Phenom II Earns a Financially Troubled AMD Less per Chip than Core 2 Quad

The global economy isn’t exactly strong right now. People are still buying, just not nearly as much as before when it felt like money grew on trees. Financially, all companies have been hurt, but AMD has much bigger issues. The table below shows net income in millions of US dollars before taxes for AMD and Intel over the past four quarters:

Net Income Before Taxes Q1 2009 Q4 2008 Q3 2008 Q2 2008
AMD -$298 Million -$1,358 Million $22 Million -$682 Million
Intel $629 Million $369 Million $2,833 Million $2,313 Million

 

Yeah. Ouch. Granted Intel going from ~$2.8B of income in a quarter down to under $400M must’ve hurt, but AMD has lost over $2.3B in the past four quarters. The company isn’t profitable and unfortunately is stuck between a rock and a hard place when it comes to fixing that.

The financial issues extend beyond simple CPU sales but I must point out the obvious issue with AMD’s current strategy. We know that the Phenom II is competitive, but look at what it’s competing against:

Processor L1 Cache L2 Cache L3 Cache Total Cache (4-core) Transistor Count Die Size
AMD Phenom II 128K per core 512KB per core 6MB 8.5MB 758M 258mm2
Intel Core 2 Quad Q9x50 64KB per core 12MB N/A 12.25MB 820M 214mm2
Intel Core 2 Quad Q9x00 64KB per core 6MB N/A 6.25MB 456M 164mm2
Intel Core 2 Quad Q8xxx 64KB per core 4MB N/A 4.25MB 456M 164mm2

 

Every single Phenom II uses a single 258 mm2 45nm die, that’s nearly Nehalem-sized. The problem is that the Phenom II parts generally compete against Intel’s Core 2 Quad Q9x00 and Q8xxx series, both of which have a total die size of 164mm2. AMD’s Phenom II die is 57% larger.

AMD and Intel both manufacture on 300mm wafers, but Intel can get nearly 60% more CPUs for each wafer than AMD can thanks to its die size advantage. That translates into more revenue per wafer and a significant profit advantage for Intel.

AMD’s Phenom II is very competitive, but the strategy does not have much long term staying power. AMD needs to introduce smaller die versions of its CPUs soon.

The deeper ramifications of AMD’s current situation are troubling. I’m not sure what impact all of this is having on the development of AMD’s next-generation architectures, but I suspect that it can’t be good.

The Test

Motherboard: Intel DX48BT2 (Intel X48)
MSI DKA790GX Platinum (AMD 790GX)
Chipset: Intel X48
AMD 790GX
Chipset Drivers: Intel 9.1.1.1010 (Intel)
AMD Catalyst 8.12
Hard Disk: Intel X25-M SSD (80GB)
Memory: G.Skill DDR2-800 2 x 2GB (4-4-4-12)
G.Skill DDR2-1066 2 x 2GB (5-5-5-15)
Qimonda DDR3-1066 4 x 1GB (7-7-7-20)
Video Card: eVGA GeForce GTX 280
Video Drivers: NVIDIA ForceWare 180.43 (Vista64)
NVIDIA ForceWare 178.24 (Vista32)
Desktop Resolution: 1920 x 1200
OS: Windows Vista Ultimate 32-bit (for SYSMark)
Windows Vista Ultimate 64-bit
Core 2 Quad Q9400 vs. Q8400: An Extra $30 buys you 6% SYSMark 2007 Performance
Comments Locked

60 Comments

View All Comments

  • TA152H - Friday, May 8, 2009 - link

    Anand,

    Have you ever asked Intel why they don't lower the latency of their processors when they cut the cache size? Why should a processor with two MB cache run at the same latency as one with six MB? Certainly it does not need to.

    I could understand it where they are just cutting off some of the cache because it's faulty, but when they are actually two different dies, with the different cache sizes designed into the chip, why do they artificially slow down the chips with the smaller cache? They should have no trouble lopping off one cycle, since the 4 MB Conroes were 14 instead of 15, and these processors are 2 MB cache per core, so it could allow 13 cycles, but surely can handle 14 easily.

