Why Barton will be a better OC'er

[Copenhagen]

Reading various OC reports on the new T-bred seems to indicate that heat is it's Achilles heel. This is not exactly what one would expect, as a die-shrink normally results in lower temperatures. And the T-Bred actually does have a lower power-consumption than the Palomino, but it also has difficulties in getting rid of the heat, due to the reduced surface of the small core and lack of heat-spreader. This is why I believe that the added L2 cache of Barton, which will result in a larger core, will allow for better heat dissipation, and thus, higher clock-speeds than the T-bred is currently showing.

Notice how extreme cooling helps the T-bred to reach very high frequencies.

Wonder if the large core of the P4 was something Intel designed "on purpose", in order to improve heat dissipation at very high frequencies.

<i>/Copenhagen - Clockspeed will make the difference... in the end</i>

 

[eden]

Personally I was also predicting Barton's OCing will be better. The core design and layout will be improved and maybe AMD can do a Palomino-style rearrangement, resulting in lower temps and possibly lower voltage?
But I am more positive that Barton will have better OCing.

--
Intel and AMD sitting under a tree, P-R-O-C-E-S-S-I-N-G!

 

[tommygun]

You may be right about this.

I have used an AthlonXP 1600+ and a Duron 1000(Morgan) on almost (but not exactly) the same system and the Duron was hitting higher temps.
I believe the analogy might be similar with the Duron having less cache, although I haven't really checked the die sizes.
So maybe the smaller die size is really the issue with the Tbred.

I wonder what Texas Techie & Matisaro might have to say on the subject.

 

[castle]

I donot think so.
CPU with larger cache normally donot oc well, though perform better at the same frequency.
Besides, AXP has quite large 128K L1, so doubling its L2 wouldnot help the performance as efficiently as it does to P4.

 

[IIB]

the bigger L2 cache will boost higher Cache-Hit rates and thus the new processor will access the memory less.
this should bring down temps becouse memory access creates an effect of Capacitance and thus lifts core tempertures.


This post is best viewed with common sense enabled

 

[bront]

the P4 L1 and the AMD L1 actualy work differently, so it might give more of a benifit than you think.

Also, if CPUs with a larger cache don't OC well, then how do you explain the P4 NW overclock compaired to the P4 Willy?

The cache should have little to do with the OC by itself, however, with the increased surface area, that should help heat disapate better, which could help.

So, better heat disapation + lower initial clock on the same basic architecture should = better OCability.

One other thought. It's also posable that the extra cache on the CPU could help with heat by simply not letting the CPU heat up as much because the CPU works more efficiently. Not sure if that would work, just a curious thought. (Less heat because the CPU does less to get the same ammount of work done, so less electricity flowing through places where it can heat up.)

English is phun.

 

[FatBurger]

Also, if CPUs with a larger cache don't OC well, then how do you explain the P4 NW overclock compaired to the P4 Willy?

Because it was a die shrink. Tbred -> Barton will not be. And he's right, processors with more cache typically OC less.

<font color=blue>Hi mom!</font color=blue>

 

[M_C_Hammer]

"This is why I believe that the added L2 cache of Barton, which will result in a larger core, will allow for better heat dissipation"

this makes no sense, sure a larger surface area dissipates heat better, but the extra core area (the extra 256 kb of cache) will produce its own heat surely? adding to the total heat output, therefore the heat dissipated per unit area will remain the same. There is more heat but more surface area to compensate.

You are assuming that the extra 256kb of cache doesnt produce any heat.

I predict you are wrong and we will be disapointed once again, but i really hope you are right.

I need a 1.5 Ghz Athlon + 512mb ddr ram to write emails......honestly

 

[Copenhagen]

<blockquote><font size=1>Svar på:</font><hr><p>I donot think so.
CPU with larger cache normally donot oc well, though perform better at the same frequency.
Besides, AXP has quite large 128K L1, so doubling its L2 wouldnot help the performance as efficiently as it does to P4.<p><hr></blockquote><p>You missed my point, which was that the T-Bred core seems to have problems dissipating heat, due to its small size. A heatspeader or a larger core surface would improve heat dissipation.

About whether large cache OC well or not: Look at the Northwood, it doesn't seem to have any problems in that respect.


<i>/Copenhagen - Clockspeed will make the difference... in the end</i>

 

[Copenhagen]

<blockquote><font size=1>Svar på:</font><hr><p>You are assuming that the extra 256kb of cache doesnt produce any heat.<p><hr></blockquote><p>You are assuming that L2 cache generates the same amount of heat per unit area as, say, a FPU unit or an ALU. I'm almost sure that it is not the case.


<i>/Copenhagen - Clockspeed will make the difference... in the end</i>

 

[IIB]

I dont like the fact that evryone disregarded my post about the Barton being cooler as a result of a highe Cache-Hit rates and less access to memory per amount of time.


This post is best viewed with common sense enabled

 

[M_C_Hammer]

think about it logically, heatsink would be closer to core than extra cache, heat from core is not magically going to transfer into the extra area cache provides.

I need a 1.5 Ghz Athlon + 512mb ddr ram to write emails......honestly

 

[FatBurger]

Just because nobody commented on that directly doesn't mean everybody ignored it.

