AMD releases 7000 series Carrizo-L (Puma+)

[ShintaiDK]

http://www.anandtech.com/show/9246/amds-carrizo-l-apus-unveiled-12-25w-quad-core-puma

No baseclock and no GPU clock listed tho.

Also AMDs press. Including the new APU prices (But not for the Carrizo-L chips):
http://www.amd.com/en-us/press-releases/Pages/new-7000-series-apu-2015may06.aspx

 

[coercitiv]

I was expecting to see 2.6Ghz+ for their top tier SKU at 25W. Let's hope those 2.5Ghz are somewhat sustainable. (as in let's hope base speeds are good)

 

[jji7skyline]

Are these FM2 CPUs?

 

[ShintaiDK]

Are these FM2 CPUs?

No. FP4 BGA.

 

[DrMrLordX]

Carrizo-L is not puma+. It is puma. In other words, rebadged Mullins with higher clocks and maybe some different GCN options. Allegedly, Nolan will be the same thing, with different GCN options.

Puma+ is MIA.

 

[erunion]

Nolan is cancelled, as far as I can tell. AMD removed all 20nm parts.

 

[Fjodor2001]

Kind of confusing naming it Carrizo-L isn't it? Gives you the impression it has a close resemblance to Desktop Carrizo, when in fact it hasn't, and instead is Puma core based.

It would be like naming the Intel Atom CPUs as Intel Core-i5-L or something.

 

[ShintaiDK]

Kind of confusing naming it Carrizo-L isn't it? Gives you the impression it has a close resemblance to Desktop Carrizo, when in fact it hasn't, and instead is Puma core based.

It would be like naming the Intel Atom CPUs as Intel Core-i5-L or something.

Thats just a codename. So that part is irrelevant.

However the final naming is a problem.

AMD A8-7410 vs A6-7400K for example.

 

[ikachu]

So my insider info is getting more and more out of date, but if I remember correctly, Puma was just the codename for Jaguar fabbed on GF instead of TSMC. Puma+ incorporates a few performance improvements, and was already released with Mullins.

One thing I remember that was added in Puma+ was some string stuff controlled by MSRC001_1020.

From the Mullins BKDG:

52 LsWcPlusFastStringEn. Read-write. Reset: 0. BIOS: 1. 1=Enable string optimization on destination
memtype of WC+.
51 LsWcFastStringEn. Read-write. Reset: 0. BIOS:1. 1=Enable string optimization on destination
memtype of WC.
...
24 WcPlusEn: write combining plus enable. Read-write. Reset: 0h. BIOS: 1. 0=Memtype WC+ is
treated as write protect (WP). 1=Treat WC+ as per APM: loads are not cacheable, stores are coherent
WC.

 

[Fjodor2001]

Thats just a codename. So that part is irrelevant.

I don't think it's irrelevant. It signals the CPU generation and uArch used for the complete Carrizo-L series.

 

[Enigmoid]

However the final naming is a problem.

AMD A8-7410 vs A6-7400K for example.

Agree. They need an 'm' for mobile or something.

 

[el etro]

Puma is the Jaguar fabbed at GF, but with some other power management improvements.

AMD did not put a real new cat arch since the first Jaguar back in 2012. Intel heavy charge with Bay Trail(and many bilions of $) probably scared them from invest here more.


Puma+ is the Carrizo-L.

 

[DrMrLordX]

Nolan is cancelled, as far as I can tell. AMD removed all 20nm parts.

Actually, what it looks like is that they've taken Mullins and added some GCN cores to it and named it Nolan. 20nm Puma+ Nolan is canned so far as I can tell.

Puma is the Jaguar fabbed at GF, but with some other power management improvements.

Eh? That doesn't seem right. Kabini is Jaguar, Mullins is Puma, and Mullins was faster/better than Jaguar at the same clock with the same number of cores, was it not? I know that Mullins was not Puma+.

Puma+ is the Carrizo-L.

That contradicts what I've heard, and what AMD has said in press releases. They admitted as far back as February that Carrizo-L would not be Puma+. Puma+ doesn't look like it's going to show up anywhere now that Nolan has been . . . subverted.

 

[coercitiv]

Eh? That doesn't seem right. Kabini is Jaguar, Mullins is Puma, and Mullins was faster/better than Jaguar at the same clock with the same number of cores, was it not? I know that Mullins was not Puma+.

From Anandtech's perf preview, Beema/Mullins was Puma+. It even contains AMD slides to confirm, so it seems to me AMD is even more confused than we are.

Like their predecessors, Beema and Mullins combine four low power AMD x86 cores (Puma+ this time, instead of Jaguar) with 128 GCN based Radeon GPU cores.

Kabini/Temash was fabbed on TSMC 28nm, while Beema/Mullins was fabbed on GF 28nm. The chips were a bit different, Puma+ had more transistors. (the ARM A5 core I think). AFAIK all the performance difference was due to the new turbo boost speeds, Beema saw a 2.4Ghz top speed compared to the 2.05Ghz of Kabini.

