Smithfield: 2Q05, 2.8Ghz for $240!!!

[Mephistopheles]

We all know that Smithfield is not as technically interesting as Toledo, AMD's alternative that is going to be launched by the 2H05. The obvious speculation leads us to believe that smithfield is probably too hot and too underperforming... But recently, a little more info has been leaked that shows that there might be at least one small advantage for them: Prices.

Smithfield 840 $528
Smithfield 830 $314
Smithfield 820 $240
<A HREF="http://www.theinquirer.net/?article=20826" target="_new">(pricing info from the inquirer)</A>

In addition to that, Smithfield will be launched in <A HREF="http://www.theinquirer.net/?article=20806" target="_new">late 2Q05</A>, which probably means it might get launched before toledo. (I wonder if it'll be another paper launch?... I also wonder if Intel could flood the market with 820s... and make it all sound as if it was their idea, not AMD's, and they haven't just slapped two cores together...)

This will mean a lot if, as we're told, AMD only gets dual-core Toledo Athlon FX processors with FX price tags (500+). I mean, a dual-core processor for $240 by july (if that schedule is to be believed in) might just make for an interesting product. The only problem is that you'll probably have to buy one hell of a powerful heatsink and fan for the processor.... And you'll spend extra money if you don't want it to sound like a jet engine...

 

[Starfishy]

Do we know any projected prices for the Toledo though?? How do we know that AMD won't still undercut the Smithfield??

 

[peteroy]

Dual core CPU might use BTX which is better in heat dissapation than ATX.

==================================
<A HREF="http://www.clancas.net" target="_new">clan CHAOS</A>
<A HREF="http://www.daynighthosting.com" target="_new">DayNightHosting.com</A>

 

[Mephistopheles]

I don't know of any projected prices for toledo, but it's slated to be introduced as Opteron CPUs first and FXs first, which basically means $500+ prices. This is of course only my speculation.

In any case, I think Intel's production capacity, even if it is for a sub-par product technically, is always traditionally better than AMD's, which means they actually could price their products lower than AMD's if they wanted to.... in theory.

 

[wolverinero79]

Because AMD can't take the ASP hit in both the flash market and CPU market. I have no idea how Intel's going to make those magic price numbers for dual core chips - very interesting indeed.

I'm just your average habitual smiler =D

 

[Mephistopheles]

I have no idea how Intel's going to make those magic price numbers for dual core chips - very interesting indeed.

Exactly. US$240 is quite buyable, FX prices and Opterons are not...

 

[P4Man]

I've not been keeping up, so please clue me in here. Smithfield is a dual core prescott ? If so, *yawn* @2.8 GHz it will be like an SMP Celeron: pretty good at precious few things. By that time $240 will buy you a cpu that is considerably faster on 98% of the apps out there. oh wait, $250 already buys me that today.

Wake me up when we get higher clocked 64 bit enabled Dothans. Dual core is nice, but no substitute for single threaded performance unless you run a server or workstation app that can take advantage of it.

= The views stated herein are my personal views, and not necessarily the views of my wife. =

 

[G_K]

It would be interesting ot see if current Socket T mother boards will get bio's to support Dual core processors. If so the upgradability of intel would trounce that of AMD.

 

[sobelizard]

Don't kid yourself - Intel will make more money with a new chipset requirement.

<b><i>Powered by <font color=blue>V</font color=blue><font color=purple>E</font color=purple><font color=red>R</font color=red><font color=purple>T</font color=purple><font color=blue>O</font color=blue></b>
Fueled by <b><font color=blue>CL-</font color=blue><font color=red>ONE</font color=red></b>

 

[endyen]

They have already said that dual core will require a lot more power. No socktT made today can take thay kind of abuse.

 

[juin]

A dual core presscott? That make 140 million logic transistor that 5 time larger that a NW core.The yield of this chip will be very very bad and at this price Gross profit margin will be close to 0%.

I personnaly believe it a Nw core that consume less and will performe better intel did got 2 year to rework it or miracle have happen with the prescott.

i need to change useur name.

 

[Unolocogringo]

Welcome Back

I aint signing nothing!!!

 

[jammydodger]

I still havent seen anything indicating that people will get a performance boost in programs that are not programmed to take advantage of multiple cores. In games dual core will make [-peep-] all difference.

