Looking to the future...

[hans007]

realistically, the mainstream nehalem wont even be out until 2009 summer.

they will probably trickle out some high end 3 channel ddr3 "high end" workstation / "xeon" style ones and call them extreme edition in december. but the 2 channell ddr3 one doesnt come out until mid 2009 and the ddr2 capable budget versions not till fall 2009.

unless you are planning onbuying a super high end machine with ddr3 at the end of 2008 you might as well just buy something now and buy something again 2009 summer.

 

[Idontcare]

Originally posted by: hans007
realistically, the mainstream nehalem wont even be out until 2009 summer.

they will probably trickle out some high end 3 channel ddr3 "high end" workstation / "xeon" style ones and call them extreme edition in december. but the 2 channell ddr3 one doesnt come out until mid 2009 and the ddr2 capable budget versions not till fall 2009.

unless you are planning onbuying a super high end machine with ddr3 at the end of 2008 you might as well just buy something now and buy something again 2009 summer.

And to be realistic, probably the best upgrade you could invest in come mid to late 2009 will be to put an SSD in your system. (other than the obvious GPU upgrade)

I miss my system response time when I had two iram's in raid-0. 8GB wasn't hardly enough to do much with though so I took them out. Can't wait for affordable (<$500) good sized (128GB min) SSD's.

 

[bryanW1995]

Originally posted by: hans007
realistically, the mainstream nehalem wont even be out until 2009 summer.

they will probably trickle out some high end 3 channel ddr3 "high end" workstation / "xeon" style ones and call them extreme edition in december. but the 2 channell ddr3 one doesnt come out until mid 2009 and the ddr2 capable budget versions not till fall 2009.

unless you are planning onbuying a super high end machine with ddr3 at the end of 2008 you might as well just buy something now and buy something again 2009 summer.

how much benefit is the tri-channel going to be vs the dual channel?

thanks dmens, that's a load off of my mind. mostly. there is still a strong incentive for them to somehow try to limit oc'ing. I'd be very surprised if it isn't at least discussed.

 

[Idontcare]

Originally posted by: bryanW1995

Originally posted by: hans007
realistically, the mainstream nehalem wont even be out until 2009 summer.

they will probably trickle out some high end 3 channel ddr3 "high end" workstation / "xeon" style ones and call them extreme edition in december. but the 2 channell ddr3 one doesnt come out until mid 2009 and the ddr2 capable budget versions not till fall 2009.

unless you are planning onbuying a super high end machine with ddr3 at the end of 2008 you might as well just buy something now and buy something again 2009 summer.

how much benefit is the tri-channel going to be vs the dual channel?

thanks dmens, that's a load off of my mind. mostly. there is still a strong incentive for them to somehow try to limit oc'ing. I'd be very surprised if it isn't at least discussed.

Take a stick out of your dual-channel system and see if you notice a performance change when it drops out of dual-channel and starts operating at single-channel...that real-world performance delta (if you notice) will be bigger than the performance delta between dual vs. tri channel DDR3 with those IMC's.

Now the memory benches tuned to highlight superior bandwidth situations will light up like a christmas tree no doubt. Bring on the winrar and SPi benches!

 

[Drsignguy]

Originally posted by: hans007

unless you are planning onbuying a super high end machine with ddr3 at the end of 2008 you might as well just buy something now and buy something again 2009 summer.

Hopefully by then the price will come down considerably. Gain a couple more good 45nm chips, and wait until the excitement settles from Nahalem.

 

[hans007]

Originally posted by: bryanW1995

Originally posted by: hans007
realistically, the mainstream nehalem wont even be out until 2009 summer.

they will probably trickle out some high end 3 channel ddr3 "high end" workstation / "xeon" style ones and call them extreme edition in december. but the 2 channell ddr3 one doesnt come out until mid 2009 and the ddr2 capable budget versions not till fall 2009.

unless you are planning onbuying a super high end machine with ddr3 at the end of 2008 you might as well just buy something now and buy something again 2009 summer.

how much benefit is the tri-channel going to be vs the dual channel?

thanks dmens, that's a load off of my mind. mostly. there is still a strong incentive for them to somehow try to limit oc'ing. I'd be very surprised if it isn't at least discussed.

i dont think most cpus are particularly bandwidth starved.


i supposed nehalem could take advantage of 3 channels more than current core 2 can. the reason you can run a core 2 on single channel sometimes and not notice a difference is that the fsb is saturated.


i.e. a 1066 bus core 2 duo, can run on a single channel ddr2-1066 and its bus is saturated anyway.

but with no fsb on nehalem i think it might see more benefit if you are say.. i dunno doing certain tasks that are very memory intensive ... like say i dont know reading and writing to a giant in memory 4gb database or something,.

that said 2 channels of ddr3 is still a lot of bandwidth, so i doubt a dual ddr3 will be hugely slower than 3 ddr3 channels.

