AMD Carrizo Pre-release thread

[Abwx]

And do you believe in yourself?

Infinitely more than in your pathetic thread crapping and ad hominem, typical of a disgruntled AMD hater.

Edit : Other than that the perf/Watt advantage over Core M should be 30-35% according to those numbers...


Personal attacks are not allowed here.
Markfw900

 

[DrMrLordX]

Carrizo without iGPU (on Mini-ITX board) would just be the same thing. (ie, Carrizo with defect to Media or display that won't sell in a notebook, etc). Except in this situation the part doesn't even need a new socket (just use BGA).

AM1 isn't pin-compatible with Carrizo to the best of my knowledge, though, even if you ignore the power planes responsible for powering the iGPU. They'd have to change the socket or redo the pinouts or . . . something.

It would probably be easier for them to dump Carrizo with failed iGPU on FM2+.

Still, that assumes that they're going to have a significant number of dud chips. By this point, 28nm planar yields should be very high. The thing to remember about AM1 is that they dump non-failed chips on it in quantity. Just stuff they couldn't sell elsewhere. Sure you get some failed iGPU units on it, but that's an edge case. Those chips alone weren't enough for AMD to bother releasing AM1 in the first place.

So is Carrizo actually coming out, or did AMD not get any OEM deals?

Yes. Probably around the same time that they launch the 300-series Radeons. I'm guessing Computex for Carrizo.

 

[el etro]

Loved Carrizo benchs on Geekbench. The hope now is to see Carrizo delivering what the benches indicate.
And why not use Carrizo for a 14nmLPP pilot test in 1H 2016?

 

[geoxile]

Loved Carrizo benchs on Geekbench. The hope now is to see Carrizo delivering what the benches indicate.
And why not use Carrizo for a 14nmLPP pilot test in 1H 2016?

Maybe it'll use the mythical FD-SOI

 

[mikk]

Dont know exactly what is the reliability of Geekbench, i suppose that comparison of a same uarch evolutions can be accurate enough while comparison between different uarches like in the case below must be weighted at some point, anyway and very grossly the FX8800P should perform like a 2.7GHz Kabini if Geekbench is of any relevance.

http://browser.primatelabs.com/geekbench3/compare/1863979?baseline=2614473

According to the informed at SA the chip is likely running at 2.3, i think that this should still be within a 15W thermal envelloppe, in wich case Carrizo perf/Watt advantage over Core M would be the one provided by a full node advantage, quite ironic given those chips respective processes...

How can you be sure 8800P is a 15W SKU? Isn't it a bit more logical to assume a TDP of 35W since AMD itself confirmed Carrizo TDP range will go up to 35W? 8800P will be the highest Carrizo SKU at launch, it is the first contender for a 35W SKU.

 

[monstercameron]

http://wccftech.com/amds-6th-genera...-3rd-gen-gcn-cores-directx-12-hsa-10-support/

 

[AtenRa]

http://linustechtips.com/main/topic/37825 http://wccftech.com/amds-6th-genera...ort/9-amd-carrizo-launched-15-ipc-gcn-12-hsa/

both links are broken

 

[cbn]

Here is the wccftech link:

http://wccftech.com/amds-6th-genera...-3rd-gen-gcn-cores-directx-12-hsa-10-support/

 

[AtenRa]

Here is the wccftech link:

http://wccftech.com/amds-6th-genera...-3rd-gen-gcn-cores-directx-12-hsa-10-support/

thanks,


this one is all the money.

 

[firewolfsm]

Loved Carrizo benchs on Geekbench. The hope now is to see Carrizo delivering what the benches indicate.
And why not use Carrizo for a 14nmLPP pilot test in 1H 2016?

Translating a chip for a new process is not trivial, so although it might be easier than using a new architecture, it would still involve a lot of work on a potentially obsolete chip (given zen).

 

[Fjodor2001]

55% performance increase according to the slides. Isn't it quite amazing what AMD is able to squeeze out of 28 nm?

 

[frozentundra123456]

How can you be sure 8800P is a 15W SKU? Isn't it a bit more logical to assume a TDP of 35W since AMD itself confirmed Carrizo TDP range will go up to 35W? 8800P will be the highest Carrizo SKU at launch, it is the first contender for a 35W SKU.

Somehow, unless we are not looking at the same geekbench, looks to me like core M is much more efficient. Geekbench scores are all over the place, but the best scores for 5Y10 are very close to those posted for Carrizo, but that is in a 4.5 watt TDP. Even taking the lowest scores, they are about half that of Carrizo, but that is in a 4.5 watt TDP, vs 15 watt for carrizo or even worse if it is 35 watts. And that is not even top of the line core M. And yes, I am sure you know who will argue that core M uses more that 4.5 watts, but there is no assurance Carrizo does not use more than its rated TDP either.

 

[Enigmoid]

55% performance increase according to the slides. Isn't it quite amazing what AMD is able to squeeze out of 28 nm?

Full node shrink?

:whiste:

I'll believe it when I see it.

 

[inf64]

Full node shrink?

:whiste:

I'll believe it when I see it.

That is just CB R15 benchmark and both Kaveri and Carizzo are at 15W. You can see the perf. advantage drops at 35W which is expected.

Good news is that Excavator brings at least 10% IPC improvement in Cinebench which bodes well for Zen. Now Zen should be drastically faster than Piledriver IPC-wise, in the region of 55+% in this popular benchmark at least.

 

[cbn]

thanks,


this one is all the money.

So if I am reading that right at 35 watts for 1C, Excavator is not losing much frequency compared to Steamroller.

 

[Enigmoid]

That is just CB R15 benchmark and both Kaveri and Carizzo are at 15W. You can see the perf. advantage drops at 35W which is expected.