Dual core CPU might use BTX which is better in heat dissapation than ATX

BTX is layed out so as too better circulate air over the hottest components. You are still going to need a HUGE heatsink and fan to get rid of the heat of a dual core prescott! A single core prescott can dissipate up to 102w of heat! Think about a dual core one, my waterblock can only get rid of 200w!
You seen to make out that BTX is the miricle cure for all our heat problems.

 

[Mephistopheles]

Well, it is more or less safe to assume that prescott's dark transistors were causing all of the heat problems. Remember, prescott has over 2x the logic transistors that NW had... If Intel got rid of the garbage and got the core working more smoothly, then maybe smithfield will work as a product...

 

[Xeon]

Rest assure they are both prescott's and also rest assure it will not be as bad as everyone thinks.

All the thermal control IC's from P-M and Itanium series(s) will be in their, on the fly clock speed adjustment, IC wait,stall,hybernate states.

As for software performance it will be on par with...

Xeon

<font color=red>Post created with being a dickhead in mind.</font color=red>
<font color=white>For all emotional and slanderous statements contact THG for all law suits.</font color=white>

 

[Vapor]

I'm extremely skeptical of this chip's performance, especially the 820 and 830.

For one, a single Scotty core eats bandwidth like no other, imagine two. Second, the only type of RAM, DDR2, that could provide this bandwidth is expensive and has horrible latency at high speeds (newer modules can run 3-2-2-8 at 266MHz, which isn't shabby, but still isn't enough bandwidth and still costs a fortune). Sure, Scotty doesn't care THAT much about latency, but I have a feeling that the way the arbiter unit on Smithfield will work will be in a way that that won't effectively cache (as it doesn't have any of its own cache) requests and will therefore NEED extremely low NB and mem latencies (not just timings, but true latency, something Intel's off-die mem controller <b>cannot</b> provide) to handle the requests of both cores. DDR2 does not have these required timings to even begin to make up for the lack of an ondie mem controller, either.

So what <i>could</i> that mean? Slower performance than a single core at the same clock (maybe that's why HT was disabled...) by a small (likely very small) amount on single-threaded apps. It will also mean poor multi-threaded scaling as opposed to what AMD could do. Not only that, but the clockspeeds of these chips are puny by modern Intel (somewhat required) standards. The power dissipated is also astronomical at all three clockspeeds and will likely only get worse since these per-core-power-dissipation numbers are among Intel's lowest thought possible (55-60W is low for a Scotty).

Intel, IMO, NEEDS to make Yonah soon. Dothan doesn't use all the bandwidth provided by dual-channel DDR2 and doesn't seem to mind the higher latencies of DDR2. Not only that, but the arbiter unit won't have to work as hard due to the lower clocks per core and won't be tied down by bandwidth arbitration. ON TOP OF THAT, Dothan's performance is as great as the best P4, runs at an extremely low 27W of power and has the overhead (in terms of power) available for 64-bit, NX, etc.

Looks like another ploy by Intel to exercise their marketting department's dominance of AMD's. But for people looking for a processor, AMD's dual core solution (although later and likely more expensive) was built from the ground up for dual core and will really be the only solution.

Maxtor disgraces the six letters that make Matrox.

 

[P4Man]

>For one, a single Scotty core eats bandwidth like no other,
>imagine two

Well, it won't be worse than Xeon; further more, if the second core can't be used (because the software isn't multithreaded), that core wn't need a lot of bandwith either If the core can be used, its gonna help regardless of bandwith, just don't expect 2x performance.

> Second, the only type of RAM, DDR2, that could provide
>this bandwidth is expensive and has horrible latency at
>high speeds

Again, won't be worse than either Xeon DP/800 or a single P4 on single threaded apps.

>Dothan's performance is as great as the best P4, runs at an
>extremely low 27W of power and has the overhead (in terms
>of power) available for 64-bit, NX, etc.

Yeah, we would all want a 20W, 2.8 GHz dual core Dothan, with AMD64, NX, integrated memory controller, HTT, FB-DIMM, Itaniums FPU's and Power5's FSB.. and ..maybe toss in VIA's embedded encryption engine as well Oh, and for no more than $200 Did I mention 9Mb L3 yet ?

