The latest news out of Fudzilla has some information on the new line of CPUs from Intel (NASDAQ: INTC), code-named Haswell. Haswell is supposed to be the chip that once and for all buries AMD (NYSE: AMD) under an avalanche of better graphics performance that will negate the only advantage Chipzilla's struggling rival has left in the market. For more than a year it's looked, thanks to the rumor mill and Intel's PR staff, that the harder AMD tried to regain relevance by advancing its APUs, the closer it would be flying into 'coffin corner' versus the inevitability of Intel's superior everything else. And Haswell's improved graphics performance would be the cause for AMD to finally stall and spiral to its death. But it is looking more and more that Haswell may be the chip that finally reveals just how far off course Intel is in its designs and causes Intel, and not AMD, to fall precipitously.
The latest news release, courtesy of Fudzilla, has the first Haswell part coming out in June and it's a 2GHz Core i7 running at 35W TDP, the Core i7-4765T. Now the prevailing wisdom is that Haswell will see instructions per clock improve by at most 15%, but more likely 10%, itself a solid achievement - you are free to disagree - given that CPU speed was not Intel's over-arching design goal here; power management and graphics performance were. The 4765T will be married to the GT2 (not GT3) HD4600 GPU which can turbo up to 1.2GHz. Okay, sounds good right? But so what?
There's an Ivy Bridge SKU now that runs at 2.2GHz and 35W TDP with the HD4000, the 3632QM. So, in effect the IPC increase is offset by the lower speed at the same thermal envelope. It doesn't matter that one is a notebook product (3632QM) and one is a desktop product (4765T). 35W TDP is 35W TDP. This is a real red flag to me and these engineering sample benchmarks support this, which show even worse performance than this theoretic comparison. So, the CPU side of Haswell may not be anything to write home about.
Surviving Contact with the Price Point
Haswell as blueprinted was going to be everything to all people and, when one looked at everything Intel had planned, it was hard to argue with that conclusion. But the problem with blueprints is that eventually it has to meet with the reality of the cost of production. So, over the course of development in the past 16 months, initial features of Haswell got lopped off, most notably Crystalwell - the massive 8 GB integrated memory block to feed the new GPUs - which is where a great deal of the performance increase was going to come from. But, at this point, Crystalwell's cost is causing OEMs to scream. Even if Haswell with Crystalwell makes it into actual systems, at what price point will they be selling?
They certainly will not be in mid-end laptops and ultrathins where price matters most. No OEM can afford to build one without handing all the profit to Intel. And people wonder why Ultrabook sales have been sluggish and estimates for 2013 are being cut by more than 30%?
Haswell won't have any issues continuing to win the CPU performance race, but that's a race that AMD and ARM (NASDAQ: ARMH) are not playing. Slap a discrete graphics card in your desktop and go if you want a fast CPU. Even better, save yourself $100 and buy an Ivy Bridge to get more than enough performance. Heck, buy a Trinity desktop SKU and add a discrete card and save even more money. Everything that makes Haswell not Ivy Bridge is in its ability to perform similarly with lower power requirements with better integrated graphics. Where Haswell has to be evaluated is in its performance at the power usage and price points where the competition is playing because there is no future in high end laptops that sport any other logo than an Apple.
Enter the Fat Dragon
AMD's Richland APUs, which will be launched on March 19th, are unabashedly process refreshes of the Trinity line with a number of small improvements including support for PC-2133 DDR3 memory. They will sport graphics improvements in mobile SKUs of 20-40% over Trinity, if AMD's numbers are to be believed. This is a refresh, folks, not an entirely new generation on a mature 22nm process - no GCN, no 28nm, no Steamroller core -- just a better 32mn part with power management algorithm tweaks and higher clock speeds per watt TDP.
And while Haswell will undoubtedly be better than Ivy Bridge, which really couldn't hold its own with Trinity, there's a good chance it won't be good enough overall at a price point to be competitive with even Richland, forget trying to compete with Kabini. As I said in my article about it last week, Kabini and Temash will likely redefine the low end mobile x86 market with the complete failure of Clover Trail Atom. Given the hype surrounding Haswell, it didn't look possible, but the closer we get to the parts Intel is actually going to ship the more I'm convinced Intel doesn't know what it's doing.
Intel showing us a 10W Haswell part is one thing. Intel selling a Haswell part OEMs want to buy at a price they can sell the computer built around it is another. The battery performance on Haswell has to be lights out better than Richland to begin to justify Intel's pricing. I'm not talking an hour here; I'm talking 50% or more. It's possible but I'm not convinced. Apple (NASDAQ: AAPL) will pay for Intel's best parts because their margins are high enough to absorb their cost but everyone else is trying to figure out how to compete with tablets at a lower price point than an iPad.
For Whom the Bell Tolls
In a nutshell Haswell is beginning to look DOA to me as a product. Trinity's CPU compares favorably to a Core i5 and is significantly cheaper. Richland will ship at current Trinity prices and include better power management, maybe not as good as Haswell's but when it comes to power and graphics Intel has to show me. AMD still has to show me they can publish a road map and hit it. So far with Kabini and Richland they are doing so. Without Crystalwell Haswell's GPU will not come close to Richland's performance so on price and usable performance it is very possible that Richland takes Haswell to the OEM cleaners. And all Intel can do is strong arm them not to do so like they did with Ultrabooks, but since they are mostly a failure how much leverage will that threat still carry?
I still think Richland will be squeezed hard by Intel and Kabini is where AMD's real opportunity lies, but if we see a bunch of design wins for Richland in Q2 the stall indicator will be sounding for Intel and the beginning of an epic fall from a great height could ensue.