 

[myocardia]

Originally posted by: bryanW1995
how much benefit is the tri-channel going to be vs the dual channel?

If you just go by the numbers, it will be 50% more bandwidth than dual-channel (3/2 = 1.5, or 150%), but like hans said, two channels of DDR3 is alot of bandwidth already. And since you and all of the other folders will be using the higher speed DDR3, it will make even less difference, compared to the people buying pre-builts, that will have either PC8500 or PC10600 DDR3. Of course, it may actually be very beneficial to F@H; if anything would benefit, I can see it being F@H. I guess we'll just have to wait and see.

 

[jones377]

Personally I think Intel added the 3rd memory channel for the sake of extra capacity rather than performance. But no one needs that much RAM on the desktop so the performance benefit will be more important there for the extreme models.

 

[zsdersw]

Originally posted by: hans007
the reason you can run a core 2 on single channel sometimes and not notice a difference is that the fsb is saturated.


i.e. a 1066 bus core 2 duo, can run on a single channel ddr2-1066 and its bus is saturated anyway.

Somewhere I read results that indicated there is no bus saturation.. if there was, the gain from going with 1333 instead of 1066 would be more than it usually is.

 

[biostud]

I thought that the extra bandwidth only was a real problem in server applications, and not that much in desktop applications. Of course extra bandwidth gives some extra speed, but not to such a degree that the bus is saturated. If that was the case, going from a 1066 to 1333Mhz bus would give you ~15-20% performance increase which clearly isn't the case.

 

[Idontcare]

Originally posted by: zsdersw

Originally posted by: hans007
the reason you can run a core 2 on single channel sometimes and not notice a difference is that the fsb is saturated.


i.e. a 1066 bus core 2 duo, can run on a single channel ddr2-1066 and its bus is saturated anyway.

Somewhere I read results that indicated there is no bus saturation.. if there was, the gain from going with 1333 instead of 1066 would be more than it usually is.

Ding ding ding, we have a winner folks!

Yes, this is exactly the case and that is why many rightfully consider the FSB speed increases from 800 -> 1066 -> 1333 -> 1600MHz to all be effectivelly a forced obsolescence/upgrade path and not a true performance increase opportunity.

I personally checked out whether any of these FSB speeds made a difference in the performance of my applications of interest (using my QX6700, so I could keep the CPU speed the same while I switched around the FSB speeds) and I saw <3% performance changes as I crossed all these FSB speeds. YMMV.

 

[magreen]

Right. The reason going with single channel instead of dual channel on a c2d makes little difference is not because the bus is saturated. Rather, it's because of the high latency of the non-integrated memory controller. That high inherent latency dwarfs the latency of the memory itself, so that increasing bandwidth does little to change system performance. That's why ddr2-533 performs about as well as ddr2-800 on a core 2 system.

Contrast that to an AM2 cpu with its IMC where you see a huge overall performance increase going from ddr2-533 to ddr2-800 or 1066. AT wrote this all up at the introduction of the c2d in summer '06.

That's why intel wants triple channel ddr3 with nehalem - with its IMC the memory bandwidth will actually increase system performance.

 

[Idontcare]

Originally posted by: magreen
Right. The reason going with single channel instead of dual channel on a c2d makes little difference is not because the bus is saturated. Rather, it's because of the high latency of the non-integrated memory controller. That high inherent latency dwarfs the latency of the memory itself, so that increasing bandwidth does little to change system performance. That's why ddr2-533 performs about as well as ddr2-800 on a core 2 system.

Contrast that to an AM2 cpu with its IMC where you see a huge overall performance increase going from ddr2-533 to ddr2-800 or 1066. AT wrote this all up at the introduction of the c2d in summer '06.

That's why intel wants triple channel ddr3 with nehalem - with its IMC the memory bandwidth will actually increase system performance.

It will be the saviour for the memory guys that's for sure. Triple-packs of DDR3 and the price premium for higher clocked stuff will once again be justified outside of SPi benchers.

I would add though that I fully expect the addition of the L3$ to Nehalem to further erode the performance impact (good or bad) of the underlying memory subsystem.

