AMD - Piledriver chitchat

[CAT-THE-FIFTH]

The Athlon II X4 845 is tested:
http://www.overclock.net/t/1591886/e...ah-there-is-it

For some reason its extremely fast in SuperPi,matching a 5GHZ Kaveri at 3.8GHZ and in CB 11.5 is the equivalent of a 4.3GHZ Kaveri even though it is running at 3.5GHZ to 3.8GHZ!! The X4 845 only has half the L2 cache of Kaveri too!!

I have a feeling Bristol Ridge will be a decent improvement all round IMHO!!

For around £50 its not too bad it appears!!

[Kumagoro]

Having just had a quick read the link above it would seem my knowledge of overclocking recent (to me) cpus is poor. Changing from AHCI to IDE can get you a higher overclock??? I better make sure I research properly when I upgrade next. Makes me wonder what else I don't know.

It's to do with everything being tied to BCLK for APUs for anyone interested.

[scaryjim]

A lot depends on motherboard. Quite a few FM1 motherboards had dividers for some of the peripheral lines so when the BCLK got up to 125MHz/133MHz/150MHz the peripheral lines would drop back to a reference 100MHz - that meant that an overclock might be unstable at 115MHz BCLK but rock solid at 125MHz or 133MHz. Not sure if that's the case on more recent FM2/FM2+ motherboards though.

Frankly if you're building specifically for an overclock then just get an unlocked CPU and bump the multipliers up - not as much fun, doesn't require as much skill/knowledge, but removes all the uncertainty.

[watercooled]

I wondered what would happen with AMD's embedded stuff and their cat cores: http://semiaccurate.com/2016/02/23/a...-soc-families/

It looks like they're pushing Excavator down into that segment now. I wonder what node these new dies are on though, as it's not mentioned in the article? The dual core Kabini-like chip must be pretty tiny too, even if it's still on 28nm!!

[DanceswithUnix]

It looks like they're pushing Excavator down into that segment now.

It sounded to me from previous articles like there won't be new cat cores once Zen comes out.

[watercooled]

Same, hence what I wondered what they'd do about the low power segment with no 'small' cores. Seems they're just extending the reach of the big cores.

[DanceswithUnix]

Same, hence what I wondered what they'd do about the low power segment with no 'small' cores. Seems they're just extending the reach of the big cores.

I guess with leakage so improved by finfets you don't get much lower power using by small cores, so just undervolting and down clocking Zen they can get very low power. Won't be as cheap to make as a small core, but then AMD aren't targetting phones & tablets where every penny and tiny packaging counts, they usually end up in big laptops.

[CAT-THE-FIFTH]

This is quite interesting:

https://translate.google.co.uk/trans...%2F&edit-text=

The demo is just over 4000 bytes,and seems to be mostly compute based - AMD cards do very well in it!!

[CAT-THE-FIFTH]

Dresdenboy has managed to elucidate some more information about the Zen core:

http://dresdenboy.blogspot.co.uk/201...s-emerged.html

Two front ends and one shared back end??

[CAT-THE-FIFTH]

http://wccftech.com/amd-radeon-r7-35...card-launched/

The R7 350 is a rebranded HD7750 or HD7770. However,one of the cards is literally a single slot,bus powered HD7770.

[scaryjim]

... one of the cards is literally a single slot,bus powered HD7770.

Indeed, very tasty. Be interesting to see if anyone brings those over the the EU: sadly with a $90 RRP and current exchange rates/tax they'd be likely to cost at least £80 here, which is a bit close to faster cards from both AMD and nvidia to be viable. If we had better exchange rates so they could come over at nearer £60 they'd be very interesting indeed...

[CAT-THE-FIFTH]

More X4 845 benches:

http://imgur.com/a/DTsNz

CB R15 seems fine unlike the earlier one!

[watercooled]

The single threaded performance of Stoney Ridge (Excavator-based low-TDP solution) looks pretty impressive! http://wccftech.com/amds-stoney-ridg...ning-detailed/

[scaryjim]

The single threaded performance of Stoney Ridge (Excavator-based low-TDP solution) looks pretty impressive! http://wccftech.com/amds-stoney-ridg...ning-detailed/

OK, haven't even read the article yet (I'll do so and come back with comments on the tech, no doubt) but I'm already annoyed - at WCCFTech themselves:

Starting today each and everyone of you will have the chance to have your own stories featured on WCCF. All you need to do is submit your article via email to tip@wccftech.com.

How generous. They'll let you write content for them without paying you anything for it! Go WCCFTech, how progressive of you! That kind of thing really gets on my nerves.


SECOND EDIT:

Now read the article, and I can't help but notice some similarities between these engineering samples and something else we've been talking about recently: http://forums.hexus.net/cpus/354386-...ml#post3611837

1 module, 3CUs, single channel memory sounds like a mid-way harvest between the I and J series embedded APUs - and it's possibly the first time that AMD have launched an embedded product that has then been harvested into the consumer line up?

[CAT-THE-FIFTH]

It seems the X4 845 in one of the benchmarks(7-Zip IIRC) was around 10% clock for clock faster than an X4 860K,and is similar clock to clock in the ST Cinebench 11.5 benchmark to a Phenom II X4 980BE. Considering it has less L2 cache than either an X4 860 or a Phenom II X4 980 that is not too bad.

If Bristol Ridge does have improved Excavator cores and has 4MB L2 cache,that together with the DDR4 could make it have the highest IPC of any AMD core to date IMHO!! Considering that AMD only have had one new process node since 32NM and that is a bulk process,in terms of performance and performance/watt they have done some decent improvements over the last 4 to 5 years.

Plus,considering that the AMD CPUs are still 28NM and Intel is two nodes ahead,its amazing they manage to even compete in anyway.

Basically go from Trinity/Richland,there has been a near 20% improvement in IPC and a decent reduction in power consumption too.

[watercooled]

OK, haven't even read the article yet (I'll do so and come back with comments on the tech, no doubt) but I'm already annoyed - at WCCFTech themselves:



How generous. They'll let you write content for them without paying you anything for it! Go WCCFTech, how progressive of you! That kind of thing really gets on my nerves.


SECOND EDIT:

Now read the article, and I can't help but notice some similarities between these engineering samples and something else we've been talking about recently: http://forums.hexus.net/cpus/354386-...ml#post3611837

1 module, 3CUs, single channel memory sounds like a mid-way harvest between the I and J series embedded APUs - and it's possibly the first time that AMD have launched an embedded product that has then been harvested into the consumer line up?

According to Semiaccurate, these new parts are actually new dies rather than harvests, as harvests apparently wouldn't allow them to scale cost and power down as well.