Monday, March 31, 2008
Intel's I/O stalled !!
"IO," said one manager of a major datacentre, is the great big bottleneck. He said that AMD blades performed better by between 10 and 50 per cent in his corporation, and he'd asked Intel over and over again when it would deliver this kind of throughput. If it can't produce chipsets, who can, he asked.
Badhardware has warned people on this dark side of Intel's hardware (AMD's Opteron is immune to this thanks to HT). And the obvious congestion cause at Intel is prehistoric FSB customized for single processor personal computer:
We will wait *WWW) almost the whole year to see QPI and Nehalem delivered by Intel. However, being B3 BArcelona revision is ready now, AMD delivers it . Is delivering NOW.
Badhardware has warned people on this dark side of Intel's hardware (AMD's Opteron is immune to this thanks to HT). And the obvious congestion cause at Intel is prehistoric FSB customized for single processor personal computer:
QPI vs. FSB | ||
---|---|---|
| Intel FSB(Front Side Bus) | Intel QuickPath Interconnect(QPI) |
Topology | Shared Bus | Point to Point Link |
Physical Bus Width(bits) | 64 | 20 |
Data Transfer Width(bits) | 64 | 16 |
Requires Side-band Signals | Yes | No |
Total Number of Pins | 150 | 84 |
Clock Per Bus | 1 | 1 |
Embedded Clock | No | No |
Bus Direction | Bi-directional | Uni-directional |
We will wait *WWW) almost the whole year to see QPI and Nehalem delivered by Intel. However, being B3 BArcelona revision is ready now, AMD delivers it . Is delivering NOW.