1. Home >
  2. Computing

Once more unto the breach: AMD's Trinity takes on Intel's Ivy Bridge

Today marks the official launch of AMD's Trinity APU on the desktop, and hopefully heralds the beginning of a brighter future for the company's products.
By Joel Hruska
Trinity-Feature

Today marks the official launch of AMD's Trinity APU on the desktop, and hopefully heralds the beginning of a brighter future for the company's products. The past 18 months have been ugly; Bobcat's strong debut was marred by AMD's decision to fire then-CEO Dirk Meyer, and Llano sales were undercut by manufacturing difficulties at GlobalFoundries. Bulldozer missed its performance and thermal targets, Sunnyvale was forced to cancel its Krishna/Wichita follow-up to Brazos, and a number of important engineers/executives were either laid off or left.

Through the midst of this turmoil, the company's remaining engineers have persevered in unenviable conditions. Trinity AMD's first desktop part to fuse Bulldozer's architecture and a new GPU. The mobile version met with reasonably good reviews, but AMD still sells millions of desktop CPUs a year. Can the higher power envelope of the desktop form factor help Trinity hit higher performance targets?

Evolutionary adaptations, competitive pricing

We've discussed both Bulldozer and Trinity at significant length over the past year; I'd recommend you consult our various stories if you need a refresher on either architecture. For our purposes today, it's enough to understand that Bulldozer had significant problems that AMD couldn't immediately fix. GlobalFoundries was already ramping Trinity production when BD launched 12 months ago, and that means AMD had only a limited window in which to improve the architecture.

Piledriver improvements

Incremental performance improvements, in other words, are the order of the day. With Trinity, AMD focused on solving the thermal issues that kept the chip from reaching high clock speeds and on improving IPC when and where it could. Here's the company's new desktop lineup:

AMD APU lineup

The 5800K is $122, the 5600K is $101, and the dual-core 5400K is just $67. At those prices, AMD has set itself up to compete mostly against low-end, last-generation Sandy Bridge parts with low-end, Intel HD 2000 graphics. Intel's Pentium G2120 is AMD's $100 competition (Intel's chip is based on Ivy Bridge and clocked at 3.1GHz with a 3MB L3 cache and no Hyper-Threading. The Core i3-2220 is AMD's stiffest challenge. Intel launched the chip just a few weeks ago; it's a 22nm Ivy Bridge dual-core at 3.3GHz with Intel HD 2500 graphics and Hyper-Threading support.

The best thing about AMD's price structure for this launch is that the company appears to have learned from its mistakes. One of the reasons Bulldozer had such a rough launch was AMD's insistence on keeping the chip priced well above the six-core, 45nm Thuban core it was supposed to replace. Despite its age, the older chip offered better overall performance, and left the FX-8150 sitting in no-man's land.

At $122, the A10-5800K avoids this problem. The chip's pricing is no accident; at $122, the A10-5800K is going to match graphical blades with Intel's cut-back HD 2500 graphics, not the full HD 4000 solution. The lower-end processors will face off against a melange of Pentium-brand Sandy Bridge hardware, which rely on the older HD 2000 graphics. That said, there are already a few IVB-based Pentium chips and Intel may reinforce this product divisions if it feels the need.

Next page: Cheap overclocking, integrated graphics...

Trinity retrenches around cheap overclocking, integrated graphics

When AMD was slugging it out with Intel in the late 1990s, the company chose to extend the life of the Socket 7 platform, even though Intel had abandoned it for the Pentium II's Slot 1. AMD's K6 and K6-2 processors lacked the floating-point muscle to keep up with the Pentium II, but compensated with much lower prices and broad compatibility (you could plug a K6/K6-2 into a lot of motherboards).

For enthusiasts on a tight budget, a K6-2, Thermaltake Golden Orb, and a bit of overclocking was a great way to boost the performance of a system if you couldn't afford to replace the original motherboard. We suspect that Trinity's highly competitive pricing is designed to echo this arrangement by providing a "good enough" x86 CPU combined with absolute best-in-class graphics relative to what Intel sells at the same price point.

Trinity upgrade path

Pushing a 4.2GHz A10-5800K to 4.6 or 4.8GHz isn't going to alter the balance of power between AMD and Intel, but the fun of tinkering with overclocking on the cheap could appeal to certain buyers.  AMD is also getting back to its roots as a company that kept platforms around for more than a generation or two. Socket FM1 (Llano) was a one-shot affair, and there were concerns that the company would take a similar route with Trinity and Socket FM2.

Trinity's GPU-CPU connection has evolved from what Llano offered but still resembles a conventional northbridge architecture. Piledriver incorporates shared virtual addressing and incorporates bi-directional power management. This means that the CPU can down-clock to give more headroom to the GPU and vice-versa; it's a useful mechanism for expanding performance flexibility and it's one that Intel has had in place for awhile now.

Trinity-UNB

Trinity doesn't incorporate anything like Intel's shared ring bus, and the GPU's performance remains sensitive to total RAM bandwidth.

 

Performance data (or the lack thereof)

Our original plan with Trinity was to focus on the chip's graphics performance, particularly when paired with a low-end discrete Radeon. The price difference between the A10-5800K and chips like Intel's Core i5-3550 allows for this sort of budgeting, and we wanted to see if the Trinity + Radeon combination would be faster than the i5-3550 with the same card. Our plans had to be adjusted at the last moment, when our motherboard -- an MSI FM2-A85XA-G65 -- caught fire.

If you've never experienced the joys of unexpectedly molten electronics, some of the finer points of the experience may be lost on you. Allow me to help fill the gap:
  • The volume of smoke produced at the per-component level is truly astonishing.
  • Burning silicon smells terrible. Fry a motherboard or CPU for 30 seconds and you'll be wishing for a nice Styrofoam-and-tire fire for the next six months. Startling skunks just outside an open window may become a way of life as the epithelium cells inside your nostrils beg for relief, at any price.
  • If the fire starts while you are in the bathroom, it may be proof that a divine being is nursing a serious grudge.
Dead boardThe capacitors in question are normally covered by a heatsink.

In eleven years of hardware testing, this is only the third time I've seen a system spontaneously ignite. I confirmed that all components, including the CPU cooler, were properly secured and functional. There were no abnormal temperatures, BIOS warnings, or voltage fluctuations prior to the burnout and the system's PSU remains fully functional. The system was plugged into a surge suppressor, which in turn was hooked to a UPS, and neither unit tripped or reported any problems.

This torched our chance of having much in the way of independent review coverage. Replacement hardware is on the way, and rather than toss up a few uneven skimpy graphs, we've decided to hold back on publishing until we've got a more robust set of data to publish.

Tagged In

GlobalFoundries Graphics AMD Processors Review

More from Computing

Subscribe Today to get the latest ExtremeTech news delivered right to your inbox.
This newsletter may contain advertising, deals, or affiliate links. Subscribing to a newsletter indicates your consent to our Terms of use(Opens in a new window) and Privacy Policy. You may unsubscribe from the newsletter at any time.
Thanks for Signing Up