AMD - Piledriver chitchat

[watercooled]

A10-5800K undervolting article:

http://www.tomshardware.co.uk/a10-58...iew-32536.html

The thing about these underclocking articles, they will either have to disable features like CnQ/Speedstep or set a constant voltage (for reasons I've covered in another thread), depending on what the BIOS allows, but either way can badly harm idle consumption, where an average desktop PC will spend most of its time. It's all well and good if you can save 10w from full load power, but is it worth it if you're doubling idle power?

The problem, is they often compare to a control test where power saving features are also disabled and/or only show load graphs (as is the case here) which is very misleading IMO.

[CAT-THE-FIFTH]

The thing about these underclocking articles, they will either have to disable features like CnQ/Speedstep or set a constant voltage (for reasons I've covered in another thread), depending on what the BIOS allows, but either way can badly harm idle consumption, where an average desktop PC will spend most of its time. It's all well and good if you can save 10w from full load power, but is it worth it if you're doubling idle power?

The problem, is they often compare to a control test where power saving features are also disabled and/or only show load graphs (as is the case here) which is very misleading IMO.

Good point,as most systems will be running at low load.

[CAT-THE-FIFTH]

Regarding the single threaded or lightly threadedbenchmarks,I manage to find some from Toms Hardware.







However,I wonder if the audio encodes are done from a disc drive or from a ISO on a hard drive or SSD??

[scaryjim]

I did like the Tom's review - good analysis of the power consumption issue along with reasonable performance analysis. It looks like Ivy Bridge has a 10% - 30% advantage in lightly and single threaded applications (be interesting to see this revisited with Windows 8 at some point - presumably Piledriver suffers from some of the thread scheduling issues that plagued bulldozer?). Given the vast majority of their test set is the ~ 1 hour spent compiling Chrome, and the Ivy Bridge processors have a 10% lead in that, I'd assume it's a lightly threaded task, so it's possible the A10-5700 might be able to get to a reasonable turbo speed doing that and perform close to the 5800K - in which case it would still have similar performance to the i3s but at a much more comparable power consumption.

I'm definitely very sad that we've not seen an A10-5700 reviews - it really does look to be the pick of the line up (and a much better comparison to the i3s, since it doesn't have the unlocked mulitpliers to aid overclocking...). At £90 it's a shade cheaper than the 3220, and whilst it would undoubtedly lose out in CPU tests, it's going to have so much more graphics Ooomph than the HD2500....

[scaryjim]

The thing about these underclocking articles, they will either have to disable features like CnQ/Speedstep or set a constant voltage.

Not true, I'm afraid - you can undervolt and keep Cool and Quiet turned on, and the motherboard uses a fixed step reduction in voltage which means idle consumption *also* drops when you undervolt. My undervolted Sempron dropped right down to 0.75v at idle...

[watercooled]

But you run a high risk of BSOD while idle doing that, and you're likely to run into that problem before problems with load voltage i.e. idle voltage is usually much closer to threshold than load. I've seen strange problems/BSODs (arguably BSOD is preferable as it's obvious something is wrong then) pop up with tiny negative voltage offsets on both Intel 1155 and FM1 systems. AM2/3 may behave differently, and process node may also have an impact. It's really not easy to test it properly either; stress tests will obviously cause the CPU to come out of idle mode.

[DanceswithUnix]

The main issue appears to be that load power consumption,seems to have not improved over the previous generation,although some reviews say it does,many don't. At least low load power consumption has improved somewhat though.

Would you expect it to?
If it has a TDP of 100W, and is under load, I want the thing to turbo everything it can to get to 100W. I would expect the old one to do the same, so the only change aught to be the new chips should be faster.

As a side note, whatever happened to programmable TDP in the BIOS? Thought that was a great idea.

[CAT-THE-FIFTH]

Would you expect it to?
If it has a TDP of 100W, and is under load, I want the thing to turbo everything it can to get to 100W. I would expect the old one to do the same, so the only change aught to be the new chips should be faster.

As a side note, whatever happened to programmable TDP in the BIOS? Thought that was a great idea.

I am more surprised by the ones showing it consuming 20W more than an A8. Then you get the ones which seem to show it consumes less!!

Perhaps I need to see what software was used to load the CPUs.

[watercooled]

Yeah it's not terribly helpful when you get 'load consumption' without any indication what program was used, or even if it's the same between tests. You can easily get 10-20w difference between different apps, all appearing to load CPU cores to 100%.

IMO it's nice to know what a CPU will use with something like OCCT, but it's not indicative of real world use and isn't terribly important unless you're working with limited PSU headroom for instance.

[scaryjim]

Yeah, the load figures seem to vary massively depending on what load was used: presumably different workloads hammer different parts of the silicon and Trinity is more efficient under some loads than others.

Anyone know if the Turbo threshold is changeable in the BIOS? I wonder how fast you could get it to run on just one thread. 4 steps / 4.2GHz doesn't seem like a lot of turbo to me (given that the 65W versions can do 6 steps/4.0GHz)...

[DanceswithUnix]

Scan finally have parts listed.

http://www.scan.co.uk/shop/computer-...ore-socket-fm2

http://www.scan.co.uk/shop/computer-...socket-fm2-atx

[CAT-THE-FIFTH]

It would be interesting to see whether Windows 8 will improve Trinity performance(or not).

[scaryjim]

Curious, Scan have the 65W parts more expensive than the 100W parts...

[DanceswithUnix]

It would be interesting to see whether Windows 8 will improve Trinity performance(or not).

If Windows 8 is the answer, I'm not sure I like the question
I can see it being another Vista/Me unloved release. Shall stick to Fedora/RedHat for work and Win 7 for gaming and sit that one out.
Some decent Linux benchmarks would be nice, but only Phoronix seem to bother and I tend not to be too impressed with their efforts tbh

[scaryjim]

Maybe you should start up your own Linux review site and see if you can talk AMD into sending you parts to test...

Hmmm, maybe *I* should do that...

[CAT-THE-FIFTH]

If Windows 8 is the answer, I'm not sure I like the question
I can see it being another Vista/Me unloved release. Shall stick to Fedora/RedHat for work and Win 7 for gaming and sit that one out.
Some decent Linux benchmarks would be nice, but only Phoronix seem to bother and I tend not to be too impressed with their efforts tbh

Actually as I mentioned in another thread I started using the 3 month RTM trial on my mini-ITX box I built in this thread:

http://forums.hexus.net/pc-hardware/...-possible.html

With Start8 in place,it helps a lot.

OTH,whether I would change over my main system from Windows 7 is another question.