Also to note: Puma was used to designate a AMD mobile platform back in 2008.

 

[DrMrLordX]

Huh, weird. At this point the puma/puma+ designation seems almost meaningless if they insist on using it in such a way.

 

[Enigmoid]

Eh? That doesn't seem right. Kabini is Jaguar, Mullins is Puma, and Mullins was faster/better than Jaguar at the same clock with the same number of cores, was it not? I know that Mullins was not Puma+.

There are no meaningful IPC improvements (if any which is most likely to be the case) between Jaguar and Puma.

 

[Abwx]

There are no meaningful IPC improvements (if any which is most likely to be the case) between Jaguar and Puma.

Granted but this chip perfs are simply very good, too bad they didnt upgrade it with a second memory channel and say 128SPs, IPC would be more than enough given the 1.8-2.5 frequencies.

As said it still hold the perf/Watt crown in its TDP segment despite the release of the much marketed Core M, it will be interesting to see how AMD s BDW opponent fare against its own little brother, GPU wise it s a no contest but CPU wise the little core has still some strength left.

http://browser.primatelabs.com/geekbench3/compare/2561640?baseline=1863979

http://browser.primatelabs.com/geekbench3/compare/2559986?baseline=1863979

The detailed scores match more or less the comparisons made at Hardware.fr between this kinds of chips (they used a C2Q 6600 instead).

 

[Shivansps]

Lets hope for dual channel, since they are desgined to share the same FT4 BGA with Carrizon there is still hope, if is still SC they dont have the big IGP lead over Braswell. And the E1-7010 is pure and complete crap, 1333 max and a dual up to 1.5ghz? wth AMD.

And TDP starts to be a problem now if they are in the same sector as Braswell.

 

[Abwx]

Carrizo-L is a Beema/Mullins in disguise so its specs are definitly the same, memory bandwith and GPU size wise.

Despite the presence of this odd 1.5 dual core the rest of the line is quite solid with base frequencies starting at 1.8 with the 4C A4-7210.

As for competing with Braswell it shouldnt be a problem for the low power Mullins variants, recent reviews of the former point to IPC being unchanged while 14nm doesnt bring what it was supposed to in perf/Watt, as such Braswell GPU has good scores but its perf/Watt is still unknown.

 

[Phynaz]

So this is just a rebadge?

 

[Abwx]

So this is just a rebadge?

Thermal management has been changed and TDP extended from 15 to 25W for short periods, thoses should boost with all cores at close to/up to max frequencies.

That s better than nothing and indeed in thoses power enveloppes there s not much you can do, you ll be limited by the power at disposal to the point that it s likely that Carrizo-L, aka Jaguar cores, will still be the best in perf/Watt at 14-15W in the coming months, this allow AMD to push back any new design.

 

[lefty2]

According to this Anandtech article Puma+ is a new core: http://www.anandtech.com/show/9246/amds-carrizo-l-apus-unveiled-12-25w-quad-core-puma

Carrizo-L features ‘Puma+’, which by virtue of the naming scheme suggests an updated version of Puma which was seen in Beema.

That is confusing because latest AMD documents say Beema is Puma, while older ones say its Puma+.
I think the older documents referencing Beema as Puma+ are wrong. Wikipedia has Beema using Puma: http://en.wikipedia.org/wiki/Puma_(microarchitecture)

 

[coercitiv]

According to this Anandtech article Puma+ is a new core: http://www.anandtech.com/show/9246/amds-carrizo-l-apus-unveiled-12-25w-quad-core-puma

That is confusing because latest AMD documents say Beema is Puma, while older ones say its Puma+.
I think the older documents referencing Beema as Puma+ are wrong. Wikipedia has Beema using Puma: http://en.wikipedia.org/wiki/Puma_%28microarchitecture%29

See my previous post. Even AMD designated the cores inside Mullins as Puma+. I'll take AMD over Wikipedia when AMD products are concerned.

 

[Abwx]

These Puma and Puma + should be about the same chips, Mullins/Beema are apparently Pumas while Carrizo-L is allegedly a Puma+, dont know what is the difference if ever there s one.

Also the TSMC fabbed chips top at 2.0 with about 18-19W TDP, a long shot from GF s fabbed Beemas/Kabinis (the Athlon 5350 is likely a Beema with features disabled) that are at 14-15W at the same frequency, the slides published by AMD are actualy accurate about the huge perf/Watt improvements when porting this design at GF.

 

[lefty2]

See my previous post. Even AMD designated the cores inside Mullins as Puma+. I'll take AMD over Wikipedia when AMD products are concerned.

The latest AMD document shows Beema as Puma: http://images.anandtech.com/doci/9246/Mobility Roadmap_678x452.png