Published: Monday 16th October, 2006 | Author: James Morris
Companies: AMD (All AMD content)
We’ve had a stormy relationship with AMD of late. But we’re hardly alone in our concern over the underdog CPU manufacturer’s ability to cope with Intel’s recent reinvigoration. Ever since news started filtering out about Conroe, the AMD fanboys have been deserting their old object of worship faster than it takes to cook an Athlon XP. It was a ‘no-brainer’: Conroe was turning the tables on the Athlon 64, and ‘ass mastering’ it at lower clock speeds – with faster versions already on the way.
There is no denying that Intel’s Core Microarchitecture has the upper hand in virtually every benchmark, and usually by a significant margin. As a result, most hardware enthusiasts have scratched the Athlon 64 FX off their shopping list and written in a Core 2 Duo instead – or even Core 2 Quad. Before jumping on the bandwagon express, though, it’s worth stepping back from the present a bit and putting the current situation in a little more context – both past and future.
Intel has clearly made a huge comeback, and intends to drive home its advantage still further with the Kentsfield quad-core part. From a business perspective, though, the bigger deal is the Xeon version of the Core Microarchitecture. For all the excitement over how much better an Athlon 64 was than a Pentium 4, even more significant was how much better a dual-core Opteron was than a single-core Xeon 800 (for there was no dual-core version until very recently). AMD’s advantage in the workstation and server market was really starting to turn into a whitewash. A few months ago, I tested five £5,000 workstations, and there wasn’t a Xeon in sight – every single one was based on a dual-core Opteron.
You might think that overnight all these manufacturers would be deserting AMD and returning to the new Xeons, which have a similar performance benefit compared to a dual-core Opteron as a Core 2 Duo does over a dual-core Athlon 64. Certainly, if I asked five manufacturers to send me £5,000 workstations now most of them probably would be the new Xeons.
But businesses don’t think quite so short term as this, and they also don’t just think about raw performance. This is particularly true when it comes to mass purchases for corporate use, which has traditionally been an area where Intel has been virtually impregnable. Whatever is happening regarding the performance crown, AMD is making inroads into the business market like never before.
Let’s take Dell for example – one of AMD’s big wins of the last year, and the one everyone is saying looks stupid now Intel is back. As a business customer, you can either buy the Dimension E521 for £499 + VAT (with an AMD Athlon 64 X2 4200+), or the E520 for £50 more (with an Intel Core 2 Duo E6300). They’re both dual-core, and the performance difference is essentially irrelevant to a business customer. But if you’re buying 100 of them, you’d save £5,000 by going for the E521. That’s a fairly easy decision for a financial director to make.
Even in the performance space, you’d be a bit silly to think that AMD’s current position in Intel’s rear-view mirror is for the foreseeable future. Sure, the 30 to 50 per cent lead is going to be hard to overtake. But the difference won’t be anywhere near so obvious this time next year, and the two companies could even be close to parity again. In fact, over the next six months or so AMD looks likely to claw back quite a few per cent already.
AMD has been showing off its 65nm wafers for a few months now, which means the Rev G core is on its way. Even if the DDR2 memory controller which arrived with the Rev F only had a small performance benefit, Rev G has a few more improvements than just the die shrink. The latter will enable higher clock speeds and a lower price, plus allow AMD to compete on an equal playing field to Intel, which has been manufacturing 65nm processors since the Pentium XE 955 at the end of 2005. Spyshots of the new processor revision also show it has some extra Level 2 circuitry, which allegedly includes redesigned branch predictors and prefetchers – one of the areas where the Core Microarchitecture improved over Netburst.
Although AMD processors are far less susceptible than Intel’s Pentium 4 to cache mis-hits, because they have much shorter pipelines, the Core Microarchitecture also has a shorter pipeline. So AMD no longer has this advantage. Reducing cache mis-hits further essentially squeezes more performance out of the processor for free, so it wastes less of its clock cycles. AMD could of course have thrown more cache at the problem, which Intel has also done with its 4MB SmartCache. But in terms of long-term manufacturing costs keeping the cache size smaller and using better prediction circuitry means cheaper processors. With AMD back to the value segment for the time being, this makes a lot of sense.
There is also a lot of confusion over when exactly AMD will be releasing its quad-core parts. Most sources are still saying the middle of 2007, but an AMD representative recently mentioned to me something about Q1, and there are even some rumours of a November 3rd launch. However, the latter could be confused with the launch of 4x4, which appears to be two dual-core Opterons on Socket F rebranded as Athlon 64 FX and placed on a dual-socket motherboard.
