22-11-2009, 19:12 | #11 | |
Moonshine
Join Date: Sep 2007
Location: Southampton
Posts: 3,201
|
Quote:
__________________
|
|
23-11-2009, 05:53 | #12 |
Stan, Stan the FLASHER MAN!
Join Date: Jul 2006
Location: In bed with your sister
Posts: 5,483
|
Intel in confusing nomenclature non-shocker
I was looking for information on the i7 920 and one of the Google links took me to an article about the Pentium D 920 (I think I still have one of those ).
__________________
Just because I have a short attention span doesn't mean I... |
23-11-2009, 13:16 | #13 |
HOMO-Sapien
Join Date: Jun 2006
Location: Chelmsford
Posts: 6,692
|
I would personally go the 1366 route. The bandwidth is wasted on gaming but it'll come into it's own for video compression and other simular CPU high intensive tasks. The new breed of CPUs such as the gulftown and will work fine with an x58 chipset for some future upgrading.
The new Samsung 1TB F3's look pretty decent too. I might invest in some RAID'ing with these myself.
__________________
I just got lost in thought.. It was very unfamiliar territory. Techie Talk | My gaming Blog | PC spec | The Admirals log |
23-11-2009, 17:43 | #14 |
Preparing more tumbleweed
Join Date: Jun 2006
Location: Hawaii
Posts: 6,038
|
For me at the moment the 1366 is just silly extra money that I can't really afford, particularly the motherboard, and for very little benefit.
We're talking at least another year until we see 8 cores in CPUs and Intel confirm that you need that at least 6 before triple channel really begins to show any benefit. About the main time I imagine I'll push all 4 cores to their max is some of the stuff I do under Linux, other than that games are still not particularly wonderfully multi-threaded, which is where the Turbo feature of the chip comes into advantage, automatically speeding up cores on the chip if others aren't being utilised. Video compression is increasingly offset to the GPU through CUDA/OpenCL (and decompression through VDPAU and similar.) With the DMI and PCIe links integral in the chips rather than Northbridge on the LGA1156 that's a boost in it's favour, plus video creation benchmarks for the i5-750 seem to leave it fairly even with the i7-920. As far as I could see the pros of cheap and comparable performance to the i7-920 (whilst consuming less power) with only a few negatives against it for situations I'm not likely to be in for a fair while (P55 CrossFire/SLIs nicely but marginally slower than X58 based systems due to the on die PCIe controller sharing 16x bandwidth) Benchmarks like these seem to suggest the only times the i5-950 starts to lose out to the i7-920 is where HyperThreading shows advantages (which isn't all that often), and will knock the i7-920 into a top hat when it comes to single threaded apps: http://anandtech.com/cpuchipsets/showdoc.aspx?i=3634
__________________
Mal: Define "interesting"? Wash: "Oh, God, oh, God, we're all gonna die"? |
23-11-2009, 19:17 | #15 |
Screaming Orgasm
Join Date: Jul 2006
Location: Newbury
Posts: 15,194
|
Very interesting debate here. I keep considering an upgrade and really don't know enough to decide whether to go i7 or i5 (was planning i7, but that was before i5 existed).
In the past, I discovered HT was a waste of time for multi-core/multi-CPU environments. Perhaps my choice of running SETI wasn't a good real-world example, but I found it saturated the memory bandwidth before HT got a look in. This might not happen now with newer cores/bigger caches/faster memory. I just don't know. |
23-11-2009, 19:43 | #16 | |
Preparing more tumbleweed
Join Date: Jun 2006
Location: Hawaii
Posts: 6,038
|
Quote:
Both threads operate on the same core and so use the same on-die cache. Regularly if a thread discovered the data it needed wasn't cached it would flush the cache and fetch it. In a fair number of cases the HT 'virtual CPU' would be handling something different from the primary CPU, resulting in a large waste of time purging CPU cache and refilling it. Depending on the task and the general work of a system HT could actually cause it to be slower! I discovered that to be true even on dedicated MySQL boxes, which I assumed would be fine given the similarity of the task. For the most part the threads on the CPU would be doing totally different tasks with the data, and we'd see a boost from disabling HT.
__________________
Mal: Define "interesting"? Wash: "Oh, God, oh, God, we're all gonna die"? |
|
23-11-2009, 19:46 | #17 |
Moonshine
Join Date: Sep 2007
Location: Southampton
Posts: 3,201
|
I saw some benchmarks that showed HT causing performance drops. I can only imagine they did something stupid to manage that but i'm not personally convinced HT is all that useful. There isn't much an i5-750 won't monster through and the stuff it can't, I suspect more raw speed would be of more benefit than HT.
I honestly think 1366 i7 is now the domain of really hardcore users - people who want/need the fastest CPU they can get, or need a capability of X58 like tri SLI etc. 'Normal' people just don't need it.
__________________
|
23-11-2009, 19:55 | #18 | ||
Preparing more tumbleweed
Join Date: Jun 2006
Location: Hawaii
Posts: 6,038
|
Quote:
Quote:
__________________
Mal: Define "interesting"? Wash: "Oh, God, oh, God, we're all gonna die"? |
||
29-11-2009, 18:57 | #19 |
Preparing more tumbleweed
Join Date: Jun 2006
Location: Hawaii
Posts: 6,038
|
Engadget managed to get hold of a roadmap. Once again Intel aiming with the confusing naming scheme at their high volume sales point. It reads almost like they're deliberately out to fox people, e.g an i5 651 will actually be worse than a i5 650, and will be $90 cheaper. Surely that's an arse about tit way to do it? You'd think a higher number full stop would be better.
__________________
Mal: Define "interesting"? Wash: "Oh, God, oh, God, we're all gonna die"? |
29-11-2009, 19:11 | #20 |
Screaming Orgasm
Join Date: Jul 2006
Location: Newbury
Posts: 15,194
|
|