Its not like intel had a lot of options, 64 bit just isnt ready for Dothan yet, and a high end, dual core chip that does not support that is going to be a hard sell once windows X64 ships

In all fairness though, this would make a great chip for a cheap workstation. its just not gonna be a gamer chip, thats for sure.

= The views stated herein are my personal views, and not necessarily the views of my wife. =

 

[P4Man]

>on the fly clock speed adjustment, IC wait,stall,hybernate
>states.

All of that is great... when you are not feeding the core with instructions ! When you need the power, none of this is going to help squat.

>As for software performance it will be on par with...
>Xeon

Agreed. But it will be harder to cool since it will be a fairly impressive ammount of power in a tiny package...

= The views stated herein are my personal views, and not necessarily the views of my wife. =

 

[Xeon]

For one, a single Scotty core eats bandwidth like no other, imagine two.

The Prescott’s core logic bundles data even if it's not double precision data, bandwidth issues will be moot with considerations P4's run on a quad data rate memory subsystem, CPU1 takes rise fall of the 1/2 tick CPU2 gets the rise fall of the second 1/2 of system tick. In theory if both processors keep normal busy, ei: game there should be no real world performance loss.

Now with regards to heavy memory access software such as encoding or synthetic benchmarks is where the restrictions will become quite apparent. But Intel has shown they do very well with threaded vectorized code thus far, with known information it's likely not to change.

It will also mean poor multi-threaded scaling as opposed to what AMD could do.

I beg to differ Intel has put up a very good fight IMO with even the Xeon's bandwidth starved, IPC crippled, and lack there of a on die memory controller. With Intel's current HT technologies showing good performance with good code, I would have to say that Intel should in theory be on par or slightly ahead of AMD in threaded applications. Until we see NVIDIA’s Nforce5 chipset which will undoubtedly breath life into the performance starved Intel camp, it's anyone’s guess how/if memory will be an issue with smithy.

The power dissipated is also astronomical at all three clockspeeds and will likely only get worse since these per-core-power-dissipation numbers are among Intel's lowest thought possible (55-60W is low for a Scotty).

Final thermal output and electrical draw have yet to be "finalized". I know Intel stated worse case scenario draw and heat output but the race isn’t over yet.

Looks like another ploy by Intel to exercise their marketting department's dominance of AMD's. But for people looking for a processor, AMD's dual core solution (although later and likely more expensive) was built from the ground up for dual core and will really be the only solution.

Again without real world performance numbers there is no winner or loser at this point. AMD may be faced with production issues long before they ready the release of the FX and A64 lines for dual core operation. They have trouble meeting market demands as it is, doubling processors per die without double production capacity will equal market shortages. Sure they will be fast as snot, but no one will have those, makes it very moot in the end.

All of that is great... when you are not feeding the core with instructions ! When you need the power, none of this is going to help squat.

If code worked in that manner yes but it is sequenced and ordered, 100% CPU usage isn’t a real world situation unless you have some sort of a virus or the code ends up in a infinite loop. Now will there be more "power" in a smithy vs. K8 not likely, the K8 will enjoy its IPC dominance yet another generation.

Agreed. But it will be harder to cool since it will be a fairly impressive ammount of power in a tiny package...

I don’t see that being a real issue either double the silicon on a 200 watt spec heat sink just means the heat sink is operating at its max specifications. As for power draw ya it’ll be pretty sick but the race isn’t over yet. Intel might pull another rabbit out of the hat similar to the "j" series.

Xeon

<font color=red>Post created with being a dickhead in mind.</font color=red>
<font color=white>For all emotional and slanderous statements contact THG for all law suits.</font color=white>

 

[Mephistopheles]

I think you assume too much. I mean, I understand your concerns, and while they do seem to apply from what we know at this point, we cannot presume to foresee the future.

For one, a single Scotty core eats bandwidth like no other, imagine two. Second, the only type of RAM, DDR2, that could provide this bandwidth is expensive and has horrible latency at high speeds (newer modules can run 3-2-2-8 at 266MHz, which isn't shabby, but still isn't enough bandwidth and still costs a fortune). Sure, Scotty doesn't care THAT much about latency, but I have <b>a feeling </b>that the way the arbiter unit on Smithfield will work will be in a way that that won't effectively cache

While I see your point, I don't think smithfield will be tremendously bandwidth starved - like someone else here already said, it can't be worse than current Xeon platforms. And if the arbiter chip is well designed, Intel might manage. And have you ever thought that AMD will also share only one memory controller, as sophisticated as that controller might be? This <i>will</i> also have an impact - although probably much smaller - on single-core performance... perhaps comparable to going back from s939 to s754 processors.

