backtop


Print 22 comment(s) - last by Argon18.. on Nov 30 at 1:06 PM

Servers are where thread-friendly new architecture shines the brightest

Today, Advanced Micro Devices, Inc. (AMDannounced the Opteron 6300 Series, code-named Abu Dhabi.  Equipped with a Piledriver core, the new chips serve as an enhanced replacement to the Bulldozer-equipped Opteron 6200 Series (which was code-named Interlagos).

As we mentioned in our piece on the consumer Piledriver launch, the new core is impressive, featuring a number of performance enhancements.  However, consumer workloads tend to be lightly threaded, so the consumer workloads tend to fall short on the price-v-performance scale.

In the server market it's a very different tail.  Workloads here tend to be heavily threaded in many cases, such as virtualized infrastructures, web hosting, mobile device data serving, etc.

AMD's numbers may be a bit biased, but it's claiming to essentially match rival Intel Corp.'s (INTC) performance in HPC (high-performance computing) applications, such as chemical simulations, with a chip that's only half the cost.

AMD performance Opterons

Intel will likely respond with some aggressive price cuts to stay competitive, but for now it's faced with the puzzle of how to compete with a foe that offers twice as many cores at half the cost.

AMD Opteron 6300 Series pricing
Pricing (click to enlarge)

Unlike the consumer market, Intel and AMD are largely competing on the same node -- 32 nm -- for server chips.  This is because Intel has yet to announce E5 series chips based on Ivy Bridge, having only announced E3 series dual- and quad-core offerings.

We spoke with Michael Detwiler, AMD’s Server Product Marketing Manager, who says that AMD's focus it to be "real targeted, instead of trying to be everything to everybody."

He argues that AMD's new Opterons provide "more computations per dollar" than Intel's mixed 32 nm and 22 nm Xeon line.

Piledriver
Piledriver bakes in a number of enhancements (click to enlarge)

The chips feature a number of subtle improvements, including better branch prediction, new instruction set support, and a more efficient cache.  In other words, everything is looking good, although third-party benchmarks weren't available at launch.

Source: AMD



Comments     Threshold


This article is over a month old, voting and posting comments is disabled

RE: Not that hard
By aicom on 11/5/2012 5:45:27 PM , Rating: 2
You're exactly right. The thing is that this higher performing core that Intel needs has already been designed. Haswell finally sees the jump from 3 ALU pipelines (since the Core 2 days) to 4. But lo and behold, 4 ALU pipelines is just the number that AMD has per module (2 per core). For those tasks with more ILP to extract, Haswell will be a monster. With HT, programs can take advantage of that new pipeline even with lower ILP. That's not even counting the performance difference that TSX makes in high core-count systems.


"A politician stumbles over himself... Then they pick it out. They edit it. He runs the clip, and then he makes a funny face, and the whole audience has a Pavlovian response." -- Joe Scarborough on John Stewart over Jim Cramer














botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki