i7-900s vs Xeon processors - identical?
- Thread starter fosch
- Start date
Don't know for a 920, except you want the D0 stepping one and check for reports on a good bin number. All W3520 are high binned. The W3520 at 9700 is the average for my friend's test. High end MB, super stable voltages, low voltage low latency memory tweaked settings (7-7-8-7-22).
That's one of the problems with benchmarks, what else is in the mix, and the friend is constantly playing with settings. Had the W3520 going at 4.3 for a while on air, huge heatsink, heatpipes routed through the HE, the fins and a pelltier. But, it would fold for two days max without crashing. Not what we want.
Rock stable at 3.9. Fold for a week without issues (his max test).
Yet on the same platform with stock settings, the stock 970 was completing folding so fast they upped the job size by 40% to 60% over the OCed W3520 and he was still getting the same completion rate as the OC W3520, and often slightly above even with the increased load.
Right Now:
If you have a choice between a 920 and a W3520 for an i7 desktop MB, go for the W3520. IF you already have an X58 1366 MB or get it for a really good price.
If you're getting everything new, the i7 2600K is a great bang for the buck. And although a 970 appears (at this early stage) to do more real work, it costs a lot more, and there's more Sandybridges coming down the road, many of which will work on the same MB as a 2600K (don't know about upper hexes or the octs). We anticipate a Sandybridge at the current cost of a 970 that should completely blow the doors off a 970, and for real work. The internal architecture of the Sandybridge is that good.
IMPORTANT UPDATE
Apparently there's more to this. We don't know how many more faster Sandybridge CPUs will be released for the 1155 socket.
I'm told the more powerful Sandybridge CPUs will require a different socket, first one, and then another, so they won't work on a 1155 motherboard.
Also, the 1155 MBs have limited PCI-E lanes available (30? 28?), so the specs for those MGs have 16x for one card only, add more PCI-E cards have to run slower with fewer lanes. Newer sockets/motherboards for up-coming Sandybridge are supposed to have 48 PCI-E lanes.
You could, theoretically, need "lots of PCIE lanes" ... ONLY if running multiple video cards and mutiple monitors. Considering the bandwidth used in market data streams, data requirement marketing hype is all a hoax! ANY modern mobo and video card(s) combo can handle market data streams by a factor of 1000x, minimum.
Yes, but the point is that it appears that the 1155 upgrade path will be limited.
Also, we don't know what other PCI-E cards he plans to use. I believe he'll also use the machine for still photography - PCI-E SSD? If any serious work with video or RAW video, unless he's willing to let the machine sit rendering for a day or two on some edits (instead of trading) he'll need an SSD and preferably a PCI-E SSD. There may also be other uses undisclosed as they're perceived as less than that of trading or photography, yet they'll need another PCI-E card.
By any standard, the GPU capability is so far beyond what we need...
But lanes, that's bandwidth...
However, when I attempted to do the math, for the six securities I follow, along with the Level II for two currently trading, and broker's platform, onto six monitors: the best we could determine is that a single lane was sufficient with a large safety factor. We don't know if we are right, nor what I may want the system to do down the road.
This one has 49 lanes... GIGABYTE GA-P67A-UD7