I was wondering if I could use a 4870 card with a llano 3850 for dual graphics power.

You could run one monitor off of the integrated graphics and a different one off of the discrete card.  That could make some sense if your goal is to run four monitors at once, with a game only running on one of them.  If you want to use asynchronous crossfire, then no, you can't.  Even if AMD allowed you to try it, you'd get worse performance from it than from the 4870 alone, as the performance gap is too big.  And that's when asynchronous crossfire works, which often it doesn't.

I did some research on this, because I was a bit curious.

Llano AMD Dual Graphics, according to AMD:

*Only pairs a Llano on-die graphics with "Select" discrete GPU chipsets (and it's a short list)
*Only works with DX10/11
*And since it only works with DX10/11, it only works on Vista/Win7
*Only works in Full Screen mode (not any sort of Windowed mode)

That being said, there are a few desktop cards that will match up for "Dual Graphics". The 4870 is not on any list.

http://www.amd.com/us/products/technologies/dual-graphics/pages/dual-graphics.aspx#1

Some performance numbers. Summed up: not really worth while.

http://www.tomshardware.com/reviews/amd-a8-3850-llano,2975-2.html

Otherwise, the two graphics adapters don't really play well together - depending on your BIOS, it could be a matter of one or the other.

Well it looks like I don't have to worry about buying a 6xxx series card along with the APU, but I think I'll probably get the APU anyways to save on some power and maybe sell my 4870.

I don't play games at high resolutions... pretty much play everything at 1360x768. I think I can take the the performance hit to save some money on the electricity bill. I don't know about the rest of the specs but from the specs I can find I'm losing 100mhz gpu clock speed and 400 shader processors if I replace the 4870 with the 3850, which I guess is actually pretty significant.

I think my current gpu overperforms for what resolution I play at right now... pretty much everything I've been playing is capped at 60vsync or higher without vsync. I don't need lot's of flash in my graphics, medium settings are fine.

I guess Anandtechs benchmarks are seriously wrong cause they were showing much better performance for these APUs. Although I'm not sure if they were using a seperate gpu instead of the integrated one. It seems they were.

I'm kinda not sure about it... I'd like to save some money on the electricity but 100 * 400 is a lot of processing power, and with games becoming more and more tasking, I'm not sure how long the APU would work out.

i think the equvalent in there is a 6550 if i am right then yes you could run in hybrid crossfire!

Originally posted by jusomdude

I don't play games at high resolutions... pretty much play everything at 1360x768. I think I can take the the performance hit to save some money on the electricity bill. I don't know about the rest of the specs but from the specs I can find I'm losing 100mhz gpu clock speed and 400 shader processors if I replace the 4870 with the 3850, which I guess is actually pretty significant.

More important is that you lose something like 80% of the memory bandwidth.  Llano is greatly constrained by memory bandwidth, because it has to feed both a GPU and a CPU.  Some review sites paired a Llano APU with 1333 MHz memory or some such, and that cripples the graphics in it.  Anandtech tested different memory speeds with it and found that it makes a big difference, so that might be what you were seeing.

If you're going to play games on a Llano system, then you might want to consider springing for 1866 MHz memory like this:

http://www.newegg.com/Product/Product.aspx?Item=N82E16820231455

For just the processor, it doesn't matter, but for feeding the graphics, it sure does.

If the goal is to save electricity, then the Radeon HD 4870 is one of the worst cards you could get.  The problem is idle power consumption, not load.  GDDR5 memory was still very new (to the degree that AMD wasn't sure if they'd be able to launch the card at all, if they could get the GDDR5 chips for it), and AMD hadn't yet figured out how to clock down the memory at idle.  So it has idle power consumption of something obscene like 90 W.  Newer cards with comparable performance can bring that down to perhaps 10-20 W.

For your electricity bill, idle power consumption is usually what matters, not load, because the system is nearly always idle (i.e., not playing graphically intensive games).

If I downclock my gpu clocks without messing with the voltage when I'm not playing games will that save electricity?

BTW do you have an idea on how many clocks of GPU memory equals the clocks of desktop memory when paired with the APU?

like if the gpu was 100mhz, how many mhz would the desktop memory need to perform similarly?

The GPU and memory clocks will generally reduce as far as they can without causing problems.  The problem with the 4870 was that it launched before AMD had figured out how to reduce the memory clock speeds without causing crashing, artifacting, and various other problems.

If you mean for a Llano system, then it will automatically do a much better job of reducing clock speeds at idle than you could manually do yourself.

Right now I have my gpu at 500mhz and the memory at 600mhz. I kept the clocks at the same ratio as the default clocks and havn't noticed any problems.

I noticed that with GPUZ, it shows the core cconstantly fluctuate between 500 and the default clock when it's not down clocked

like it will be 500 one second then 750 the next, and just keeps fluctuating every second.

