""You only compared two dimensions, what about the third?""
sorry, I don't live on these boards...
depth is more or less irrelevant, since racks of these type are basically standardized. It's not like they're going to have to build extra deep racks for Xserves. And I really don't mean anything as a jab at you, but if you've worked with racks, doing network or server builds, you'd know that most non-backbone type network gear is 'half-depth' (but still 19" wide), and servers are varying depths, but all deep enough that they need a full depth rack. so whether it's a Power Mac or an Xserve doesn't matter for depth, they're both deeper than workgroup switches, etc. and require fullsize racks. Normally the extra depth is used for cooling and cable management stuff.
But I still don't understand why this is turning into a big debate...exactly how many PM's or Xserves you can fit into a give rack space isn't that big of a deal. None of it is relevant to predicting future processor releases.
Not trying to cause trouble, really just trying to help, and steer discussion in a more interesting direction (bigmac potentially taking #1 spot, and the implications of that).
As for the 'custom' racks of bigmac 1, it looks from the pics (though I'm NOT sure) like they're standard 19" racks, some of which have divider side panels mounted vertically on the fronts, probably for additional cooling and cable management. and by the way, no one will ever understand what a HUGE headache cable management really is until they've wired an entire network closet / server rack. especially an install like this, where there are actually two network topologies, power, and presumably some type of hardware / temp reporting.
Anyway, let's get back on track to have a nice constructive discussion.
3 PM's wide x 8.1 inches = 24.3 inches wide as a minimum.
Ok, so what's the point? I think we all agree the racks are custom and would have to be replaced as they won't handle the 19" standard xServe form factor.
Ok, so what's the point? I think we all agree the racks are custom and would have to be replaced as they won't handle the 19" standard xServe form factor.
They are custom racks. They also came with custom flanges that narrow the width to a standard 19" since VT knew they'd be replacing the Desktop units. They will not have to replace the racks.
They are custom racks. They also came with custom flanges that narrow the width to a standard 19" since VT knew they'd be replacing the Desktop units. They will not have to replace the racks.
Oh that's cool. I didn't realize that. It almost looks like the inner vertical rack mount rails are fixed in all the pictures I've seen.
One potential speedup I haven't seen mentioned is a new compiler backend. I believe sometime after BigMac made its appearance, IBM announced a commercial compiler for PPC which takes advantage much better of the G5 (than does gcc --- which Mac OSX uses). I believe this was to account for a 25-40% speedup. Oddly, I have never seen mention of this compiler technology again, though I would have expected it in IBM's best interests to share this technology with Apple (to fold into gcc) or even simpler for UV to buy one the compiler to recompile and run on the BigMac (thought perhaps the executables don't run on OSX?). If such a PPC-specific backend were introduced, that alone should boost the raw horsepower of each G5 process by 25% which would be a freebie more or less.
I'd like to add in with the question of whether you can simply predict a liner speed increases by adding an indefinite number of nodes (or a couple of thousand).
I have no idea, but I would assume that there may be some sort of network latencies and other related efficiency loss with the additional nodes that at some point may eat into those projected Rmax and Rpeak #'s.
I'd like to add in with the question of whether you can simply predict a liner speed increases by adding an indefinite number of nodes (or a couple of thousand).
I have no idea, but I would assume that there may be some sort of network latencies and other related efficiency loss with the additional nodes that at some point may eat into those projected Rmax and Rpeak #'s.
Well Rmax is theoretical so it never decreases but the efficiency drops with increased numbers of nodes so no it isn't linear.
I would like to think that the original post does seem to be very likely as VT still has not been delivered with all units. (Atleast there is no Public Confirmation).
Whatever problems IBM may have had fabbing the 970fx chips, I am sure that they would have fabbed the reqd no by now.
I am betting that there will be a new Xserve at WWDC which will help VT go past the no. 3 slot.
I would like to think that the original post does seem to be very likely as VT still has not been delivered with all units. (Atleast there is no Public Confirmation).
Whatever problems IBM may have had fabbing the 970fx chips, I am sure that they would have fabbed the reqd no by now.
It's unlikely, but possible, that the same chips are also going into other Apple's product(s) like iMacs, and Apple is putting 99% of them into iMacs in anticipation of wild demand.
Would you rather supply customers with machines that they have ordered for or use the chips to manufacture and stock machines that are not even announced as yet???
There is no way a company like Apple could afford to manufacture and stock new machines at the expense of shipping existing orders.
And I read that VT is going to be expanding from 2200 970-CPUs to 3200 970fx-CPUs. That IS the xServe G5 and its target date for starting was May 2004. Also, it looks like they have 2 more upgrades planned for 2005-6 and future.
Would you rather supply customers with machines that they have ordered for or use the chips to manufacture and stock machines that are not even announced as yet???
I wouldn't. That's why I said 'unlikely'.
Quote:
There is no way a company like Apple could afford to manufacture and stock new machines at the expense of shipping existing orders.
Who knows what's going on right now at VT? Maybe they are building a 10 000 node-cluster. Maybe, Apple offered a huge discount for shipping delays. Maybe, there are only 100 working 970FX chips at 90nm on the planet?
