Silicon Investor (SI) -- The First Internet Community

STOCKTALK

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.

Technology Stocks : Advanced Micro Devices - Moderated (AMD) -- Ignore unavailable to you. Want to Upgrade?

To: eracer who wrote (215234)	10/27/2006 8:49:37 PM
From: pgerassi	Respond to of 275872

Eracer: I was just showing the absurdity of his argument by throwing the same flawed "logic" back at him. You are the one with the flawed logic. You take estimates and treat them like gospel. You also don't go into details on how you calculate or change my method and then claim the changed method was wrong. So the GPUs coming from Intel would be 2.5 times as fast as the ones from AMD according to your calculation ;) It really isn't my calculation. I just borrowed it from Pete. :) No its your method, not mine. t's difficult to say, how fast a K8 would run, if produced in TSMC's foundry process. I'd guesstimate something in the ballpark of 1.6-2GHz for the same TDP if the design would be adapted to the other process. That's a bit lower than mine, but not much. I was thinking TSMC would probably be at 2.2GHz or so for an A64 FX type of CPU with mainstream CPUs at even lower frequencies. Well there I disagree with you strongly. AMD stated that 90nm ssSOI is either faster than plain SOI by 40% or used 1/10th the power at the same clock. That is comparing to their mature 90nm standard unstrained SOI. From their statements between bulk Copper to SOI copper, at 130nm, SOI added either 20% more clock or cut power by 50%. From 180nm, Copper added 20% faster speed versus Aluminum. I don't recall if any power savings were stated. It may not have because power wasn't as big a deal back then as now. 1.2 * 1.2 * 1.4 = 2.016 at roughly the same power. So the 2.8GHz DC 90nm Copper ssSOI A64 FX-64 AM2 or Opteron x220 F would run at 1.4GHz on 90nm Al bulk at the same TDP. Conversly, the 90nm Cu ssSOI at 1.4GHz uses about 15W per the cubed clock rule. That is a bit high probably due to that AMD would use high threshold very low leakage transistors. One reason it wouldn't drop much lower is the need to drive 4 slots of DDR2 and 3 cHT ports and that is fixed no matter how fast the CPU cores are. The 90nm Cu ssSOI FX-64 has 227.4 million transistors on a 230mm2 die and has a 120W TDPmax. The 80nm Al Bulk R600 has 64 pipelines (the same number of total pipelines in the R580 (Radeon X1950XTX)) and 500+ million transistors on a 383mm2 die at an estimated 250+W TDP. The 90nm Al bulk R580 has 384 million transistors on a 352mm2 die topping at 650MHz using 125W. Now I know that without a die plot of the latter, we can not estimate what the ratio of actual core (the pipelines themselves) transistors to outside transistors (cache, bus, memory I/O, PCI-E interface, etc) is. Now you don't like my assumption that the stages have the same number of gate delays. So I will approach it from the ratio above, twice the speed at same power. A R580 uses 384 million transistors at 650MHz at 125W TDP on 90nm Al bulk. It would get 1.3GHz on 90nm Cu ssSOI. To get to 2.8GHz clock, it would use 1250W TDP. Half of a R580 would use 176mm2 of die and use 625W at 2.8GHz with 2.15 times the performance. Dropping it to 1.3GHz to get the same performance would use 62.5W. 1/4 R580 at 1.3GHz would use 88mm2 of die, about the amount used by one K8F core and 512KB of L2, use 32W and have 1/2 R580's performance. This would be equal to a 1x1 89W 90nm Fusion CPU or a 2.6GHz A64 4000+ SC and a Radeon X1600XTX system. Not a bad mainstream gaming rig for today's games. Going down to 65nm which starts shipping soon, if not already, makes that a 103mm2 45W 1x1 Fusion CPU. A 2x2 Fusion based on a QC K8L die would use 283mm2, 89W at 2.6GHz. That would be the equal of a X2 5200+ and a X1950XTX, a 2007 mainstream gaming rig. Extreme gamers could have dual X2 5000+s and X1950XTXs in addition to 4 R600 discrete GPUs. The two internal X1950XTXs do the physics calculations. Going down to 45nm which is when Fusion is released, a 1x1 2.6GHz Fusion CPU would use 60-70mm2 and 25W. It would be decent for 2008 games for those occasional gamers. The 2x2 2.6GHz Fusion would use 150mm2 and 50W. Pushing that to the standard 35W, 65W, 89W and 120W TDPs would get 2.4GHz, 2.8GHz, 3.2GHz and 3.6GHz speeds. That would cover the mainstream gamers. 4x4 2.6GHz Fusion CPUs would use 300mm2 and 89W. Pushing that to 120W gets the clock to 2.9GHz. That would take care of the sporting gamer (more than mainstream, but not extreme). Extreme gamers probably would opt for dual 2x6 3GHz Fusions with 4 discrete dual R620s which needs over 1500W! Going the other way, 1x1 1GHz Fusion CPUs would use 1.4W and be good enough for set top boxes, OLPCs, PICs and palmtops. They would easily supplant the Geode line. Which is AMD's initial target. Pete