. Then again, NostaSeronx, doesn't say when this slide was created/distributed.

 

[NostaSeronx]

Then again, NostaSeronx, doesn't say when this slide was created/distributed.

October 23, 2013 is when it popped up on the internet.

Ilya Perminov (Can't Identify)
Timur Paltashev (Senior Manager at AMD)
are the authors.

 

[CHADBOGA]

Kaveri Mobile^

20% faster than A10-5750M(CPU)
30% faster than 8650G(GPU)

via pre-silicon projection

If this slide is current/representative of Kaveri's performance characteristics, Kaveri will underwhelm big time.

 

[AtenRa]

If this slide is current/representative of Kaveri's performance characteristics, Kaveri will underwhelm big time.

Yeap,

1. 20% faster CPU performance, we dont know at what frequency but i bet lower than last gen. Making the SR core really powerful.
2. Up to 30% faster in GPU performance, is that on 3D or Compute or both ??
3. More features than Trinity/Ritchland. New UVD and VCE, PCIe Gen 3, Lightning Bolt and more.
4. Lower power consumption with higher battery life.
5. Smaller die size than Trinity/Ritchland, on a cheap fab process.

Yes it will be a big failure

 

[piesquared]

Kaveri Mobile^

20% faster than A10-5750M(CPU)
30% faster than 8650G(GPU)

via pre-silicon projection

Looks awesome! And this is just in traditional/dated old school software. Can't wait to see all the goodies in Kaveri get exploited. It's going to just keep getting more powerful as time goes on. Kaveri would make a great developer platform. I wish I were a developer right now, HSA enables all kinds of innovation.

 

[Vesku]

The estimated battery life is good, but again it is estimated. They've had 28nm GF samples in hand for quite a while now so to be putting out information based on pre-silicon projections is underwhelming. Even more so coming from a company that has a hard time meeting projections.

 

[NostaSeronx]

1. 20% faster CPU performance, we dont know at what frequency but i bet lower than last gen. Making the SR core really powerful.
2. Up to 30% faster in GPU performance, is that on 3D or Compute or both

Since, it is referring to the mobile platform:

1.8 GHz and 0.5 GHz

Richland is:
2.5 GHz and 0.72 GHz.

Format: CPU Clock and GPU Clock.

 

[IntelUser2000]

The estimated battery life is good, but again it is estimated.

11 hours in MobileMark07, which uses very light tests and has only 60 nits brightness, with a 62WHr battery.

The Vaio Duo 13 Haswell gets 11 hour battery with MobileMark 2012 which has more demanding tests and 150 nits brightness, with a 47WHr battery. The Duo 13 with MobileMark07 test achieves somewhere in the 16-17 hour territory. Most Haswell U based systems achieve 8-9 hours in a MM12 scenario with ~45WHr battery.

That means the battery life of Kaveri isn't really better than Richland.

 

[piesquared]

So dual graphics with Crystal series. Combined with Kaveri that should make a pretty bad ass mobile platform.
Really will be interesting to see what project discovery is all about. No doubt there will be some cool innovative designs there.

 

[Vesku]

11 hours in MobileMark07, which uses very light tests and has only 60 nits brightness, with a 62WHr battery.

The Vaio Duo 13 Haswell gets 11 hour battery with MobileMark 2012 which has more demanding tests and 150 nits brightness, with a 47WHr battery. The Duo 13 with MobileMark07 test achieves somewhere in the 16-17 hour territory. Most Haswell U based systems achieve 8-9 hours in a MM12 scenario with ~45WHr battery.

That means the battery life of Kaveri isn't really better than Richland.

Lenovo X1 Carbon, 7h 19m with a 45 WHr battery MobileMark 2007.

http://www.pcmag.com/article2/0,2817,2408289,00.asp

So projected battery use is good, especially considering Kaveri is not on Intel's 22nm process but GF's 28nm. However, as I've been saying, it still remains a projection from a company with a shaky track record regarding such things.

 

[bullzz]

Lenovo X1 Carbon has ivy bridge, not haswell..

 

[NostaSeronx]

Lenovo X1 Carbon has ivy bridge, not haswell..

Not 100% sure, but the battery life target is for the 35W TDP Kaveri.

i5-3427U -> dual-core Ivy Bridge @ 17w TDP.
i7-4500U -> dual-core Haswell @ 15w TDP.
A10-6750M -> quad-core SteamrollerB @ 35w TDP.

 

[CHADBOGA]

Yeap,

1. 20% faster CPU performance, we dont know at what frequency but i bet lower than last gen. Making the SR core really powerful.

Not 20% faster, but rather up to 20% faster.

You do know what this generally means when translated from marketing speech, to the real world, don't you?

2. Up to 30% faster in GPU performance, is that on 3D or Compute or both ??

As it is going up to 30% faster in GPU, what will its competition be bringing later in 2014?

 

[NostaSeronx]

You do know what this generally means when translated from marketing speech, to the real world, don't you?

You do realize in one benchmark or application or workload that they expect SteamrollerB to be 20%(exact) faster than Richland.

It up to 20% because they don't expect it to be a 20% across the board.
---
CPU Benchmark -> PCMark 8
GPU Benchmark -> 3DMark Fire Strike
^--based on previous marketing slides show the tendency for PCMark and 3DMark

^We'll see this most likely.

For "2013 A-Series 'Richland'" -> "2014 A-Series 'Kaveri'"

 

[USER8000]

It will be interesting to see the improvement over Trinity too. It will be also interesting to see how much power consumption on the desktop models has gone down when compared to Richland.