<font color=blue>Hi mom!</font color=blue>

 

[castle]

Don't compare apple and orange. T-Bred=> Barton is just (or mostly) L2 change, while Willemette => Northwood is so much more.
Plus, if Athlon architecture won't go high freq at all, nothing could help oc.

 

[Kelledin]

You are assuming that L2 cache generates the same amount of heat per unit area as, say, a FPU unit or an ALU. I'm almost sure that it is not the case.

Cache is actually one of the most active portions of the chip. Execution pipelines can and do stall quite frequently, but cache is almost always being accessed and updated, even when the CPU is "idle." So the extra die space used by cache will probably put out <i>more</i> heat...

<i>I can love my fellow man...but I'm damned if I'll love yours.</i>

 

[IIB]

but they didn't.

anyway Cache is hardly that active - it holds many many adresses and is required to store/load only a few of them in the same clockcycle + it has a latncy of a few clockcycles in which only the controling mechminzem of the cache works. the cache is self - which is the larger part - has really low usage...



This post is best viewed with common sense enabled

 

[Phob]

i have a BIG question here...

Will Barton outperfor thoughbred so much, to even wait before buying, lets say, an athlon xp 2200???

 

[IIB]

most esstimates go for around ~7% improvement.
and my guess is that the desktop version of the barton would lunch at the begining of Q4' at the earlyest.
then you'd be wanting to wait for somthing else..

This post is best viewed with common sense enabled

 

[Quetzacoatl]

That's not a big question, yes and no, it will gain from the additional L2, but not as much as the Northwood, considering the Athlon is less sensitive to mispredicted branches. It won't have a large gain, but comparable to the Willy to NW

"When there's a will, there's a way."

 

[eden]

Personally I found your theory and fact to be very realistic, and I didn't disregard it at all, I just read it!

It makes sense that the more the helping components, the less the CPU has to heat up. I would suppose that the Willy has a large idle-load delta compared to NW's, but we'd need proof. So for one: The added cache will increase die space, therefore cooling is more reaching. The cache will also serve as less time to do an operation, thus less heat. In overall the delta should be reduced and the OCing just might be better. Hell we've seen recent Tbreds doing better and better in new steppings. Fugger has an interesting 2.6GHZ Tbred on LN2.(I really wish the guy would stop being so anti-AMD, he has really good info)Granted, it's extreme and no SSE, but it sure as hell shows the Tbred CAN get there soon, and that maybe Gigabyte's 2.6GHZ Tbred OC is not so farfetched. In any case, we'll just have to wait and see how Barton turns out, but for now, IIB's claim is pretty logical and if he's experienced in this, I guess his credentials are less to be debatable if someone who hasn't taken a course in such was disagreeing.

--
Intel and AMD sitting under a tree, P-R-O-C-E-S-S-I-N-G!

 

[juin]

Let the overhype start again

1 more cache do not result in less speed for the cpu(mghz)
2 Actual speed of a chip is a combination of many factor
R300 107 million transitor 350 mghz core speed 0.15 micron
Pharelia 512 80 million transitor 220 mghz core speed 0.15 micron

That off topic but prove that micron prosces alone is not the only factor for speed.

cheap, cheap. Think cheap, and you'll always be cheap.AMD version of semi conducteur industrie <P ID="edit"><FONT SIZE=-1><EM>Edited by juin on 06/25/02 11:49 PM.</EM></FONT></P>

 

[eden]

I dunno why an added cache would need less speed, I thought as we add things, we also increase clock?

--
Intel and AMD sitting under a tree, P-R-O-C-E-S-S-I-N-G!

 

[Copenhagen]

<blockquote><font size=1>Svar på:</font><hr><p>think about it logically, heatsink would be closer to core than extra cache, heat from core is not magically going to transfer into the extra area cache provides.<p><hr></blockquote><p>You certainly have a point there.

I don't know exactly how a CPU is build. Between the inner core and the heatsink there is some kind of protective material, but I guess its heat conductivity must be pretty good. It could however turn out to have less conductivity than the heat conductivity within the core itself. This makes the whole arrangement of the die very important. You would like to place very hot spots near cold ones, instead of hot spots next to other hot spots. I'm sure AMD has already optimized the die in this respect. Question is whether the added L2 cache will allow a better die arrangement with respect to heat.


<i>/Copenhagen - Clockspeed will make the difference... in the end</i>

 

[Copenhagen]

IIB, if I don't comment on a post it doesn't mean that I disregard that post. It could of course be the case, but in your case it didn't.


<i>/Copenhagen - Clockspeed will make the difference... in the end</i>

 

[baldurga]

Sorry, but I disagree. I would like to see Barton as a good overclocker, but right now Tbred has some problems, that, maybe, are due to lower contact with HSF so more difficult to transmit heat. Ok, more L2 cache can help. Now another point.

Barton will be initially out at 2400+ or something like this, isn't it? So bearing in mind that the frecuency will be higher and, from my point of view, there is an arquitecture limitation in the K7 core, there will be less room to overclock, simply because we will be near the limits of the CPU.

Also I don't think AMD will spend money and time to rearrange the core to allow higher frecuencies, because they are focused on Hammer.

For all this reasons I'm affraid that Barton won't be a good overclocker. Better we stay in the low spectations range, or we will be disappointed again.

DIY: read, buy, test, learn, reward yourself!