AMD Confirms 32nm Yield Issues at Global Foundries

37 Comments

View All Comments

• Penti - Thursday, September 29, 2011 - link

  They don't really and they are packaged in AMD's facility in Malaysia, Silicon can either be from Dresden or Singapore for AMD CPUs. The old Chartered AMD-process fab still isn't on the latest process though. Packaging mean that they are cut, from the wafer that is, assembled on the cpu-package and tested in Malaysia and therefor ships from there. Which is of course closer to the computer manufacturer (or their contractors/EMS's and ODMs any way).

• BSMonitor - Thursday, September 29, 2011 - link

  We need something to go right for AMD.
  
  Already, Intel is holding back performance. 65W and 95W Sandy Bridge only (vs AMD 125W Phenoms). Where are the Socket 1155 -125W parts?
  
  Oh right, not needed. Intel could probably ship an air cooled 4+ GHz base SNB now. But they'd only be competing against themselves.

• silverblue - Thursday, September 29, 2011 - link

  Yes, they probably would be.
  
  In any case, we've been over this again and again - AMD TDP != Intel TDP. What's more, AMD's higher end products are still 45nm, and in comparison with 32nm second generation fabrication, of course Intel is going to bring out more frugal processors. That probably won't change even with Bulldozer. That said, Llano hasn't done too badly on that front.

• JarredWalton - Thursday, September 29, 2011 - link

  Generally speaking, at least up until Llano and Brazos, AMD's real power numbers tended to be worse than Intel's. More importantly, even though people say the TDP measurements aren't the same, they're close enough for practical purposes. TDP is a spec for the manufacturers of laptops, desktops, heatsinks, etc. more than anything, so when a company says "max TDP = 95W", that will be pretty close to the most you can pull from just the CPU. What is not the same is AMD's ACP and Intel's TDP, but if I'm not mistaken AMD generally quotes ACP for Opterons and not for other parts.
  
  Even if the two companies did measure the "same" TDP, it's important to remember that TDP (Thermal Design Power) isn't the same as power use. Depending on the load, power use could be quite a bit lower, but it should never be higher for a "thermally significant" portion of time. TDP is generally a worst-case result running a worst-case processor under a worst-case load. So when AMD says that the Phenom II 1100T, 1090T, 1075T, 1055T, 980, 975, 970, 965, and 955 are all "125W TDP", variations in the voltage requirements, clock speeds, CPU batch, etc. all play a role. The same goes for 95W, 65W, etc. CPUs from both AMD and Intel, naturally.
  
  That means when we measure real-world power use, the above gives us charts like this (and if we grabbed 20 different CPUs of each type, we'd probably see a spread of at least 10%).
  
  http://www.anandtech.com/bench/CPU/51
  http://www.anandtech.com/bench/CPU/52
  http://www.anandtech.com/bench/CPU/63
  http://www.anandtech.com/bench/CPU/64

• silverblue - Thursday, September 29, 2011 - link

  Very happy to concede the point. What doesn't help though, as I'm sure you'll agree, is when AMD releases two very different parts with the same TDP, such as the A8-3800 and the A4-3300. Unless I'm missing a trick (very happy to be educated here), the A4 is bound to consume noticably less power than the A8 even if they've switched the bias back towards the CPU cores, and even if there's leakage through any deactivated cores (I presume the A4 has two of those along with a crippled GPU). So, AMD's TDPs aren't exactly reliable here (and when you consider that the 3850 is a 100W part compared to the 65W 3800...). I'd love to see a review concerning both parts so I can be sure of how much each part actually consumes.
  
  Incidentally, the A4-3300 is only $12 more (per 1K units) than the new Atom D2700. Well... it certainly made me smile.

• JarredWalton - Thursday, September 29, 2011 - link

  What actually can keep TDP for dissimilar parts relatively close is that the parts are still binned. So, your i3-2100 and i5-2500S are both 65W parts, and both are dual-core. The i5-2500S has Hyper-Threading as well, and more cache enabled. The i3-2100 runs at a max of 3.1GHz, but with no Turbo Boost, where the i5-2500S runs at 2.7GHz but can Turbo up to 3.7GHz. So which actually uses more power?
  
