I usually don't post much, but this one deserves it. I am surprised by the big loads of cr@p posted by several members. So let's clear it out.
- The K8 HAS NO FSB!!!! Read the white papers. The integrated memory controller + the Hypertransport(HT) link "replace" the FSB. The HT link is NOT a kind of FSD, neither is the memory controller.
- PCI express to the K8 requires NO change to the CPU. You know the reason. The CPU communicates to the chipset through the HT link, and that doesn't change. A chipset with PCI express for the K8 was late because the maker were late to PCI erxpress in general. In fact, a K8 chipset is much easier to design than a P4 / P-M chipset... it is just a peripherials controller (SB) plus the HT link
- Centrino is the name of the bundle that contains chipset, wireless LAn and Pentium M CPU. ALL 3 of them have been optimized for battery life.
- The pentium M is a really good one CPU. Since the original banias it was an excellent performer. Dotham is even better. However, for RAW horsepower it is still no match for the brute force of the K8.
- Some reviews show a K8 and a dotham neck and neck, so why I say the P-M is still no match for the K8? Well, many of those reviews contain a lot of becnhmarks that are very cache friendly (super-pi the clearest example). The 2 MB L2 of Dotham surely help it. But once the cache is exhausted and the memory plays, performance drops quite drastically. Gaming is also very sensitive to cache (3.4 P4EE beats prescott 3.8 in the majority of the games) so Dotham gets good performance and gives a good fight. Office applications also get help from a large cache (a laptop is mainly used for office applications, so that is why Dotham got the 2MB L2) However, any content creation test and any data intensive test will show the K8 beating the living cr@p out of it (CAD and database for example) Video encoding is another example where Dotham will be beaten handily. Dotham is very good performer, very well balanced but those who think it can beat a K8 need to read more. Analyze the nature of the benchmarks!!
- Power consumption between the CPUs is quite close. The least hungry mobile Athlon 64 is already at 25W max versus 21 for the P-M. Even the notebooks based on the 62W K8s are getting 3+ hours of battery life from 8 cells 4400 mAH (typical consumption 25W) . A notebook is more than just the CPU, and that is where AMD had nothing. The K8T800 is a desktop chipset. Once the chipset reduces the power consumption the field will be more balanced.
- In the long run, SOI will give AMD the edge in battery life. Dotham frugality is due in a great parts to the ability to shutdown parts of the L2 cache. this cache gating technique saves a lot of energy in leakage current. The L2 cache is the biggest spender in the CPU, that is why the 25 W K8s have only 256 KB. AMD need a larger cache with gate switching.
- Cache in the K8 is less of a factor because of the integrated memory controller. That is why the 3800+ and the FX-53 run that close in benchmarks. However, a 128 KB is too little. The CPU in Turion need to have at least 512 KB.
- Dotham relies a lot in the cache because of the slow FSB. Even at 533 MHz, they need the big cache. Being the cache the largest contributor to performance, the 533 FSB won't be the silver bullet many believe.
- The 64 bits OS is a big deal. There are additional general purpose registers that the OS can use freely. If you have installed the 64 bits beta of XP AMD64 you noticed that video encoding performance using the same 32 bits applications is between 7% and 35% FASTER (TMPGenc of the low side, Ulead video studio int he high side) The OS uses the additional registers if the application doesn't command specifi use of them. Think of these registers as SS2 on stereoids.
- Turion will be a complete platform of a low consumption chipset, a very low power mobile Athlon 64 (L2 gate clocking or similar) and low power components.
Alex
PS. Dotham lovers, I know you won't like the fact that I said it is still no match for the K8. My M6805 is ready for becnhmarks. Just get ready to play rough. No super pi or sysmark garbage. We will go onto AutoCAD, solidworks, PSpice, DV encoding and similars. You want a machine to brag about 3dmark, get a P4EE.
- The K8 HAS NO FSB!!!! Read the white papers. The integrated memory controller + the Hypertransport(HT) link "replace" the FSB. The HT link is NOT a kind of FSD, neither is the memory controller.
- PCI express to the K8 requires NO change to the CPU. You know the reason. The CPU communicates to the chipset through the HT link, and that doesn't change. A chipset with PCI express for the K8 was late because the maker were late to PCI erxpress in general. In fact, a K8 chipset is much easier to design than a P4 / P-M chipset... it is just a peripherials controller (SB) plus the HT link
- Centrino is the name of the bundle that contains chipset, wireless LAn and Pentium M CPU. ALL 3 of them have been optimized for battery life.
- The pentium M is a really good one CPU. Since the original banias it was an excellent performer. Dotham is even better. However, for RAW horsepower it is still no match for the brute force of the K8.
- Some reviews show a K8 and a dotham neck and neck, so why I say the P-M is still no match for the K8? Well, many of those reviews contain a lot of becnhmarks that are very cache friendly (super-pi the clearest example). The 2 MB L2 of Dotham surely help it. But once the cache is exhausted and the memory plays, performance drops quite drastically. Gaming is also very sensitive to cache (3.4 P4EE beats prescott 3.8 in the majority of the games) so Dotham gets good performance and gives a good fight. Office applications also get help from a large cache (a laptop is mainly used for office applications, so that is why Dotham got the 2MB L2) However, any content creation test and any data intensive test will show the K8 beating the living cr@p out of it (CAD and database for example) Video encoding is another example where Dotham will be beaten handily. Dotham is very good performer, very well balanced but those who think it can beat a K8 need to read more. Analyze the nature of the benchmarks!!
- Power consumption between the CPUs is quite close. The least hungry mobile Athlon 64 is already at 25W max versus 21 for the P-M. Even the notebooks based on the 62W K8s are getting 3+ hours of battery life from 8 cells 4400 mAH (typical consumption 25W) . A notebook is more than just the CPU, and that is where AMD had nothing. The K8T800 is a desktop chipset. Once the chipset reduces the power consumption the field will be more balanced.
- In the long run, SOI will give AMD the edge in battery life. Dotham frugality is due in a great parts to the ability to shutdown parts of the L2 cache. this cache gating technique saves a lot of energy in leakage current. The L2 cache is the biggest spender in the CPU, that is why the 25 W K8s have only 256 KB. AMD need a larger cache with gate switching.
- Cache in the K8 is less of a factor because of the integrated memory controller. That is why the 3800+ and the FX-53 run that close in benchmarks. However, a 128 KB is too little. The CPU in Turion need to have at least 512 KB.
- Dotham relies a lot in the cache because of the slow FSB. Even at 533 MHz, they need the big cache. Being the cache the largest contributor to performance, the 533 FSB won't be the silver bullet many believe.
- The 64 bits OS is a big deal. There are additional general purpose registers that the OS can use freely. If you have installed the 64 bits beta of XP AMD64 you noticed that video encoding performance using the same 32 bits applications is between 7% and 35% FASTER (TMPGenc of the low side, Ulead video studio int he high side) The OS uses the additional registers if the application doesn't command specifi use of them. Think of these registers as SS2 on stereoids.
- Turion will be a complete platform of a low consumption chipset, a very low power mobile Athlon 64 (L2 gate clocking or similar) and low power components.
Alex
PS. Dotham lovers, I know you won't like the fact that I said it is still no match for the K8. My M6805 is ready for becnhmarks. Just get ready to play rough. No super pi or sysmark garbage. We will go onto AutoCAD, solidworks, PSpice, DV encoding and similars. You want a machine to brag about 3dmark, get a P4EE.