Good news is that Excavator brings at least 10% IPC improvement in Cinebench which bodes well for Zen. Now Zen should be drastically faster than Piledriver IPC-wise, in the region of 55+% in this popular benchmark at least.

You know what this looks like? AMD playing games (just like intel) with TDP.

15W Kaveri doesn't exist. Its 20W + 2-4W FCH for the lowest TDP 2M chip. It looks like they are comparing the 15W SoC Carrizio to Kaveri possibly with or without the FCH included in that number downclocked to 15W (probably without touching voltages).

Note: I'm assuming that is for the entire carrizio chip because even 35W/module is too high for mobile and high even for desktop (no problem with desktop kaveri replacement with 70W for the CPU). The graphics slide comfirms this with "optimized for 15W design point".

Lets look at actual scores. I am assuming the power is CPU only excluding FCH.

Lets look at the top 20W and 35W kaveri chips (FX 7500 and 7600P).

http://forums.anandtech.com/showthread.php?t=2419488

20W FX-7500 (CB11.5)

0.65 - ST
1.80 - MT

http://www.anandtech.com/show/8119/amd-launches-mobile-kaveri-apus/3

35W FX 7600P

0.87 - ST
2.50 - MT

Now its showing kaveri @ 15W so lets assume linear scaling and knock off 25% of the FX-7500's score putting it around 0.49 ST and 1.35 MT. Add AMD's scaling numbers and you get.

15W Carrizio quad
0.67- ST
2.10 MT

At 35W TDP its far simpler
0.97 ST
2.9 MT

Looks like a nice improvement but more on the lines of Haswell-U (little absolute performance increase for the U level chips but a large power savings).

 

[monstercameron]

carrizo adverts ramping up https://www.youtube.com/user/amd nda lift soon?

 

[AtenRa]

You know what this looks like? AMD playing games (just like intel) with TDP.

15W Kaveri doesn't exist. Its 20W + 2-4W FCH for the lowest TDP 2M chip. It looks like they are comparing the 15W SoC Carrizio to Kaveri possibly with or without the FCH included in that number downclocked to 15W (probably without touching voltages).

Note: I'm assuming that is for the entire carrizio chip because even 35W/module is too high for mobile and high even for desktop (no problem with desktop kaveri replacement with 70W for the CPU). The graphics slide comfirms this with "optimized for 15W design point".

Lets look at actual scores. I am assuming the power is CPU only excluding FCH.

Lets look at the top 20W and 35W kaveri chips (FX 7500 and 7600P).

http://forums.anandtech.com/showthread.php?t=2419488

20W FX-7500 (CB11.5)

0.65 - ST
1.80 - MT

http://www.anandtech.com/show/8119/amd-launches-mobile-kaveri-apus/3

35W FX 7600P

0.87 - ST
2.50 - MT

Now its showing kaveri @ 15W so lets assume linear scaling and knock off 25% of the FX-7500's score putting it around 0.49 ST and 1.35 MT. Add AMD's scaling numbers and you get.

15W Carrizio quad
0.67- ST
2.10 MT

At 35W TDP its far simpler
0.97 ST
2.9 MT

Looks like a nice improvement but more on the lines of Haswell-U (little absolute performance increase for the U level chips but a large power savings).

Yea Haswell performance at the same 15W TDP with HIGHER than HD6000 graphics. And to think this is still on 28nm PLANAR when Haswell is on 22nm FinFets AND with Bulldozer architecture. It is amazing what they have accomplished.

 

[AtenRa]

So if I am reading that right at 35 watts for 1C, Excavator is not losing much frequency compared to Steamroller.

Yea looks like Turbo clocks on the 35W TDP SKUs will be almost the same minus 100-200MHz.

 

[csbin]

AVX2、MOVBE、SMEP、BMI1/2

 

[Enigmoid]

Yea Haswell performance at the same 15W TDP with HIGHER than HD6000 graphics. And to think this is still on 28nm PLANAR when Haswell is on 22nm FinFets AND with Bulldozer architecture. It is amazing what they have accomplished.

Note sure how you think 0.65/2.1 is Haswell performance unless you are comparing to the lowest level chips. Its only a tad over Beema (A8-6410 is 0.6/2.0 in CB 11.5).

Furthermore, my comparison is for 15W without the FCH (if the FCH is included in the 15W kaveri number, carrizio performance drops by ~ 10%).

3dmark is disingenuous to compare igp's. Too much driver optimization (especially on intel's side) and almost no CPU load during that benchmark. Too short for long term throttling.

 

[AtenRa]

Note sure how you think 0.65/2.1 is Haswell performance unless you are comparing to the lowest level chips. Its only a tad over Beema (A8-6410 is 0.6/2.0 in CB 11.5).

Furthermore, my comparison is for 15W without the FCH (if the FCH is included in the 15W kaveri number, carrizio performance drops by ~ 10%).

3dmark is disingenuous to compare igp's. Too much driver optimization (especially on intel's side) and almost no CPU load during that benchmark. Too short for long term throttling.

Core i3 4030U 15W TDP Haswell Cinebench 11.5 ST = 0.8 , MT = 2.0

 

[coercitiv]

Core i3 4030U 15W TDP Haswell Cinebench 11.5 ST = 0.8 , MT = 2.0

I doubt that i3 is anywhere near it's rated TDP in Cinebench. See bellow, i7 is around 10W estimated package power at 1.9Ghz - link to HWInfo log for this run.

 

[el etro]

Teams that made the Bulldozer updates did their work very well. There's not much to do in a fundamentally failed uarch.

First thing to do in Zen is to put more execution and Backend resources so them will be going out of module approach back to the big core(like Phenom!).

 

[monstercameron]

http://anandtech.com/show/9319/amd-...p-leap-of-efficiency-and-architecture-updates