HYPER-THREADING: THE NEXT GENERATION
46 |MAMAMAXIMXIMXIMXIMUUUUMMPPPCC| HOLIDAY 08 | http://www.maximumpc.com
Hyper-Threading got a bad rap under Pentium 4 for being more a
hindrance than a help to performance. Our tests then showed that
HT generally helped, but the lack of threaded applications made the
feature pretty near worthless. Intel has reintroduced Hyper-Threading
with the Core i7 and says it’s worth another look. We ran a hand-
ful of our multithreaded applications with HT both on and off and
determined that this time around, it’s good stuff. We generally saw a
healthy double-digit boost in performance with HT enabled. Using the
latest version of ProShow Producer, we actually took a 26 percent hit
by turning off Hyper-Threading. MainConcept’s encoder experienced
a drop of 17 percent without Hyper-Threading. So, if you ask us, you
oughta leave it on.The Core i7 Dissected
The CPU sports some unique features—we test their merits
IN T E L’S CO R E I 7
BENCHMARKS
HT On HT Off Difference
MainConcept (sec) 958 1,153 1,153 1,153 -17%-17%-17%
ProShow Producer 3.5 (sec) 642 868 868 868 -26%-26%-26%
Cinebench 10 32-bit 15,398 13,451 13,451 13,451 -13%-13%-13%
Cinebench 10 64-bit 18,963 16,613 16,613 16,613 -12%-12%-12%
POV-Ray (sec) 408 416 416 416 -2%-2%-2%
3DMark Vantage CPU 39,725 35,623 35,623 35,623 -10%-10%-10%
Best scores are bolded.TINKERING WITH TURBO MODE
Intel’s Turbo Mode gives the user fi ne-grain control over individual
cores. By shutting down individual cores that aren’t used during, say, a
single-threaded game, you can pick up what is essentially free perfor-
mance by overclocking, or rather, Turboing, from 3.2GHz to 3.8GHz. We
dialed up the allowable, um, Turbos from the stock 24 to 27 to see if the
feature works. Indeed it does. In our mostly single-threaded Photoshop
CS3 test and World in Confl ict, we saw the scaling you’d expect from
a 10 percent overclock. Since we didn’t choose to overclock for two
threads, we didn’t see much of a change in Quake 4. Our verdict is that
it’s a worthwhile proposition, the caveat being that you will need liquid
cooling or a big, fat heatsink to truly exploit its potential.Turbo On Turbo Off Difference
Photoshop CS3 (sec) 102 110 110 8%8%
Quake 4 (fps) 233 228 228 2%2%
World In Conflict (fps) 272 250 250 9%9%
Best scores are bolded.BENCHMARKS
TRI-CHANNEL MEMORY TESTED
Core i7’s tri-channel DDR3 memory controller presents a radical alter-
native to the standard dual-channel confi gurations. Since the controller
lets you run single, dual, or tri mode, we decided to take a look at the
actual bandwidth off ered by each scenario and the resulting real-worldimpact. Using three Qimonda 1GB DDR3/1066 DIMMs and a single
Corsair 2GB DDR3/1600 DIMM (set at DDR3/1066), we ran two RAM
benchmarks and Quake 4. The upshot is that for the best performance,
you should populate three channels.BENCHMARKS
3 DIMMs (3GB) 2 DIMMs (2GB) 1 DIMM (2GB) 1 DIMM (1GB)
Sisoft Sandra RAM Bandwidth (GB/s) 18.1 12.7 7.1 6.7
Everest Ultimate MEM Read (MB/s) 15,167 14,388 8,317 8,236
Everest Ultimate MEM Write (MB/s) 12,041 13,590 8,285 8,187
Everest Ultimate MEM Copy (MB/s) 15,583 14,848 9,062 7,798
Quake 4 (fps) 228.0 172.4 213 167
Best scores are bolded.