Also, I find it funny that you're basing your predictions off a <b>feeling</b> you've got... I mean, what were I to predict if I did the same thing and was feeling constipated? Oh boy

 

[Mephistopheles]

>on the fly clock speed adjustment, IC wait,stall,hybernate
>states.

All of that is great... when you are not feeding the core with instructions ! When you need the power, none of this is going to help squat.

Ahhh yes, I was almost getting afraid that noone else noticed that. I don't care about idle temps, I care about 100% CPU usage temps!!! Thanks for reminding me of that point.

The only scenario I could think of is if you're watching a DVD and you want your CPU fan to slow down for less or no noise, but that's about it. And also, if at all possible, it would be best if the damned system was cool and quiet<i>all the time</i>, even when being stressed out!

>As for software performance it will be on par with...
>Xeon

Agreed. But it will be harder to cool since it will be a fairly impressive ammount of power in a tiny package...

Hm, yes, but that already tells us some things... I mean, a dual-core A64 will probably not be faster than a similar-clocked 2-way Opteron system. But currently, Opteron is quite superior to Xeon, which leads us to believe that, barring minor fluctuations, the same pattern will more or less repeat itself. So what we actually needed to be doing is trying to list comprehensively what is going to be different. So what sets dual-core apart from traditional dual-cpu workstation setups? What will change?

- Smithfield will support Dual DDR2-667, not DDR2-400 like Xeon. (could be a big difference, who knows)
- Intel's current dual-cpu NBs will be replaced by the arbiter chip. Changes to performance due to this change are <b>unknown at best,</b> but will probably not exceed a few percentage points from standard Nocona setup.

There are many other factors. Firstly, both Intel and AMD will put updated versions of their cores, and those might feature better performance or more features (read: SSE3 for Toledo). But in any case, Intel has the lower hand now and it won't be easy. What I can't quite grasp is why they don't introduce smithfield with 1066Mhz FSB... thermals, perhaps?...

Anyway, anyhow, let them fight it out and let us have the best of both.

 

[Mephistopheles]

Until we see NVIDIA’s Nforce5 chipset which will undoubtedly breath life into the performance starved Intel camp, it's anyone’s guess how/if memory will be an issue with smithy.

You're quite right... Personally, I'm hoping nForce5 can actually make a true difference - maybe even before i945 and i955 show up. I wonder if Nforce5 supports Smithfield already?... It does support dual-channel DDR2-667, at least, which is probably the turning point for DDR2 memory... hopefully, at not too high a price.

 

[P4Man]

>The Prescott’s core logic bundles data even if it's not
>double precision data

Sounds like you have no idea what you are brabbeling here. What does "logic data" have to do with "double precission" ? Exactly: nothing.

> bandwidth issues will be moot with considerations P4's run
>on a quad data rate memory subsystem, CPU1 takes rise fall
>of the 1/2 tick CPU2 gets the rise fall of the second 1/2
>of system tick.

LOL. If that where true, then there would be no difference between a quad pumped FSB (4x200) and a double pumped FSB (2x200) on a single P4. Well, I suggest you think again.

>ame there should be no real world performance loss

Here I agree, there won't be any significant loss over a single core cpu, just because an idle core doesnt require or consume any bandwith. And when it does (implying threaded code, or severall active threads), a second core will help getting the best execution rate for the available bandwith. The only cave at I see is cache trashing, I don't know how Smithfields cache will look like (unified L2 or not ?). IF it is unified, some code might suffer pretty bad, just like it did on early HT implementations for the same reason. But HT as been around long enough for compilers to be aware of this, so its probably not going to be a major issue.

> Until we see NVIDIA’s Nforce5 chipset which will >undoubtedly breath life into the performance starved Intel >camp,

What is special about nForce5 that you expect a significant performance gain from it ? AFAICT, intel is second to none when it comes to desinging memory controllers. I just don't think there is a lot of room for improvement there, let alone nVidia would be able to do it. nForce1 & 2 only rocked so much because every other K7 controller sucked so badly.