Either way, it does look like AMD’s quad-core ‘Barcelona’ CPUs will be with us sooner than a lot of people expected – perhaps because this is a necessity for the company to have a hope of keeping up with Intel’s onslaught. AMD’s penchant for calling its quad-core parts ‘native’ in contrast to the dual-die approach found in Kentsfield seems a little snide. But the quad-core architecture is also likely to claw some of Intel’s lead back. The shared Level 3 cache will further reduce wasted clock cycles and keep the cores running at full pelt, plus all four cores can already access each-other’s Level 2 at core speed, as with the Athlon 64 X2.
The Barcelona will also allegedly include a HyperTransport 2 interface, which will give it a distinct advantage in multi-socket arrangements – like 4x4. Looking towards the end of next year, AMD also has HyperTransport 3 on the roadmap, and a new connection called Torrenza. This will allow peripherals (ATI graphics cards perchance?) to be connected directly to the processor via a bus with much more bandwidth than even PCI Express 2.
So although Intel is on top for now, AMD still has a few cards yet to play. These may not amount to the same kind of winning hand the company had for the last couple of years. But to say AMD is done for, and has failed to capitalise on the Athlon 64’s success, is simply short-sighted. The market moves in phases, and there are bigger gameplans than just having the fastest gaming chip around.
Copyright © 1998 - 2009, HEXUS.net. All rights reserved. Terms, conditions and privacy information.
HEXUS® is a registered trademark of HEXUS Limited.
HEXUS.community :: your right2reply
And the reason I reckon quad-core might be earlier than Q3 is that a pretty senior AMD executive said Q1 to me, just a few weeks ago. I've also had a reseller drop hints about stuff he has he can't talk about, although I think that is Rev G dual-core processors.
Thanks for sharing! I guess there's little to do but wait and see how the cards fall into places over the next 12 months, then. Exiting times, aren't they?Quote
But if you’re buying 100 of them, you’d save £5,000 by going for the E521. That’s a fairly easy decision for a financial director to make.
I liked the article a lot, but I do have one problem; the article assumes that business buyers buying a tranche of 100 machines are going to even consider buying Dell's Dimension range. If they ARE that size of business, they'll be set up for Dell's Premier programme, they'll be buying Optiplexes, and getting them rather below the headline price, and there's no Athlon-based Optis. Businesses will buy them because they get reasonably nice discounts, 3-year on-site cover, and when you call for support, you're talking to a nice person in Eire who's technically competent, rather than a (probably equally nice but woefully undertrained) call-centre worker in the subcontinent.
I've got no problem with A64s in business machines (I've got a dual-Opteron HP server running Exchange, and there ARE a couple of A64 desktop machines in the building), but I think you've got the wrong end of the pricing stick - over the service life of a desktop machine (say, 3 years), £50 is absolutely irrelevant. On the other hand, having a replacement motherboard/hard disk/whatever on site next day, fitted and working without spending three days being bounced around call centres is PRICELESS.Quote
Based on what?
Its the same core as the current FX-62 and I've seen number suggesting the Kentsfield 2.66 can more than double the performance of an FX-62 in some benchmarks. Things don't scale perfectly as you increase the number of cores so even against a pair of 3GHz dual core FX's the Kentsfield should still have the edge.
Kentsfield numbers have been out on Xtremesystems for quite some time but I've seen nothing to suggest the 4x4 can touch it.
Well, the Kentsfield is crippled by the FSB. Two Dual Core's feeding off the FSB. So, they do have an enough L2 cache to offset those bottlenecks.
The thing about AMD's 4x4 is that each Dual Core will have there own dedicated Ram & there own dedicated L2 cache, along with the high speed HTT point to point interconnect.
What I meant to say in my last post was that the Kentsfeild will take some & will lose some, the same goes for the 4x4. But reading about it on AMD's website a few weeks ago, they seem quite confident with there Quad-Father.
Only time will tell, :cool:Quote
But reading about it on AMD's website a few weeks ago, they seem quite confident with there Quad-Father.
Its generally inadvisable to be pessimistic in your advertising material.
Well, the Kentsfield is crippled by the FSB. Two Dual Core's feeding off the FSB. So, they do have an enough L2 cache to offset those bottlenecks.
Scaling seems to be right on the button from what I've seen on XS compared to single and dual Opteron systems, FSB bandwidth doesn't seem to be an issue for Kentsfield. Unless the FX's can achieve more than 100% scaling (they can't, obviously) I can't see them being ahead in anything.
Throw in the added cost of a 2 socket motherboard (and its looking like the choices will be very limited) and having to have at least 2 pairs of DIMM's and its going to be very expensive to go 4x4 regardless of the performance.Quote
Reply