AMD - Piledriver chitchat

[DanceswithUnix]

If steamroller is actually on average 30% faster, it will be fast enough that no one but the most die hard of enthusiasts will actually need any more CPU performance for the next few years.

The more I think about it, the less the info we have heard so far adds up on the "Steamroller will be much faster" thing.

In essence they are doubling the issue width, which should improve multi threaded performance. It won't change the single thread performance though where it will go from sharing a decoder with an idle core giving it full use of it, and having it's own decoder giving it full use of it, ie no real change at all.

So I can see it being a monster server chip, and better at many benchmark scores, but the headline change I can't see it helping gamers much which is where is lacks atm. Lets hope the other tweaks they do help in that regard.

[watercooled]

As I said back in post #50, it's also getting something like a μop cache which could help massively with some code. The dedicated decoders are what's mentioned everywhere, but it's from the only significant change. As can be seen in some benchmarks, even first gen Bulldozer is fairly powerful architecture, just it's held back by the front end in some desktop apps, and the initial GF 32nm process issues didn't help.

@Noxvayl: I agree. Unless you're using the CPU for intensive tasks like gaming/encoding/whatever, I imagine you'd struggle to notice the difference in general use between pretty much any modern (Core2 or Athlon64 upwards) CPU. Having repaired a few netbooks, I can say that's not necessarily true for Atom; Bobcat feels fine, but the AMD V105 netbook I used felt a bit sluggish at times, though it could be at least partly due to it using Windows 7 on limited RAM vs XP on the Atoms.

[DanceswithUnix]

Bit surprised they are going back to uop caching. The K6 had the entire L1 instruction cache pre-decoded. The K7 didn't because whilst very trendy at the time it was considered a waste of transistors. I'm guessing this must be a very small cache

[watercooled]

Intel implemented it in SNB, it should really help with misprediction penalty considering the long pipeline.

[Bagnaj97]

This article has appeared over on Phoronix about the "bdver3" compiler support that AMD's added to GCC. It's got a bit of info about scheduler changes which may be of interest.

[CAT-THE-FIFTH]

I just saw this FX4170 against Core i3 3220 article:

http://www.tomshardware.com/reviews/...arks,3314.html

The FX4350 is probably going to quite competitive the Core i3 3220 I suspect,although hopefully power consumption should improve.

[CAT-THE-FIFTH]

OcUK seem to have a Piledriver FX section up on their website:

http://www.overclockers.co.uk/produc...967&subid=1825

[Noxvayl]

Hopefully that means they'll be available soon.

[DanceswithUnix]

Reports that in 3 months they are going to nearly halve the price of the already cheap A4-5300 part to $30.

http://www.fudzilla.com/home/item/29...-some-fm1-apus

I can understand them doing a fire sale on the FM1 parts, but that surprises me.

[Terbinator]

Why in three months time though? Nothing is going to change in those three months other than Intel probably releasing more higher-binned Pentium's.

They'd be best reducing the price now and sniping all those Christmas builds/sales away from Intel - at least you're locking people into FM2 too!

[CAT-THE-FIFTH]

Reports that in 3 months they are going to nearly halve the price of the already cheap A4-5300 part to $30.

http://www.fudzilla.com/home/item/29...-some-fm1-apus

I can understand them doing a fire sale on the FM1 parts, but that surprises me.

The problem is that the A6 5400K is only £10 more currently than an A4-5300. However,$30 with UK VAT would equate to around to around £23 to £25!!

It could be because the IB Pentium and Celerons are being released in Q1 2013. My estimations put the A6 5400K at around G630 level in single threaded applications and around 10% to 20% slower in multi-threaded applications. The A4-5300 is probably around 5% to 10% slower than an A6-5400K,so probably ballpark G530 level performance. However,the IGP is on another level though,so at under £30 the A4-5300 looks like a very good offering.

[scaryjim]

Why in three months time though? Nothing is going to change in those three months other than Intel probably releasing more higher-binned Pentium's.

This price cut is planned to come into effect three months from now, in order to directly compete with Intel Pentium CPUs.

Thing is there're already Ivy Bridge Pentiums - the G2120 that sells for £72, and the G2100T at £62. So the A6 and A4 APUs already undercut those. I think it's more likely that the price cut will be to compete with Ivy Bridge Celerons - I can't see Intel launching a Pentium at < £50, which is where the A6-5400K is already sitting.

EDIT: The current cheapest celeron is the Sandy Bridge based G440, for ~ £30, but it's a single core, single thread design @ 1.6GHz: a £30 A4-5300 would blow it out of the water in every respect.

[CAT-THE-FIFTH]

I wish more review sites actually bothered to test the Intel and AMD CPUs under £50.

[Terbinator]

Thing is there're already Ivy Bridge Pentiums - the G2120 that sells for £72, and the G2100T at £62. So the A6 and A4 APUs already undercut those. I think it's more likely that the price cut will be to compete with Ivy Bridge Celerons - I can't see Intel launching a Pentium at < £50, which is where the A6-5400K is already sitting.

I know, which is why i don't see the logic in waiting. Maybe Fud' messed up and as you've mentioned, perhaps they meant to Celerons instead.

The introduction of Celerons may spur the price-cuts on but there's little difference to the Pentum's anyway bar a loss of some cache a drop in clockspeed.

[scaryjim]

I wish more review sites actually bothered to test the Intel and AMD CPUs under £50.

I think the issue is more that AMD and Intel don't send out review samples, so any site that was going to test them would have to foot the bill themselves, and I can't see there being a great benefit to running benchmarks on low-end software: it's hard to make that kind of review relevant as minute differences in benchmark scores don't really tell you anything - the review would come down to "any of these chips is fine for general web browsing and casual games".

The introduction of Celerons may spur the price-cuts on but there's little difference to the Pentum's anyway bar a loss of some cache a drop in clockspeed.

Actually, the various SB Celerons are quite different: the range includes 2 core 2 thread, 1 core 2 thread, and 1 core 1 thread. TBH the G530 - 2 cores, 2.4GHz, 2MB cache - could be quite an interesting rival to the A4-5300, and is currently £10 less. In fact, all the Celerons on Scan are within a could of £ of each other, so I can't see why anyone would buy the cheaper, single-core versions, unless they had some pressing reason for a 35W TDP (and they didn't actually want their computer to do anything ).

[Terbinator]

I think the issue is more that AMD and Intel don't send out review samples, so any site that was going to test them would have to foot the bill themselves, and I can't see there being a great benefit to running benchmarks on low-end software: it's hard to make that kind of review relevant as minute differences in benchmark scores don't really tell you anything - the review would come down to "any of these chips is fine for general web browsing and casual games".

Lower end products to to produce, dare i say, boring reading too. Obviously its always nice to see what there lower offerings offer up in terms of B4B and possible excursions in low power draw but they don't really set the world-aligth and grab the punters like "8-Cores @ 4GHz" CPU launches.

Actually, the various SB Celerons are quite different: the range includes 2 core 2 thread, 1 core 2 thread, and 1 core 1 thread. TBH the G530 - 2 cores, 2.4GHz, 2MB cache - could be quite an interesting rival to the A4-5300, and is currently £10 less. In fact, all the Celerons on Scan are within a could of £ of each other, so I can't see why anyone would buy the cheaper, single-core versions, unless they had some pressing reason for a 35W TDP (and they didn't actually want their computer to do anything ).

That was the one i had in mind.

I guess the only way i could see the single-core being of any use is in a fileserver or something. I bet the power draw is godly (sub-20W perhaps?).