    It's maddening when Intel slows things down for no good reason. It's probably a marketing decision, and we all know how marketing decisions are.
  • Anand Lal Shimpi - Friday, May 8, 2009 - link

    It actually comes down to design resources. It's fairly easy to change a cache size, but changing the access latency requires more of a design change. I'm guessing those resources are better spent on newer architectures. e.g. Intel could go back and make even better versions of the Penryn based cores, but it makes more sense to put those efforts into engineering Westmere and its successors.

    Plus, a given architecture is usually optimized for whatever latency cache it's originally designed with. Speeding up the L2 may not yield as big of a gain as it would had the architecture been originally designed around a faster L2.

    I believe there's always a focus on lowering cache latencies, and that's what we saw with Nehalem. From what I've heard, the new focus is bringing down that L3 latency...

    Take care,
    Anand
  • Seramics - Friday, May 8, 2009 - link

    I noticed a strange trend in Phenom II X3 720 performance. They consistently performed very well, always outperforming similarly clocked buy quad core model of PII X4 920 and sometimes even besting PII X4 940. Strange... wonder why is that... the only advantage of 720 over quad models is higher L3 cache per core... but still, i would thought even 920 should be better.... strange strange
  • cfaalm - Friday, May 8, 2009 - link

    Why are we getting new Core2 models in the first place? Shouldn't Intel be selling us i5 rather sooner than later?
  • garydale - Friday, May 8, 2009 - link

    In almost every benchmark the AMD chip bested the (cheaper) Intel - often by a noticeable margin. The real competition for the 940 remains the 9400. A better comparison would have been between the 8400 and one of AMD's comparably-priced CPUs.

    Still, with VT disabled on the 8400, we're talking apples and oranges. I don't know what Intel were thinking by disabling it but seems remarkably silly with virtualisation even hitting home users these days. It's quite like disabling SSE but the idea seems like a win for marketing over engineering. Pay the extra $30 and get a CPU.
  • ssj4Gogeta - Saturday, May 9, 2009 - link

    I can't understand what all this fuss is about. As if everyone in the world is going to be running the Ultimate version of Windows.
  • LoneWolf15 - Friday, May 8, 2009 - link

    Intel VT as well.

    Sorry, I think my Q6600 is worth more than a Q8400 for that reason alone. I'm also pretty sure that AMD's hardware virtualization trickles down to a cheaper level than Intel's does.

    Intel really needs to stop cutting VT on all but perhaps Celeron-class CPU's and maybe Pentium Dual-Core on laptops. I certainly wouldn't buy anybody's quad-core that didn't have it.
  • leexgx - Friday, May 8, 2009 - link

    all AM2 or higher chips have amd-V (apart from semprons None of them have amd-V), i think some 939 cpus have amd-V not sure but not the point realy as thay not been sold for years

    all q8000 cpus do not have VT (unless thay bring one out that does), q9000 do
  • Scali - Friday, May 8, 2009 - link

    Yup, AMD leaves VT enabled on pretty much everything.
    Makes sense, because it's an added value over Intel's offerings, and it doesn't cost much extra, since it's already in the chip design.

    For Intel it makes sense to leave it disabled, because it's a feature that's mainly useful for business users, and Intel has always tried to push business users to the high-end CPUs by disabling certain features on the lower-end models. Performance alone isn't really a reason to get a high-end CPU anymore.
  • LoneWolf15 - Friday, May 8, 2009 - link

    It will be useful for far more than business users shortly. Windows 7's new XP-mode option requires Intel VT or AMD-V. While I admit that so far I haven't had app issues under Win7RC (that includes 32-bit app issues in a 64-bit environment) there IS one 16-bit app I still run that I'm sure I'll need XP mode for under 64-bit Windows. I'm also sure there will be other instances that pertain to the general market.

    Eventually, hypervisors are going to be a big deal, just like multi-core processors are becoming now. At that point, people without Intel VT will be screwed, and I think they'll be pretty ticked if they're the ones that thought that buying a quad-core CPU like the Q8400 was a good future-proofing move.

Log in

Don't have an account? Sign up now