Testing Different Memory Speeds on AMD's A8-3850 Llano APU

A place to give your thoughts on our reviews!

Testing Different Memory Speeds on AMD's A8-3850 Llano APU

Postby Apoptosis » Wed Jul 13, 2011 5:42 pm

Testing Different Memory Speeds on AMD's A8-3850 Llano APU

Legit Reviews has had a number of readers write in saying that they want to see performance numbers of AMD's A-Series APUs on the A75 chipset at 1333MHz, 1600MHz and 1866MHz. When we did our launch article on the A8-3850 we ran 1866Mhz memory, but since this platform is very budget friendly we know our readers also want to see performance numbers at 1333MHz and 1600MHz. Read on to see how 'Llano' performs at three different memory clock speeds and also with 16GB of memory in a 4x4GB configuration!

Image

We found that the memory bandwidth ranges from 13GB/s to 16GB/s with typical dual channel memory kits on our AMD A8-3850 APU and Gigabyte A75-UD4H motherboard. This difference doesn't sound that great, but a 3GB/s increase when dealing with 13GB is a very nice 23% performance increase. Most of this performance gain is seen when moving up from a 1333Mhz to a 1600MHz memory kit is where you'll see the largest performance jump. For example 2x4GB 1333MHz CL8 memory kit was able to hit ~12.7GB/s memory bandwidth, but the 2x4GB 1600MHz CL8 memory kit has ~15.0GB/s of memory bandwidth, which is a very nice 18% improvement over 1333MHz


Article Title: Testing Different Memory Speeds on AMD's A8-3850 Llano APU
Article URL: http://legitreviews.com/article/1652/1/
Pricing At Time of Print: $69.99 shipped
Find us on Facebook to discover the faces behind the names!
Follow Me on Twitter!
User avatar
Apoptosis
Site Admin
Site Admin
 
Posts: 33057
Joined: Sun Oct 05, 2003 8:45 pm
Location: St. Louis, Missouri

Re: Testing Different Memory Speeds on AMD's A8-3850 Llano A

Postby Major_A » Wed Jul 13, 2011 7:50 pm

Now I have a question, since the MC is embedded in the Intel processors do they get similar results?

I know your laptop did with faster RAM, and that was an Intel based test.
User avatar
Major_A
Legit Extremist
Legit Extremist
 
Posts: 3793
Joined: Tue May 15, 2007 2:11 pm
Location: Houston, TX

Re: Testing Different Memory Speeds on AMD's A8-3850 Llano A

Postby Tator Tot » Thu Jul 14, 2011 6:27 am

Major_A wrote:Now I have a question, since the MC is embedded in the Intel processors do they get similar results?

I know your laptop did with faster RAM, and that was an Intel based test.


With the Llano platform, you'll see bigger jumps in performance from the boosted memory speeds (which is why AMD worked so hard to improve the IMC/Northbridge, but not the CPU Core logic.)
Though, some of the biggest gains I saw were in games themselves. With 30% increases in speed from 1333Mhz C9 to 1866Mhz C9; just depending on the game.
The biggest reason for that is that the AMD APU's graphics portion shares the same buffer and RAM with the CPU portion and as you can tell from most GPU's; their default bandwidth is way above the average of 12Gb/s you find on CPU's.
Thus, that's where the bottleneck is.

On Intel systems, you won't get any GPU related performance boosts from the increased RAM speed by ~5% due to the Graphics Portion of the chip already being limited in a number of other ways. (IE: Clock speed and cores)
User avatar
Tator Tot
Legit Fanatic
Legit Fanatic
 
Posts: 128
Joined: Sat May 29, 2010 7:31 am
Location: Ellisville, MO; US

Re: Testing Different Memory Speeds on AMD's A8-3850 Llano A

Postby Apoptosis » Thu Jul 14, 2011 8:46 am

Major_A wrote:Now I have a question, since the MC is embedded in the Intel processors do they get similar results?