 

[magreen]

That's an interesting point. I anyways wonder why they're adding so much cache. Just because they can? You'd think it wouldn't have such an impact once their memory subsystem is in good shape. And it would lower costs to have less cache, no?

 

[bryanW1995]

that's the reason that I've always heard to explain why x2/phenom has so much less cache than core2duo.

@myocardia: I don't fold, I search for aliens. I'm sure the argument still applies, however. That scares me a little bit, I kinda like knowing that I can buy a low-mid system today and oc it way beyond the capabilities of a high-end rig.

 

[Idontcare]

Originally posted by: magreen
That's an interesting point. I anyways wonder why they're adding so much cache. Just because they can? You'd think it wouldn't have such an impact once their memory subsystem is in good shape. And it would lower costs to have less cache, no?

It's an insurance policy against meagerly performing memory (serves to isolate some of the damage of CAS7) as well as a reasonably cheap method if ensuring you have a way to diversify your product offerings by market segment (turn off some of the cache and you have celeron, etc).

Increasing cache sizes does not necessarily decrease yield (redundancy is added to the memory which can be fused off at test if defects caused some of the cache to be non-functional) but yes it does increase cost in that it further decreases the total number of sellable chips you get from a wafer.

You know it can't be all that much of a cost increase, otherwise Intel would have just MCM'ed a big'ole monolithic SRAM array into their wolfdale/yorkfield "CPU packages" rather than making the cache monolithic to the logic die itself.

 

[myocardia]

Originally posted by: bryanW1995
@myocardia: I don't fold, I search for aliens. I'm sure the argument still applies, however.

Ahh, I had forgotten which DC project you ran. IIRC, it seems like SETI isn't as RAM speed-dependent as folding, but is still moreso than your average app.

That scares me a little bit, I kinda like knowing that I can buy a low-mid system today and oc it way beyond the capabilities of a high-end rig.

Don't we all. I've been doing that since the first system I built, a 25Mhz 486 @ 33 Mhz. Man, I sure thought I was something, after first building, then overclocking that system.

 

[roguerower]

So...is the E8*** series going to be the last update before Nehalem? Looking at the roadmap there are no future die shrinks and there's no where to go after 9k unless you start using the old ATI branding of X1***.

 

[Idontcare]

Originally posted by: roguerower
So...is the E8*** series going to be the last update before Nehalem? Looking at the roadmap there are no future die shrinks and there's no where to go after 9k unless you start using the old ATI branding of X1***.

It would appear that way. And with essentially no competition to encourage releasing new parts, and the desire to not set the bar any higher than they have to before Nehalem comes out, I don't see the motivation for Intel to do so.

 

[Dadofamunky]

Originally posted by: Idontcare
And to be realistic, probably the best upgrade you could invest in come mid to late 2009 will be to put an SSD in your system. (other than the obvious GPU upgrade)

I miss my system response time when I had two iram's in raid-0. 8GB wasn't hardly enough to do much with though so I took them out. Can't wait for affordable (<$500) good sized (128GB min) SSD's.

Assuming they come down in price by then... along with DDR3 memory. Till then, I might go for a Q6700 (with its tempting 10X multiplier and big price drop - thank you Intel!) or a Yorkie on my current system by this fall and run that sucker for three years or so. 'Cause it looks like Intel's gonna jack their prices up.

Originally posted by: Idontcare

But at the end of the day here I am still talking about aging and obsolescent 65nm tech while you get to sit back and think about how all us poor suckers just have no idea (we can't, we don't have the insider info) how big nehalem is going to be when it hits the markets...and for that I just ask that you have some forum pity on us blind enthusiasts

Hardly that bad... the 6x50s and quads just came out last year! Geesh. I'm aging and obsolescent. I think the 1st-gen quads still have a long ways to run...

 

[Idontcare]

Originally posted by: Dadofamunky

Originally posted by: Idontcare
But at the end of the day here I am still talking about aging and obsolescent 65nm tech while you get to sit back and think about how all us poor suckers just have no idea (we can't, we don't have the insider info) how big nehalem is going to be when it hits the markets...and for that I just ask that you have some forum pity on us blind enthusiasts

Hardly that bad... the 6x50s and quads just came out last year! Geesh. I'm aging and obsolescent. I think the 1st-gen quads still have a long ways to run...

I've had my first quad since Nov 2006...to me that is ages and ages

I'm ready for an upgrade, Moore's law entitles me to 2X the performance come this fall so I have high expectations for Nehalem.

