AMD has two counters to Conroe
Depending on which mole you listen to
By Charlie Demerjian: Saturday 27 May 2006, 16:55
<A TARGET="_blank" HREF="
http://ad.uk.doubleclick.net/click%3Bh=v6|33f4|3|0|%2a|i%3B33221141%3B0-0%3B1%3B13228625%3B31-1|1%3B16375917|16393812|1%3B%3B%7Esscs%3D%3f
http://www.via.com.tw/en/C7M/"><IMG SRC="
http://ad.uk.doubleclick.net/1194645/1X1PIXEL.GIF" BORDER=0></A>
WE ARE TOLD AMD has a response to the Intel MCW cores, but until a little bit ago, no one would say what. A third mole has now surfaced - it must be mating season - to hint at what it is.
The male moles sing in lovelorn voices that they are all looking forward to what Intel would call a quad-core part from the green team. The females sing back from the other side of the hedge that the males are full of it, and AMD is going to be coming out with a hyper-clocked FX. Mating season makes them crazy.
So, of the two, I lean towards the semi-quad core, it seems like a much more feasible stopgap - ask Intel, it is really good at stopgaps, and this is their solution of choice. The interesting part is how they will be connected, a can of worms that has both up and down sides compared to the Intel FSB approach. Intel just dumps the extra cores on the bus, and puts on a pained smile when you mention bus speed. We hear 1333 is still a stretch goal for Kentsfield and Clovertown, good luck there guys. Either way, 2x2 is easy enough to do, if you are prepared to pay the price.
Enter AMD. It has a bigger problem, as it uses a point-to-point bus. If it does things the Intel way, you end up with one core slaved off the other, and an extra hop across HT to get to the remote core. This has some pretty dire effects on latency, especially on 4S systems. The other approach is to have each die with one HT-out connected to the socket, but this again has latency problems, and potential routing issues.
Of the two, they both hammer memory latency, halve memory bandwidth, and generally make things slower for all cores. Not happy. I would expect the one die slaved off the other to be the way AMD does things in the end, it seems like the lesser of half a dozen evils.
The other counter we hear is a short run of a hyper-clocked FX part, some say 90nm some say 65. Some say AMD can clock it up and it is cherry picking, others are more conspiracy-oriented and say the firm burning wafers to get a few uber-parts. Yet more say it will be a special run of large-cache, high-clock parts.
Lets just rule out the big cache ones right now: AMD may really want to retain the gaming crown, but the cost of new masks to do a large cache variant is way out of line for a PR stunt. It would most likely be cheaper to buy all the hardware sites out there off, and have them bash Intel - masks can run into seven digits each, not a PR level expenditure. OK, maybe it is for VIIV, but it needs it.
That leaves us with a clocked-to-the-moon FX part, something the firm might just be able to do. If you look at the SE parts AMD is supplying Sun, they are a bin up over normal parts, and have a 35W higher thermal envelope. 35W is quite a lot to play with, if AMD wants to push the edge, unlike what they are doing now, you can get at least a bin, maybe two out of that.
The FX-62 parts seem to be taking not much more than 100W, so with a slight voltage bump, you could get to 3.0GHz with 3.2 being a stretch if you cherry pick. That would put an FX-62++ up against a 2.97/1066 XE, will it be enough? Will Lassie get home in time to tell mommy that Timmy fell down he well?
Either way, the war goes on, in 1S, Intel put the stake in the ground, time for AMD to react. A few years ago, when we announced the P4EE for Intel, I had a meeting with AMD a few hours later. I asked the assempled reps what they thought of the new Intel gaming chip. The person sitting in the chair looked up and with a really unconcerned look on his face said 'So?' The next three or so years pointed out that he was correct.
Now I am getting the same attitude from Intel. One person there, when asked about the new, hypothetical AMD parts responded "If this is true - then we expect them to react to Core, but this seems a bit like a Hail Mary. We hear there will only be hundreds to a couple of thousand of chips total. We’re welcoming the other millions of gamers to the ultimate processor for these ultimate PC buyers: Conroe." Strangely, the same attitude on the other side of the fence.
Whatever AMD does, it sure will be fun to watch. It is not going to take MCW lying down. Will it be enough? Will there be volume? Will mommy get to the well in time to save Timmy? Tune in next week, if the launch is happening. We hear it is happening soon. µ