AMD - Piledriver chitchat

[DanceswithUnix]

I've just spotted something interesting on this article: http://www.extremetech.com/computing...come-to-market

The bit about Atom using x86 instructions internally - I wasn't aware of that (well I hadn't given it much thought TBH)! I can't seem to find any more details on it though, I wonder if it applies to old or new Atom, or both?

I think all the out of order x86 chips have been ones that translate the code internally, from the K5 to the modern stuff. Cyrix were the last ones to do native x86 execution that I know of.

So perhaps original Atom is in order and hence native, but I will still take that with a pinch of salt.

[watercooled]

I'm finding it hard to search for details as mentioning Atom, CISC, translate, x86 etc all seem to turn up results related to ARM vs x86 and/or Android binary translation on Atom.

However I threw µop into the search terms and came across Agner Fog's microarchitecture pdf. See page ~146: http://www.agner.org/optimize/microarchitecture.pdf

And in the process of typing this reply, I also found this: http://www.anandtech.com/show/2493/9

So, it sounds more like a hybrid than strictly CISC execution to me; most instructions = one µop out of the decode stage/left intact, but some are still broken down to multiple µop or fused to one. But then the author of the VIA article only said Atom uses x86 instructions, and it sounds like it does. Kind of...?

[watercooled]

Carrizo rumour about on-package memory: http://www.eteknix.com/amds-carrizo-...am-on-package/

Apply salt as required; it's not the first time we've heard about this, but it would certainly make a lot of sense.

[CAT-THE-FIFTH]

Bah,seeing people quoting the TR review as indicating the Pentium dual core is "better" for gaming in Crysis3 than an AMD chip in HUKD now. The TR reviewer's credibility is no better than that of the lot on Bit-tech now IMHO. This is the same game where my Core i3 had to be ditched for a Core i5 since it became a limiting factor for me.

Edit!!

It looks TR has fudged something here.

Earlier review using an AVX2 encoder(same test as the new review):

http://techreport.com/r.x/core-i7-4770k/x264.png



New review:

http://techreport.com/r.x/pentium-g3258-oc/x264.gif



Typo?? However,their CB results indicate otherwise. Looks like they have been mucking around with things slightly.

[CAT-THE-FIFTH]

A10-7800 spotted in a Medion desktop in Europe for just over £300:

http://www.fudzilla.com/home/item/35...4000-e-desktop

Edit!!

Tested here:

https://translate.google.com/transla...-10125917.html

It seems to use a 350W FSP PSU,but has no PCI-E power connectors.

Edit!!

Hynix lists HBM availability(nicked from SA forums):

http://www.skhynix.com/inc/pdfDownlo...hicsMemory.pdf

[watercooled]

That would give some credibility to the Carrizo rumour!

It's no secret AMD were working with Hynix on the development of HBM, after all.

[CAT-THE-FIFTH]

A well balanced opinion piece on the G3258 with some benchmarks:

http://www.eurogamer.net/articles/di...m-g3258-review

However, Crysis 3 - and to a lesser extent Battlefield 4 - demonstrate that the most advanced gaming engines cause problems that can't be fully resolved by sheer clock-speed alone. The future of games development is many-core in nature and we wonder whether a little more investment in a better CPU in the here and now could end up being the better option overall in the medium to long term.

While we had good success in overclocking on the supplied heatsink and fan, you may end up with a chip that requires an aftermarket cooler to get decent clocks - and that additional expense could take you into FX-6300 or Core i3 price territory: two processors perhaps better equipped for the many-core era and potentially better suited for a budget gaming PC. On the flipside, at least the upgrade path available for a PC built on this platform is sizeable: not only can you run this year's Core i5s and i7s, but in theory you should be able to run next year's too.

Finally an English language G3258 article that does not go too OTT with how brilliant the G3258 is. Plus shame on you TR,they actually bothered to test games like Crysis3 properly!

[watercooled]

PA review up at Anandtech.

Unfortunately they seem to have completely negated to include any AMD processors whatsoever in the discrete gaming benchmarks, and they don't include any more expensive processors as a reference in any of the benchmarks. I mean it's understandable they're only comparing cheaper CPUs, but a bit extra can make all the difference in threaded games/applications. And after all, they do like to throw an i7 or two in the charts when they're looking at APUs...

Also, the single-threaded result for the Atom C2750 in Cinebench seems...wrong...
For future reference in case it's corrected - it's now at 309 vs 149 for the i3 4360.

[CAT-THE-FIFTH]

PA review up at Anandtech.

Unfortunately they seem to have completely negated to include any AMD processors whatsoever in the discrete gaming benchmarks, and they don't include any more expensive processors as a reference in any of the benchmarks. I mean it's understandable they're only comparing cheaper CPUs, but a bit extra can make all the difference in threaded games/applications. And after all, they do like to throw an i7 or two in the charts when they're looking at APUs...

Also, the single-threaded result for the Atom C2750 in Cinebench seems...wrong...
For future reference in case it's corrected - it's now at 309 vs 149 for the i3 4360.

Agreed,but I do like the fact they say the "Pentium-AE Is A Processor We Want, But Not The Processor We Need" which is kind of spot on IMHO.

[CAT-THE-FIFTH]

It looks like Carrizo won't use HBM:

http://chinese.vr-zone.com/121197/no...hown-07162014/

[cheesyboy]

To be fair, I wouldn't expect DDR4 to make its AMD debut with mobile chips. IIRC Carrizo was signposted as "DDR3/DDR4" by AMD. May have both memory controllers on the die, like Phenom II.

will be interesting to see more detail on the desktop parts.

[CAT-THE-FIFTH]

The top end Carrizo for desktop is meant to have a 65W TDP.

[cheesyboy]

IIRC Carrizo was signposted as "DDR3/DDR4" by AMD.

In fact, it was the Opteron counterpart to Carrizo, "Toronto", which was roadmapped as DDR3/4;
http://www.extremetech.com/computing...lobalfoundries

[CAT-THE-FIFTH]

Wait - L2 cache is now down to 2MB instead of 4MB??

[scaryjim]

pinch of salt time, I feel. Listen not to the rumour mill. Little of that seems particularly reasonable to me. Particularly the alleged targeting of a BGA platform with Carrizo - I'd be amazed if the mainstream laptop platform wasn't socketed. So this could be a leak on a low-power variant with integrated FCH (I can't see them going for only 2 SATA ports on a mainstream platform either), or just a load of old tosh....

[CAT-THE-FIFTH]

The Carrizo SOC rumours have been around for a while and also the drop in TDP. That screenshot from extremetech,is actually from an AMD video I posted somewhile back(I think some of the websites quietly copied it TBH!).

However,the reduction in L2 cache,I really hope is matched by an increase in cache speed and a reduction in latency.

If not Carrizo will end just a TDP reduced version of Kaveri,which I suppose is good for laptops,but a bit meh for desktops.