  Well, some i3-2100 chips will run at a higher voltage, and perhaps the wafer/die wasn't as prime, so it ends up being potentially more leaky or whatever. I'd still expect most i3-2100 to use less power than the i5-2500S, but given the latter is a "lower power" option, it might actually win out. It's something that we really can't state with certainty unless we actually have a chance to test the parts in question. That's all from the Intel side, but we could say the same for many AMD chips.
  
  TDP ends up going in steps because manufacturers target specific limits. We see this especially in laptops, where everything is usually netbook (<10W), low power (15~25W), 35W, 45W, or in a few (rare) instances 55W. Most of the 35W parts are meant to be interchangeable, but a low-end dual-core 35W part like a Pentium B940 should typically use less than a high-end dual-core 35W part like an i7-2620M. Actually, I've noticed that the i5-2410M usually gets better battery life than the i7-2620M as a perhaps better example, and while I've never had a chance to test the exact same laptop with both chips, I still think the i7-2620M tends to draw closer to TDP under full load than the i5-2410M.

• silverblue - Friday, September 30, 2011 - link

  Understood, however I thought the i5-2500S was a quad core desktop CPU?
  
  Not trolling, you understand, just interested that's all.

• BSMonitor - Thursday, September 29, 2011 - link

  "In any case, we've been over this again and again - AMD TDP != Intel TDP. What's more, AMD's higher end products are still 45nm, and in comparison with 32nm second generation fabrication, of course Intel is going to bring out more frugal processors. That probably won't change even with Bulldozer. That said, Llano hasn't done too badly on that front."
  
  Hmmm.
  
  Phenom II x6 - Binned 125W - 157 Total System Watts
  Core i7 980X - Binned 130W - 170 Total System Watts
  
  AMD A8 - 3850 - Binned 100W - 123 Total System Watts
  Core i7 2600K - Binned 95W - 128 Total System Watts
  
  Joe Plumber, they are not the same between the two, you are correct. However, the typical 125W-130W level for each company is their Highest Performance Level.
  
  The 95W-100W is the Mainstream Performance Level.
  
  The 65W Power Savey / Efficient Performance Level.
  
  They are comparable between the two TDP calcs.. Thus my point. A 95W Mainstream Intel Part is doing MUCH more work than a Performance Level AMD Part. Look at the numbers! Regardless of the TDP calculation, you cannot argue that AMD's TDP is soo drastically different than Intel's that the 2600K should be compared to a Phenom II x6 in terms of power usage/requirements??
  
  Therefore, my point that Intel has not even released a SNB part at the power level of a Phenom II x6 is accurate.
  
  The data is in the BENCH tab, I am sure you know how to use it.

• silverblue - Friday, September 30, 2011 - link

  I do know how to use it... no need to get snotty. :)
  
  In any case, you just compared a 32nm 6C/12T CPU with a 45nm 6C/6T CPU. The AMD CPU wins but most likely only because of its relative lack of L3 cache (even if Intel's caching is very nice indeed). AMD simply cannot release a better CPU just yet so this is the best they have on offer. That 980X can feed its cores better and has better IPC off the bat.
  
  Comparing the A8-3850 to the 2600K was a bit odd on the surface - one has a complex GPU, the other doesn't as much - however as this is the x264 test, your point stands. Wouldn't mind seeing the CPU load graphs for that test - does the 2600K even break a sweat?
  
  I'm not debating the fact that Intel haven't released a 125/130W SB (even though they could), and you're probably correct in that they don't feel the need to, but again, the X6 is an old product on an old production node with lesser performance. A 32nm X6 would use a good deal less power but performance wouldn't be anywhere close to the 2600K even with Llano's minor core enhancements (2 generations old tech though, remember), so Intel doesn't need an Extreme Edition here. When we saw the X6 announced, we were a lot more enthusiastic about it than after it released - some software simply hates it. Personally, I'd love to see a 125W SB/SB-E just to see how damned powerful it can be and how much it improves over the 990X, but the price would scare most of us. ;)

• BSMonitor - Monday, October 3, 2011 - link

  As I explained, comparing target thermal product lines.
  
  The 980X was the highest 130W part on Anand's list. The x6 1100 the highest 125W part on AMD's side. FYI, the 980 is an old model too.
  
  The A8 is AMD's newest/greatest 32nm mainstream CPU. The 2600K is Intel's.
  
  We aren't talking prices here. We are talking thermals and their targets.