facebook rss twitter

Review: AMD A10-5800K Trinity: why it needs faster RAM

by Tarinder Sandhu on 3 October 2012, 08:46

Tags: AMD (NYSE:AMD)

Quick Link: HEXUS.net/qabm7z

Add to My Vault: x

Fast RAM = good graphics

AMD's latest all-in-one processor is the A10-5800K APU. It combines four CPU cores and a 384-shader GPU on to one piece of silicon. This second-generation A-series APU builds on the foundations laid down by last year's A8, yet now offering up to 40 per cent more graphics performance.

In common with all chips that house integrated graphics on the same die as the CPU portion, A10-5800K pulls the necessary GPU memory bandwidth from the system's. This memory bandwidth is a fundamental determinant of overall graphics performance, as the best-in-class GPU execution cores thrive when fed with faster RAM.

AMD states that all A10-series chips are capable of running DDR3-1,866 memory right off the bat, but chances are that premium motherboards and RAM will enable higher speeds to be reached. We've previously demonstrated that scaling system memory from DDR3-1,600 to DDR3-2,400 does little to improve overall performance on an Intel Core i7 system, whose GPU is relatively anaemic compared to the Radeon HD 7660D of the A10-5800K, so it will be interesting to see the just how the APU's performance is affected by changes in memory speed alone.

We've therefore tested the A10-5800K's CPU and, more importantly, GPU performance scaling by simply changing system memory frequency. This has been done by replicating the speeds and feeds of popular packs available from respected manufacturers. Here's a brief table outlining the memory speeds, latencies and cost.

Memory CAS latency tRCD tRP tRAS Command rate Price for 8GB DC pack
DDR3-1,333 9 9 9 24 2T £25
DDR3-1,600 9 9 9 24 2T £25
DDR3-1,866 9 10 9 27 2T £40
DDR3-2,133 11 12 12 30 2T £45

Providing full disclosure, an 8GB pack of Patriot Viper 3 memory was used to replicate the timings and speed of the modules above. We managed to boot into Windows with 2,400MHz-rated memory by using a pack of GeIL Veloce and G.Skill TridentX RAM. However, the test Gigabyte F2-A85X-UP4 motherboard, updated to the latest F3e BIOS, would only run the super-fast memory at this speed in single-channel mode. No amount of tweaking or choice expletives remedied the problem, and we're in discussions with AMD over this issue. Note that AMD does not guarantee any memory frequency over 1,866MHz.

So, in summary, what you have is the A10-5800K run with memory speeds ranging from dual-channel DDR3-1,333 to DDR3-2,133. We've focussed on gaming benchmarks because that's where the changes in memory speed will be most keenly felt.

Testing methodology

System Setup

Processor
AMD A10-5800K @ 3.8GHz
Cores/threads
4/4
Cooler
AVC reference
Motherboard
Gigabyte F2-A85X-UP4 (F3e)
Chipset driver
Catalyst 12.8 SB
Graphics
AMD Radeon HD 7660D
Graphics driver
Catalyst 12.8
Memory
Patriot 8GB IEM DDR3 (2 x 4GB)
Memory speed
9-9-9-24-2T @ 1,333MHz
9-9-9-24-2T @ 1,600MHz
9-10-9-27-2T @ 1,866MHz
11-12-12-30-2T @ 2,133MHz
Sound card
Onboard high-definition audio
Primary storage
256GB Samsung 830 SSD
PSU
Corsair AX750
Chassis
Corsair Graphite 600T
Operating System
Windows 7 Ultimate SP1 64-bit

Benchmarks

AIDA v2.60 Tests system bandwidth, cache bandwidth, and latency
HEXUS PiFast A single-threaded number-crunching test we've been using for over 10 years now
3DMark 11 Run at Entry preset
3DMark Vantage Run at Performance preset
DiRT: Showdown 1,280x720, 4x MSAA, Medium settings, DX9
Batman: Arkham City 1,280x720, 2x MSAA, High-quality settings, DX9 and Medium-quality, DX11

Notes

One last thing; for the graphics tests we've also run the fastest memory (DDR3-2,133) with the GPU overclocked from the standard 800MHz to 1,000MHz. The purpose of doing this is to see how much extra scaling the APU's graphics provide once both the core and memory frequency are raised.