Leaked Slide Claims To Spill Details On Next-Gen Bulldozer Processor
A slide leaked to occasionally-accurate Turkish website Donahim Haber purports to showcase some of the features of AMD's next-generation Bulldozer core, codenamed 'Piledriver.' This is the same CPU core the company's upcoming 'Trinity' mobile APU will be based on. Other leaked slides have implied that Trinity will be released more quickly than first anticipated, which makes this information downright timely.
As always, keep in mind that Donahim Haber is a rumor site. The slide, however, is a rumor that makes sense. A 10 percent performance jump from Bulldozer to Piledriver is a reasonable increase for a quick architectural refresh. The chip keeps BD's cache layouts and supports version 3.0 of AMD's Turbo Core!. At a guess, this may be a suite of more-advanced power gating and clocking functionalities--it may be that the primary features of TC 3.0 are aimed at Trinity.
The chip also supports new instructions. FMA3 is an Intel-supported instruction for fused multiply-adds. The current state of FMA3/FMA4 (AMD's version) support is a bit confused, but Bulldozer won't ship with FMA3 support. Adding it to Piledriver could improve performance in certain workloads, once compilers are updated to take advantage of the new capability. This last is essential, and it's part of why Bulldozer's performance may improve over time--several of the chip's features may require compiler support to function optimally. New BMI (bit manipulation instructions) could improve the performance of certain loops or conditions.
The one thing we'd have liked to see that isn't mentioned here is any difference in power consumption between the two chips. Oftentimes, companies are able to optimize die layouts from one generation to the next as a means of reducing power, even on the same process. In this case, power savings translate into higher available clock speeds, and that's an option that could only benefit AMD. The company will be taking its eight-core (and higher) products into battle in the server sector next year, and lower TDPs would give AMD room to compete with current Xeons by increasing clock speeds more aggressively.
Other options are as expected. Piledriver will continue to support DDR3-1866 via dual-channel memory (you'll have to buy a server platform if you want quad-channel RAM), AMD is apparently sticking with the AM3+ platform, and the 9-series chipset (again, according to this rumor) will form the backbone of product support. We'd be surprised if AMD doesn't debut a 10-series chipset along with a second-generation Bulldozer, particularly since the company's Radeon HD 7000 products will supposedly support PCIe 3.0.
Bulldozer is expected to drop in the very near future, with Trinity possibly shipping as early as Q1 2012. No word on Piledriver, but a refresh sometime in H1 2012 would make sense.
As always, keep in mind that Donahim Haber is a rumor site. The slide, however, is a rumor that makes sense. A 10 percent performance jump from Bulldozer to Piledriver is a reasonable increase for a quick architectural refresh. The chip keeps BD's cache layouts and supports version 3.0 of AMD's Turbo Core!. At a guess, this may be a suite of more-advanced power gating and clocking functionalities--it may be that the primary features of TC 3.0 are aimed at Trinity.
The chip also supports new instructions. FMA3 is an Intel-supported instruction for fused multiply-adds. The current state of FMA3/FMA4 (AMD's version) support is a bit confused, but Bulldozer won't ship with FMA3 support. Adding it to Piledriver could improve performance in certain workloads, once compilers are updated to take advantage of the new capability. This last is essential, and it's part of why Bulldozer's performance may improve over time--several of the chip's features may require compiler support to function optimally. New BMI (bit manipulation instructions) could improve the performance of certain loops or conditions.
The one thing we'd have liked to see that isn't mentioned here is any difference in power consumption between the two chips. Oftentimes, companies are able to optimize die layouts from one generation to the next as a means of reducing power, even on the same process. In this case, power savings translate into higher available clock speeds, and that's an option that could only benefit AMD. The company will be taking its eight-core (and higher) products into battle in the server sector next year, and lower TDPs would give AMD room to compete with current Xeons by increasing clock speeds more aggressively.
Other options are as expected. Piledriver will continue to support DDR3-1866 via dual-channel memory (you'll have to buy a server platform if you want quad-channel RAM), AMD is apparently sticking with the AM3+ platform, and the 9-series chipset (again, according to this rumor) will form the backbone of product support. We'd be surprised if AMD doesn't debut a 10-series chipset along with a second-generation Bulldozer, particularly since the company's Radeon HD 7000 products will supposedly support PCIe 3.0.
Bulldozer is expected to drop in the very near future, with Trinity possibly shipping as early as Q1 2012. No word on Piledriver, but a refresh sometime in H1 2012 would make sense.