Comments
Originally posted by oldmacfan
Ok concentricity, are you ignoring this question.
""You only compared two dimensions, what about the third?""
sorry, I don't live on these boards...
depth is more or less irrelevant, since racks of these type are basically standardized. It's not like they're going to have to build extra deep racks for Xserves. And I really don't mean anything as a jab at you, but if you've worked with racks, doing network or server builds, you'd know that most non-backbone type network gear is 'half-depth' (but still 19" wide), and servers are varying depths, but all deep enough that they need a full depth rack. so whether it's a Power Mac or an Xserve doesn't matter for depth, they're both deeper than workgroup switches, etc. and require fullsize racks. Normally the extra depth is used for cooling and cable management stuff.
But I still don't understand why this is turning into a big debate...exactly how many PM's or Xserves you can fit into a give rack space isn't that big of a deal. None of it is relevant to predicting future processor releases.
Not trying to cause trouble, really just trying to help, and steer discussion in a more interesting direction (bigmac potentially taking #1 spot, and the implications of that).
As for the 'custom' racks of bigmac 1, it looks from the pics (though I'm NOT sure) like they're standard 19" racks, some of which have divider side panels mounted vertically on the fronts, probably for additional cooling and cable management. and by the way, no one will ever understand what a HUGE headache cable management really is until they've wired an entire network closet / server rack. especially an install like this, where there are actually two network topologies, power, and presumably some type of hardware / temp reporting.
Anyway, let's get back on track to have a nice constructive discussion.
con...
Originally posted by oldmacfan
3 PM's wide x 8.1 inches = 24.3 inches wide as a minimum.
Ok, so what's the point? I think we all agree the racks are custom and would have to be replaced as they won't handle the 19" standard xServe form factor.
Originally posted by Rhumgod
Ok, so what's the point? I think we all agree the racks are custom and would have to be replaced as they won't handle the 19" standard xServe form factor.
They are custom racks. They also came with custom flanges that narrow the width to a standard 19" since VT knew they'd be replacing the Desktop units. They will not have to replace the racks.
Originally posted by Tomb of the Unknown
They are custom racks. They also came with custom flanges that narrow the width to a standard 19" since VT knew they'd be replacing the Desktop units. They will not have to replace the racks.
Oh that's cool. I didn't realize that. It almost looks like the inner vertical rack mount rails are fixed in all the pictures I've seen.
Anyone have further news or insights?
I have no idea, but I would assume that there may be some sort of network latencies and other related efficiency loss with the additional nodes that at some point may eat into those projected Rmax and Rpeak #'s.
or, maybe not...........
Originally posted by tink
I'd like to add in with the question of whether you can simply predict a liner speed increases by adding an indefinite number of nodes (or a couple of thousand).
I have no idea, but I would assume that there may be some sort of network latencies and other related efficiency loss with the additional nodes that at some point may eat into those projected Rmax and Rpeak #'s.
Well Rmax is theoretical so it never decreases but the efficiency drops with increased numbers of nodes so no it isn't linear.
IBM's XLC compiler is pretty well known; all the Mac news sites mentioned it when it came out.
Originally posted by wmf
Keep in mind that Big Mac will be bumped to #4 because Thunder is now #2 at 20 TFLOPS.
IBM's XLC compiler is pretty well known; all the Mac news sites mentioned it when it came out.
Does anyone have any more new details about IBM's XLC compiler? Will Apple
showcase it's power in OS 10.4 or is it using it already for the incremental
updates?
Any news of software companies taking advantage of the XLC compiler
(Adobe, Macromedia, Alias, etc.)?
Originally posted by Telomar
Well Rmax is theoretical so it never decreases but the efficiency drops with increased numbers of nodes so no it isn't linear.
Just a correction, Rpeak is theoretical, Rmax is actual.
http://www.top500.org/
Whatever problems IBM may have had fabbing the 970fx chips, I am sure that they would have fabbed the reqd no by now.
I am betting that there will be a new Xserve at WWDC which will help VT go past the no. 3 slot.
Originally posted by fresco
I would like to think that the original post does seem to be very likely as VT still has not been delivered with all units. (Atleast there is no Public Confirmation).
Whatever problems IBM may have had fabbing the 970fx chips, I am sure that they would have fabbed the reqd no by now.
It's unlikely, but possible, that the same chips are also going into other Apple's product(s) like iMacs, and Apple is putting 99% of them into iMacs in anticipation of wild demand.
There is no way a company like Apple could afford to manufacture and stock new machines at the expense of shipping existing orders.
Originally posted by fresco
Would you rather supply customers with machines that they have ordered for or use the chips to manufacture and stock machines that are not even announced as yet???
I wouldn't. That's why I said 'unlikely'.
There is no way a company like Apple could afford to manufacture and stock new machines at the expense of shipping existing orders.
Who knows what's going on right now at VT? Maybe they are building a 10 000 node-cluster. Maybe, Apple offered a huge discount for shipping delays. Maybe, there are only 100 working 970FX chips at 90nm on the planet?