>I don’t see that being a real issue either double the
>silicon on a 200 watt spec heat sink just means the heat
>sink is operating at its max specifications

Oh common. You will probably also claim Prescot has no heat issue, and does not require huge noisy fans.. Yes it can be done, but that doesn't mean such high wattages, especially the way intel specs their cpu's, are desirable. Further more, I highly suspect the chip will implement the powermanagement features to keep it from exceeding the spec, which simply means throtteling. Dual core Prescott just doesn't sound like a good idea to me, at least not for a consumer product (might be a nice low end workstation chip though)

= The views stated herein are my personal views, and not necessarily the views of my wife. =

 

[P4Man]

>The only scenario I could think of is if you're watching a
>DVD and you want your CPU fan to slow down for less or no
>noise, but that's about it

Actually, 99% of the time at least one core will be iddle, so having a Cool&Quiet/Speedstep whatever equivalent certainly is a good thing. It just doesn't help reducing HSF or PSU requirements, since you also want the chip to work at 2x100% load

>Hm, yes, but that already tells us some things... I mean, a
>dual-core A64 will probably not be faster than a similar-
>clocked 2-way Opteron system

Who's to say ? AFAICT, AMD is not power limited at all, so even adding a core might enable them the same clockspeeds. AMD chips are clocklimited for other reasons. The same can not be said from intel, they seem very much power limited, so doubling the cores will hurt their clock speed (on P4 obviously).

Consider Dothan; adding a second core to it, is not likely to have much impact on maximum clock frequency. Well, of course statistics come it, if you measure the average maximum overclock of 2 chips at a time, its likely to be less than for one, but not by much.

>- Smithfield will support Dual DDR2-667, not DDR2-400 like
>Xeon. (could be a big difference, who knows)

Are current xeons *still* single channel ? If so, expect a very nice boost over Xeon DP.

>- Intel's current dual-cpu NBs will be replaced by the
>arbiter chip.

Ahem.. no. From what I read, first smithfields will not feature an arbiter chip, hence presenting themselves as 2 distinct cpu's to the NB. That is exactly the same configuration as Xeon DP. Now, adding an arbiter (which they will later), might *speed up* performance over a non arbiter or DP configuration, by making more efficient use of the FSB bandwith. By how much is hard to tell, but I wouldn't expect it to be huge. Personally, I think a bigger downside is that initial Smithfields will be seen as 2 *physical* cpu's by the OS and software, meaning, more expensive licences. HT enabled chips are mostly seen as just 1 cpu, but without arbiter chip, there is no way to differentiate a dual core from a dual cpu setup.

> both Intel and AMD will put updated versions of their
>cores, and those might feature better performance or more
>features (read: SSE3 for Toledo)

I'm sure that will make a whopping 0.5% difference

>What I can't quite grasp is why they don't introduce
>smithfield with 1066Mhz FSB... thermals, perhaps?..

Not likely. Signal integrity on the MB seems like a better bet. I mean, why not release 1600 or 3200 MHz FSBs while you are at it ?

= The views stated herein are my personal views, and not necessarily the views of my wife. =

 

[Xeon]

Sounds like you have no idea what you are brabbeling here. What does "logic data" have to do with "double precission" ? Exactly: nothing.

Vectorized code such as SSE and SSE2 run-in bundles 128bit, Intel added and widened this feature for the Prescott core, where the processor tries to alleviate stress off the memory subsystem by bursting the data whether it be to a GPU, IDE controller or memory. Ill get you a link since you obviously missed that tidbit of information.

LOL. If that where true, then there would be no difference between a quad pumped FSB (4x200) and a double pumped FSB (2x200) on a single P4. Well, I suggest you think again.

Did I miss something? A P4 receives and sends 4x a clock tick so no 2x200 would not be the same as 4x200, to be sure I am amazed by what you said because it makes no sense what so ever.

But HT as been around long enough for compilers to be aware of this, so its probably not going to be a major issue.

Yes most of the ground work for this processor has been done I would have to think it would be a fairly smooth transition with regards to compiler and OS support. As far as I can tell all existing HT optimized code will work “ideally” on smithy.

