AMD - Piledriver chitchat

[DanceswithUnix]

Vishera info leak at e-tailer...

http://semiaccurate.com/2012/10/11/a...p-at-etailers/

16MB cache? So 8MB L3 then.

[watercooled]

Sounds about right. Really interested to see how much of a difference the L3 makes in various apps.

I wonder if we'll see a μop cache or similar in Steamroller to take the sting out of the branch mis-prediction penalty?
Edit: Just remembered about the 'μop queue' after re-skimming Steamroller arch info.

[scaryjim]

FX4300 looks curious - 3.8GHz Quad core with only 4MB L3 cache. Given it has the same architecture and clockspeed as an A10-5800K that L3 cache better give it some hella IPC increase or it looks utterly pointless - not fast enough to be an upgrade over a Phenom II X4, not cheap enough to be a genuinely budget option, I just can't see who'd buy one.

[Terbinator]

Bulldozer had 16MB total anyway IIRC?

[watercooled]

Yeah but the APUs have no L3.

[DanceswithUnix]

FX4300 looks curious - 3.8GHz Quad core with only 4MB L3 cache. Given it has the same architecture and clockspeed as an A10-5800K that L3 cache better give it some hella IPC increase or it looks utterly pointless - not fast enough to be an upgrade over a Phenom II X4, not cheap enough to be a genuinely budget option, I just can't see who'd buy one.

If the idle power consumption for the platform is good then actually I could be tempted.
Am considering the Trinity part for my home server rebuild, but an AM3+ motherboard would be nicer I think.

[watercooled]

The FM1/2 platform is more efficient than AM3+ in general if it's an important factor. IMO if you're going for best idle power and if you can do without the L3 cache, Trinity may be a better route, or possibly one of the Trinity Athlons (I think I read they exist). But it wouldn't hurt to wait, if possible; hopefully we'll be surprised!

[DanceswithUnix]

The FM1/2 platform is more efficient than AM3+ in general if it's an important factor. IMO if you're going for best idle power and if you can do without the L3 cache, Trinity may be a better route, or possibly one of the Trinity Athlons (I think I read they exist). But it wouldn't hurt to wait, if possible; hopefully we'll be surprised!

Well for now I am running with my HTPC re-purposed, so no great problem there.

The energy requirement isn't that important tbh, but less heat is less fan noise and I am always trying to drive that down a bit. The motherboard/cpu that failed was an old AM2 2.9GHz dual core I think, so anything modern is an improvement on that

I'm expecting the box to run routing across a couple of network connections and a couple of virtual machines (also running Linux). For me, 40 shaders in a 760g chipset and L3 cache on the CPU makes more sense than Trinity.

[badass]

I've had a rummage through the Anandtech CPU bench. I compared a Q6600 to an A10-5800K.
The A10 is substantially faster in all of the CPU tests, single and multi threaded.
The only reason I went from a Q6600 to a 3570k was maximum memory capacity, virtualisation and fast video transcoding. My Q6600 was plenty fast enough for everything else, including games.
TBH, I thought it was about the same as the Q6600.

[CAT-THE-FIFTH]

According to Anandtech bench,the A10 5800K is faster overall than a Q9550:

http://www.anandtech.com/bench/Product/675?vs=50

[DanceswithUnix]

Teaser on overclocking potential on water.

http://www.fudzilla.com/home/item/29...-hits-5%20-ghz

Doesn't sound much better than the old 8150? Too busy this morning to go look up what that could do

[scaryjim]

Legitreviews got an 8150 to 4.9GHz with a Corsair H100, which is a better AIO water cooler than the bundled AMD one the fudzilla article mentions.

http://www.legitreviews.com/article/1741/19/

If you can get more than 5GHz on a 120mm AIO water cooler, it'll be a little better than Bulldozer (although not much, given the higher base clock). I guess the real test will be power draw when overclocked, though: iirc Bulldozer's power draw went through the roof when you started overclocking.

[CAT-THE-FIFTH]

The AMD water cooler is basically an Antec Kuhler 920. Actually the FX8150 water cooler bundle is not too badly price:

http://www.scan.co.uk/products/amd-f...il-20-cashback

The water cooler itself is around £73:

http://www.scan.co.uk/products/antec...m2plus-am3plus

[Terbinator]

I guess the real test will be power draw when overclocked, though: iirc Bulldozer's power draw went through the roof when you started overclocking.

Correct. It's probably part of the reason no one could seriously recommend it for a game build. The chip isn't that bad comparatively to a 2500K but the amount you'd have to clock it up to consistently match/beat it just isn't attainable.

[badass]

According to Anandtech bench,the A10 5800K is faster overall than a Q9550:

http://www.anandtech.com/bench/Product/675?vs=50

Indeed. It looks like the i5-750 is faster overall but the difference isn't enough that it means anything in real world usage.

It just goes to show how relative this "lack" of CPU performance on Piledriver is. 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.

[Noxvayl]

@ badass

I find little real world difference between my old Q9550 and my current i5 3570K, benchmarks show a large gap between them but you wouldn't be able to tell the difference without the benchmarks.