- 新着としてマーク
- ブックマーク
- 購読
- ミュート
- RSS フィードを購読する
- ハイライト
- 印刷
- 不適切なコンテンツを報告
I just bought me an early xmas gift :)
Are there any special flags I need to turn on to get the best performance out of the i7-970, or the upcoming Sandy Bridge? If so, what areany dis-advantages to turning on those flags?
Will the special flags cause the code to not work on older CPUs, and will it cause the newer CPUs to get different numerics from the old ones?
Are there any special flags I need to turn on to get the best performance out of the i7-970, or the upcoming Sandy Bridge? If so, what areany dis-advantages to turning on those flags?
Will the special flags cause the code to not work on older CPUs, and will it cause the newer CPUs to get different numerics from the old ones?
コピーされたリンク
3 返答(返信)
- 新着としてマーク
- ブックマーク
- 購読
- ミュート
- RSS フィードを購読する
- ハイライト
- 印刷
- 不適切なコンテンツを報告
See here. If you use the /Qax form of the option the code will run on older processors (SSE2 capable by default). It's always possible that optimizations, especially vectorization, can affect results.
- 新着としてマーク
- ブックマーク
- 購読
- ミュート
- RSS フィードを購読する
- ハイライト
- 印刷
- 不適切なコンテンツを報告
Turned out great! I went with SSE4.2, that's the latest my compiler had. On x86, I got a different set of results vs. without it, butI got digit for digit agreement between the new sets of results on Pentium M vs Core 2 vs i7. On x64, even with the SSE4.2 option, I had digit-for-digit agreement vs. without the option (Core2, i7). And, oh yeah, (new run time)/(old run time) = (about 0.6) doesn't hurt either.
You have my vote to make /Qa(latest your compiler will support) the default!
You have my vote to make /Qa(latest your compiler will support) the default!
- 新着としてマーク
- ブックマーク
- 購読
- ミュート
- RSS フィードを購読する
- ハイライト
- 印刷
- 不適切なコンテンツを報告
Glad to hear you got good results. For quite a number of reasons, we aren't going to make auto-CPU-dispatch the default, but it's there if you want it.