Yeah, AMD was able to reduce the core clock.  But the problem is the memory clock, as GDDR5 was still very new.

The issue is that even when a system is idle at the desktop, a GPU still has to do some work to display the desktop.  Furthermore, the ratio of which pieces have to do how much work can be very different from a game.  Even if you have 50 times the shader power you need to display the desktop, that doesn't mean that you have 50 times the power of everything else.

There's also a problem that AMD has to make sure the clock speeds are high enough to work for everyone, including the relatively more active loads it counts as idle.  If the clock speeds work for 99% of people, and cause crashing for 1%, then the 1% complain endlessly all over the Internet that their card is defective or their drivers are defective or some such, and AMD gets a bad reputation for making defective products.

It's not strictly active/idle, either as it has some intermediate clock speeds available.  But it has to make sure that no matter what you do, it keeps the clock speeds high enough not to cause problems.

One of the things I've found interesting is that CPU to GPU data memory transfers are faster with the APU than any discrete GPU using opencl. I do some programming as a hobby and am currently working on a software renderer that doesn't even use shaders so this seems like it could be useful to me.

That's because it doesn't have to actually transfer information anywhere.  It just has to point the CPU or GPU to information already sitting there in memory.

They say there's a performance hit for the process on the current chips though, I wonder how big it is.

Well I whent ahead and ordered the A8-3850 with the memory you linked. Should have it in about two weeks since I had to pay through an online bill pay service.

Hoping I can get 1/3 of the money back from selling my 4870.

It's a downgrade for the graphics paired with the typical game but an upgrade for the cpu. I want to see what I can do with the APU and opencl though.

Oh and it looks like the A8-3850 is second in place for price vs performance on cpubenchmark.net for high end CPUs.

Just checked some reviews, testing different memory speeds with the APU and it looks like 1866Mhz memory provides almost no performance improvement over 1600Mhz memory for gaming. Although the 1600Mhz has an average of 20% better performance over 1333Mhz memory.

I can still change my order, but not sure if I will to save $20 for cheaper memory.

Anyways, just a heads up for anyone looking at the APUs.

http://www.anandtech.com/show/4476/amd-a83850-review/4

A 5% boost in gaming performance is pretty significant.  Nearly every game they tested and at nearly every resolution showed improvements.

A lot depends on how the testing is done.  A lot of reviews either didn't test a memory bandwidth difference at all, or only had some cursory testing as an afterthought.  That's the only one I saw that really dug into it.

If something is CPU-bound, then adding more memory bandwidth won't help much.  But gaming video cards use more expensive GDDR5 memory rather than cheaper DDR3 for a reason.  If the GPU in Llano isn't meaningfully restricted by memory bandwidth even at 1866 MHz DDR3, then AMD and Nvidia are wasting an awful lot of money adding more memory channels than necessary to their high end cards.

IDK, the average of 3 more fps doesn't seem like that big of a deal to me personally. $20 for 3 fps just doesn't really seem worth it.

---

Originally posted by jusomdude
IDK, the average of 3 more fps doesn't seem like that big of a deal to me personally. $20 for 3 fps just doesn't really seem worth it.

---

That depends on how you look at it.

5% improvement for no or marginal price increase isn't a bad deal really. 5% in and of itself isn't much, but if it's nearly free then hey, why not.

Also, if that 5% or 3FPS is the difference between 27 and 30 FPS (or 57 and 60 FPS) - then it does make a marked difference in game play. And in these lower performance systems, you know they are going to be constrained, it's just a matter of what settings you turn on and off to make the acceptable tradeoff between performance and visual quality: it's not like higher end systems where you are pretty well going to get 30+ FPS no matter how many options you turn on.

Originally posted by Ridelynn

 

---

Originally posted by jusomdude
IDK, the average of 3 more fps doesn't seem like that big of a deal to me personally. $20 for 3 fps just doesn't really seem worth it.

---

 

That depends on how you look at it.

5% improvement for no or marginal price increase isn't a bad deal really. 5% in and of itself isn't much, but if it's nearly free then hey, why not.

Also, if that 5% or 3FPS is the difference between 27 and 30 FPS (or 57 and 60 FPS) - then it does make a marked difference in game play. And in these lower performance systems, you know they are going to be constrained, it's just a matter of what settings you turn on and off to make the acceptable tradeoff between performance and visual quality: it's not like higher end systems where you are pretty well going to get 30+ FPS no matter how many options you turn on.

Well, it's a subjective matter. Personally I'd rather have an extra $20 over 3 fps.

Mutiply by 20 and you can lose 100% of your performance and save $400.  Or back in the real world, you could save more than $400 by not having a computer at all.

It really depends some on whether you think you'll stay with the integrated graphics or buy a discrete card.  Also, if the goal is to save power, you might want to look into undervolting the chip, as Llano does that pretty well.