What is special about nForce5 that you expect a significant performance gain from it ? AFAICT, intel is second to none when it comes to desinging memory controllers. I just don't think there is a lot of room for improvement there, let alone nVidia would be able to do it. nForce1 & 2 only rocked so much because every other K7 controller sucked so badly.

Yes then I come back to the N-Force 4 chipset, does good for AMD. N-Force 5 should be just as interesting I would think.

Oh common. You will probably also claim Prescot has no heat issue, and does not require huge noisy fans.. Yes it can be done, but that doesn't mean such high wattages, especially the way intel specs their cpu's, are desirable. Further more, I highly suspect the chip will implement the powermanagement features to keep it from exceeding the spec, which simply means throtteling. Dual core Prescott just doesn't sound like a good idea to me, at least not for a consumer product (might be a nice low end workstation chip though)

I never said the Prescott didn’t have design related thermal issues in fact I never mentioned anything in relation to that, last I recall I was talking about heat sinks spec’d for 200watts thermal output. But whatever you like to hear yourself talk.

Actually, 99% of the time at least one core will be iddle, so having a Cool&Quiet/Speedstep whatever equivalent certainly is a good thing. It just doesn't help reducing HSF or PSU requirements, since you also want the chip to work at 2x100% load

So why did I pay for XP if 99% of the time my second core will remain idle, It doesn’t work that way with HT enabled P4’s goes back and forth. I don’t see it changing with smithy.

Who's to say ? AFAICT, AMD is not power limited at all, so even adding a core might enable them the same clockspeeds. AMD chips are clocklimited for other reasons. The same can not be said from intel, they seem very much power limited, so doubling the cores will hurt their clock speed (on P4 obviously).

How does power even support a argument that a dual core K8 would be about the same speed as a 2x Opteron. Also we have all read the clock speeds for dual cores so what further points are you trying to make.

Are current xeons *still* single channel ? If so, expect a very nice boost over Xeon DP.

Been dual channel for quite some time.

Ahem.. no. From what I read, first smithfields will not feature an arbiter chip, hence presenting themselves as 2 distinct cpu's to the NB. That is exactly the same configuration as Xeon DP. Now, adding an arbiter (which they will later), might *speed up* performance over a non arbiter or DP configuration, by making more efficient use of the FSB bandwith. By how much is hard to tell, but I wouldn't expect it to be huge. Personally, I think a bigger downside is that initial Smithfields will be seen as 2 *physical* cpu's by the OS and software, meaning, more expensive licences. HT enabled chips are mostly seen as just 1 cpu, but without arbiter chip, there is no way to differentiate a dual core from a dual cpu setup.

What the hell are you talking 1st gen smithies are equipped with the arbiter chip, the arbiter chip controls bus transactions, current Northbridge’s are not equipped to deal with 2 chips on one bus. With perhaps exception with the 925, 925XE respectively, hence in 2006 they will be introducing dual next gen Northbridge’s configurations, and please notice these will not be the current 915 and 925 chipsets but the 945, 955 respectively, which will boast serial ata2, 667 DDR2, 1066 fsb and some more wireless jazz.

But bandwidth isn’t the real concern for those processors at this point, thermal output has to be managed.

As per the licensing issues for dual core systems, I honestly don’t know if MS has set something in stone for that I have yet to read it. I would have to assume regardless that MS will treat it like it did with HT, since there is going to be upwards of 7 different versions of longhorn.

I'm sure that will make a whopping 0.5% difference

Personally 3-7% is more realistic of a theory, with regards that now there are 2 independent SSE/SSE2/SSE3 engines if given fast shifting with minimal latency from CPU to CPU should prove to be most beneficial.

Not likely. Signal integrity on the MB seems like a better bet. I mean, why not release 1600 or 3200 MHz FSBs while you are at it ?

Last I checked the 945, 955 chipsets would support 1066, I would have to assume with the C0 stepping of the ICH6 rolling out that it was just easier for them to roll that out see how it works and move along to the ICH7 which would have to be released with the next generation chipset. All respective timeframes work out in the end for Intel.

Xeon

<font color=red>Post created with being a dickhead in mind.</font color=red>
<font color=white>For all emotional and slanderous statements contact THG for all law suits.</font color=white>