I know your laptop did with faster RAM, and that was an Intel based test.



I still have the Intel Core i3-2120 in the test system and this might not be the best CPU to use as it has the slower Intel HD Graphics 2000 GPU inside...

stalker-2120-memory.jpg
stalker-2120-memory.jpg (36.13 KiB) Viewed 2426 times


Very minimal performance gains. I'd expect to see a slightly larger performance difference on Intel HD Graphics 3000 GPUs, but no where near the ~20% gain seen on AMD processors. The AMD A8-3850 APU that I used for testing had 400 GPU cores and is takes up the bulk of the silicon space on the processor, so it's plumbed to be better on GPU apps and obviously needs bandwidth to utilize all the GPU cores.
Find us on Facebook to discover the faces behind the names!
Follow Me on Twitter!
User avatar
Apoptosis
Site Admin
Site Admin
 
Posts: 33057
Joined: Sun Oct 05, 2003 8:45 pm
Location: St. Louis, Missouri

Re: Testing Different Memory Speeds on AMD's A8-3850 Llano A

Postby Major_A » Thu Jul 14, 2011 11:16 am

I said it in the initial A8-3850 review. It is nice to see faster RAM translate to real world gains. Thanks for taking the time to test the i3.

Thanks Mr. Tot for explaining the situation to me.
User avatar
Major_A
Legit Extremist
Legit Extremist
 
Posts: 3793
Joined: Tue May 15, 2007 2:11 pm
Location: Houston, TX

Re: Testing Different Memory Speeds on AMD's A8-3850 Llano A

Postby Apoptosis » Fri Jul 15, 2011 9:01 am

It took a few days to get a response from AMD about the better performance with four modules versus two and I thought I'd share it with our readers:

Hey Nate,

“Why is there a small performance uplift using 4x DIMMs vs. 2x DIMMs when each config is utilizing both channels of the MC?”

Here’s the feedback I got from the engineers. But first a reminder… disclaimer…

1 DIMM-per-channel is the spec for 1866 support. That won’t keep someone from plugging DIMMs in all slots… maybe it will work right… maybe it won’t.

Here’s the hypothesis I got back from an engineer:

I suspect there are a couple of reasons for the small uplift with 2 DIMMs/channel. Most of the memory benchmark programs look at memory-to-core throughput. Part of memory feeds the graphics engine and they probably do not account for that. With 2 DIMMs, the graphics memory may be stuck on top of memory which puts it in the second pair of DIMMs while the memory benchmark may be running in the lower 8 GB space (first DIMM pair). Without the sharing, there will be less paging and less competition for the bottlenecked area. It’s not a 100% giveback as the same physical data wire are used, but overlap can happen there as well when one DIMM is doing a RAS cycle and the other DIMM pumps out data. The data they have shows a 5% delta, which would easily will fit into scenarios like this.

If they want to further test memory-to-CPU performance, they could disable internal graphics and run with a discrete card. That will eliminate any graphics memory access overlap. Otherwise things such as screen resolution will impact their scores.



So, 4x DIMMs enables a better chance of workload fan-out across both channels (over-simplified). Not far off from what your initial gut was, Nathan.


Just wanted to share the official response from an AMD engineer on the CPU design team and also the PR department.
Find us on Facebook to discover the faces behind the names!
Follow Me on Twitter!
User avatar
Apoptosis
Site Admin
Site Admin
 
Posts: 33057
Joined: Sun Oct 05, 2003 8:45 pm
Location: St. Louis, Missouri

Re: Testing Different Memory Speeds on AMD's A8-3850 Llano A

Postby gwolfman » Mon Jul 25, 2011 1:11 pm

Thanks for the review.
gwolfman
Legit Extremist
Legit Extremist
 
Posts: 662
Joined: Fri Apr 25, 2008 10:57 am
Location: USA


Return to Legit Reviews Review Forum

Who is online

Users browsing this forum: No registered users and 0 guests