AMD - Piledriver chitchat

[DanceswithUnix]

I'm not denying it's VLIW likeness BTW, just some articles are claiming so.

Hmm, so if the scheduling etc is done in software, does that mean it literally runs on the cores? I was kind of interpreting the 'software' to be something running on independent microcontrollers, kind of like the dispatch processors we see on GPUs. But reading back through, it seems that's not the case?

The translation/optimization layer needs full access to the resources of the CPU, so getting the CPU to do the work itself makes sense. Kind of like a Java VM will have a thread doing hotspot optimization.

I'm guessing if the cpu execution units are running native ARM opcodes then running code single issue should work without any optimization pass, just a bit slow.

[watercooled]

Funny, I was thinking of the parallels to JVM myself earlier.

[CAT-THE-FIFTH]

TSX disabled on IB and early Broadwell CPUs due to errata:

http://techreport.com/news/26911/err...broadwell-cpus

[shaithis]

At least it didn't make it into the wild like the old FDIV bug...

[kompukare]

TSX disabled on IB and early Broadwell CPUs due to errata:

http://techreport.com/news/26911/err...broadwell-cpus

Surely you mean Haswell?

[watercooled]

I wonder if the extension was in use, or perhaps trial use, on the Haswell E3 Xeons? Besides benchmarks I'm not aware of any software that uses TSX on desktop, but considering it's currently aimed more at the server/workstation market IIRC, the story might be different there.

I was also reading they might use it as a selling point for the later versions of Broadwell (at Anandtech I think) - but, I wonder if they'll start getting refund requests from enterprise companies considering TSX on Haswell was essentially mis-sold, being advertised as a feature then removed/faulty? It probably makes sense to not bear the cost of a complete recall since most people probably won't be using it yet either way.

[Saudrey]

C'est une merveille pour l'partage , les yeux et la découverte

[DanceswithUnix]

Anandtech have a review of a Gigabyte AM1 motherboard, including some benchmarks of games using AM1 processors and a big graphics card

http://www.anandtech.com/show/8370/g...m-s2h-review/5

[scaryjim]

Tomb Raider That's a ridiculous GPU bottleneck!

Shame they mention Mantle in the BF4 blurb but then test with a GTX770 - would've been nice to see DX11 v Mantle figures for a 280X/290. But that kind of puts CPU performance into some perspective - the 7850k puts up a reasonable showing, but the AM1 chips are definitely lacking somewhat - looks like somewhere around an R7 260X would be optimal (which is still pretty impressive, tbh....) but you're probably not going get 60fps out of any GPU with an AM1 mobo (I suspect that the PCIe 2.0 x4 isn't helping.... ).

[DanceswithUnix]

Tomb Raider That's a ridiculous GPU bottleneck!

Almost by definition there has to be a bottleneck *somewhere*

Makes you think though. I only recently stopped using an x2 3800+ as a home server, that is a 9 year old platform that is probably left for dead by one of these little chips. Perhaps the people that wrote Tomb Raider just know how to code

OTOH, what on earth are they benchmarking against an 8 core Atom storage server board that cost more than my FX8350 + mobo?? http://www.scan.co.uk/products/asroc...d-sub-mini-itx

[watercooled]

Perhaps the people that wrote Tomb Raider just know how to code

It could be part of the reason the new consoles chose to go with Jaguar cores over a few Piledriver cores, for instance, and why the game devs seem quite happy with the choices made. And perhaps in part because of the consoles, we'll see more games running like this.

And along with scaryjim I'd like to see a Mantle-DX11 comparison on a CPU like this. Also, while testing with a 770 and max settings is interesting, something more mid-range and perhaps med-high settings might be a more realistic use case.

[CAT-THE-FIFTH]

Saw this on Anandtech forums:

http://www.realworldtech.com/forum/?...rpostid=142759
http://www.realworldtech.com/forum/?...rpostid=142792

[watercooled]

Interesting. About the node comparison though, comparing Intel '14nm' to TSMC '16nm'; does the density advantage of Intel's process (given TSMC are said to be using 20nm BEOL) really directly indicate a performance difference, since TSMC are still using the smaller FETs? Last I was reading, the gate pitch of the transistors is quite similar.

[crazy566]

Ah a 193 page thread about AMD FX what's the common census on the chips lately?

Back when i started it was lots of arguments about them on varying stuff but me personally been running a lovely FX 8150 for about 2-3 years now while as expected i had teething issues which probably would never happened if i updated bios sooner if there actually was a patch then.

I got hit by crappy performance in select games one being Deus Ex: Human Revolution lowest settings no change took me a few months tracked it down to something called Valve-CEG which is a DRM on some games found patch which was a daily fix you run watched game run on 8 cores instead of 1 and frame rate's up drastically so i suggest BIOS updates as they are pretty easy mine was via desktop and it will help your computer run a lot better generally.

On the Piledriver talk anyone know if the fan still has those flat angled blades because i went from nice quite curved blades on Athlon 64 x2 6000+ cooler which was same as my FX one but the latter having flat angled blades that cooler lasted for me about 1-2months i replaced it due to noise vs load and temps compared to old CPU it was crap.

Replaced with a Coolermaster Hyper EVO 212 for most of time the fans not even needed barely hear the computer and 30c day about 52c CPU under load is pretty good.

Wonder what we will see on the 3rd gen desktop models think it will be Bobcat wasn't it? I got excited when those 9xxx ones came out but 2nd gen just higher speeds also the first cost i saw :/ about $1000 for the 5ghz one

[scaryjim]

Makes you think though. I only recently stopped using an x2 3800+ as a home server, that is a 9 year old platform that is probably left for dead by one of these little chips.

I've just rebuilt a 4050e into a music workstation (although tbf I am considering using my Sempy 140 instead). According to Hexus' Athlon 5350 review it roughly matches Barton XP 3200+, which happens to roughly match the 1.8GHz Athlon 64 3000+ (in fact the A64 and Barton are slightly faster - 64s v. 67s for the Athlon 5350). All of which suggests that it couldn't match up to a 2/2.1GHz Athlon X2 in a straight line, but there wouldn't be an awful lot in it.

Out of interest, a 2.4Ghz X2 (the 4800+, which cost $1001 at launch ) nailed PiFast in just under 50s, which isn't good enough to even get close to the A8 7600, which does 36s. Then again, the i5 4670 munches through PiFast in 17s. Now there's some perspective for you....

[watercooled]

TBH this has turned into more of a general CPU/semiconductor discussion thread rather than sticking to the Piledriver topic.

It seems to be a tradition for stock coolers for high-end CPUs to be pants these days; the AMD ones seem to have a lot of noise complaints, while the Intel ones (combined with the poor IHS bonding) are often reported to allow the CPUs to approach or exceed the 90-100C throttle temps. I can't see myself sticking with stock coolers any time soon either, even though I don't overclock.

Bobcat is one of AMD's small, low-power 'cat' cores which is more comparable to Intel's Atom. The newer processors in that range now use the newer Jaguar and Puma+ cores.

I'm really not a fan of the 9000 series CPUs; as you say they're just higher-clocked 8000 series, but they're way beyond the point of diminishing returns on the efficiency curve.

Newer cores, Steamroller, are present on the Kaveri APUs, but have not made it to the FX platform.

Out of interest, a 2.4Ghz X2 (the 4800+, which cost $1001 at launch ) nailed PiFast in just under 50s, which isn't good enough to even get close to the A8 7600, which does 36s. Then again, the i5 4670 munches through PiFast in 17s. Now there's some perspective for you....

TBF, PiFast isn't a brilliant example of a representative benchmark, especially between different microarchitectures.