 

[bryanW1995]

I thought it was more of a "guideline". kinda like the pirate code, right?

If nehalem is 25% better clock/clock than core 2 and smt gives a 50% boost in multi-threaded apps that would yield an 87.5% total improvement, which gets you pretty close to your goal.

 

[Idontcare]

Originally posted by: Idontcare
Point 3: Intel's issue with selling Nehalem versus Penryn is gross margins versus yield. The larger die will result in lower yields, this is an immutable boundary condition of manufacturing processes. The fab's defect density will cause larger dies to have lower yield than smaller dies, no way around it.

So fewer sellable chips per wafer (than Yorkfield) means the gross margins will be lower (than Yorkfield) unless the selling prices are higher (than Yorkfield).

Bloomfield will literally canabilize Intel's own gross margins on QX9650 and QX9770 while at the same time not necessarily bringing higher profits (unless they sell Bloomfield for $1500-$2000) but the gross margins will be lower (yields will be lower) unless they sell for higher ASPs.

This is Intel's issue, they must extract higher gross margins with Nehalem to justify replacing Penryn (on the desktop). If they don't then they are truly wasting their shareholder's entitled profits that could be garnered from selling more Penryn.

Speaking to my 3rd point here is this article at Fudzilla today:

Nehalem goes mainstream in Q4 2009

According to sources close to a motherboard manufacturer, Nehalem CPU in its 1366 socket won't get to mainstream by volume or price until Q4 2009. This might sound shocking to many, as you might expect to upgrade to a quad-core mainstream Nehalem in mid 2009; and unfortunately it won?t happen, at least not for a few hundred Euros / Dollars.

http://www.fudzilla.com/index....=view&id=7391&Itemid=1

 

[Foxery]

Granted it's Faud FUD, but after seeing the huge vacant spaces all over AMD's latest roadmap, 'tis sadly possible. Intel can make more money this way, until the competition gives them some, well, competition.

At least my wallet can breathe a sigh of relief.

 

[Nemesis 1]

Yep. I agree Nehalem for the desktop is going to be $$$$ for sometime. Which is fine. Its also easy to understand.

IF! Nehalem is as good as promised. Intel has to concentrate on the server Market. right off the bate. They also need a highend desktop For everyone to Have on wish list. Until Intel can ramp up.

But the forum topics are going to be bait and flame. Intel way to expensive for 25% performance gain . Until Intel ramps up that is. Than there is the NEW improved topend idea intel has . Most don't like . I do tho . If you pay top $$$ low end stuff shouldn't beable to match its performance. YOU get what you pay for.

The midrange by q4 09 sounds pertty good to me . For these reasons.

1 DDR3 will be cheaper.
2. SSD should be way cheaper.
3. This is my biggest reason why its good for mainstream. LArrabbe will be coming about same time.

Now in noway do I exspect Larrabbe to out perform AMD or NV . But what I do exspect is that Larraddee will scale almost 100% with additional cards up to 4. Maybe on the midrange it won't scale as well as the The highend with the QPlink. I also exspect that nehalem/larrabbe will fully leverade all cores GPU and CPU . IN other words I exspect Larrabbee to perform considerably better with nehalem than it will with AMD processor.

I also exspect the same from AMD in the same time frame. When it comes to multicores AMD is no hillbilly they no there shit. But both intel and amd under license can use logic to make each others parts work well,If they choose to do so .

NV has I believe a HT license with AMD But they don't have a QPlink! Normally I would be against Intel on not giving NV QPlink. But because of the way NV has acted with sli I am all for intel not giving NV QPlink license .

So basicly I see the late arrival of mainstream nehalem as a good thing for consumers . One never knows What new tech can make old teck old very quickly.

The only bad thing I see is all the whining! There will be in the forums until good performance mainstream Nehalems will be available.

Its still 5 months away and the whining has already begun . LAST but not least. Is we don't know yet how Nehalem performs. If its crappy don't look for high prices.

To me the nehalem that sounds interesting is the last one to come the 2 core model . If it can do 4 threads at once. And the improvements Intel claims it made to H/T is true . This babe is going to be cheap effficient the perfect ans todays desktop.

For me the big question is that QPlink between cpu and gpu on the highend Nehalem . How much is that going to effect performance and will larrabbee leverage it. Yep I think it will . Bottom line is this guys and girls for the first time in along time . We get exactly what we pay for . I like this . Many don't . But its how it works in the real world.