Only a 5% overclock on the T-Bred?

[Smokey]

*buddry ducks low, narrowly avoiding the huge dose of sarcasm whistling right past his head*

Sorry, next time I will reference the GREEN and RED colors discussed earlier in the thread.

(Christmas is something we call a "holiday", and a very popular one here in the US which is usually represented with those two colors)

Life's a hole...dig it. - Joe Dirt

 

[FatBurger]

<i>xazos79 says:</i>
Wouldn't that be "drove my chevy to the levy" ?

Yeup. I thought "to" and wrote "by" somehow. Thanks

<i>Eden says:</i>
BTW since when has this place stopped being technical, Fat?

Sorry? I'm just sitting on my ass at work, did you want something? :tongue:

<i>Eden says:</i>
I say THG should have tried the XP1700+ too, as it uses 1.5V.

Since when does the 1700+ run on 1.5v?

<i>Corona999 says:</i>
I probably wouldn't have become a computer enthusiast if it wasn't for AMD and I'd probably be out having fun at parties rather than debating about chip wars on forums.

Wait...so why is AMD such a great company again? You've made me forget.

<i>buddry says:</i>
AMD will not dominate the Christmas buying season. IF hammer is released by then, I doubt it will be in large enough quantities that it will totally dominate anything.

I agree, Hammer will be a small, but important release.

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

 

[AMD_Man]

Since when does the 1700+ run on 1.5v?

The lower-end T-Bred use 1.5V. Read around. I believe Sharkey Extreme talked about it.

<b><i>"A penny saved is a penny earned!"</i></b>

 

[cellbiogeek]

I thought, according to Frank and Bert's article, that the three lower clocked T-breds ran at 1.5 volts.
<A HREF="http://www.tomshardware.com/cpu/02q2/020610/thoroughbred-08.html" target="_new">http://www.tomshardware.com/cpu/02q2/020610/thoroughbred-08.html</A>

 

[xxsk8er101xx]

depends ... i think you can split that type of additude 50/50

<A HREF="http://www.anandtech.com/mysystemrig.html?id=9933" target="_new"> My Rig </A>

 

[xxsk8er101xx]

naw... everyone is gonna buy dvd+RW drives and new video cards

<A HREF="http://www.anandtech.com/mysystemrig.html?id=9933" target="_new"> My Rig </A>

 

[bront]

<A HREF="http://www.amdmb.com/article-display.php?ArticleID=187&PageID=4" target="_new">Amdmb.com</A>

Apparently, his CPU was not unlocked like some review sites were either. I don't know how valid his results are either. I think we'll see about a month after they get out if there are any good OCers out there.

"Search your feelings you know it to be true, I am your... twin sister" - Darth Vader

 

[bront]

THG mentioned that they rearanged the core to make it easy for them to add cache for Barton.

As for the resources, It makes sense to me, but that doesn't make it true.

"Search your feelings you know it to be true, I am your... twin sister" - Darth Vader

 

[eden]

Well that didn't take long before they showed you 1.5V!

My point was that here we not just discuss what's best. If Tbred is OCing bad, we don't just go stubborn just because it isn't, instead we are trying to see WHY. I said that this place is technical and thus it's not about a 2-man vs entire forum war, on defending the Tbred.

Besides, it doesn't matter the K7 being at its end, it's about some logic that makes no sense. How can a 0.13m chip not clock higher than a 0.18m! It's like saying a 400LB guy can make the marathon faster than a 150LB guy. Granted it is possible, but under logical and normal circumstances, it just wouldn't work.

Anyway more benches with other Tbreds are needed and again, why are we OCing the high end chips, those that are probably at the end of the K7? Why not those who were in the 3/4 like the XP1700 0.13m?
It's like taking the last of the NWs, say 4GHZ, and OCing it. Oh yeah expect a lot....


--
Meow

 

[FatBurger]

The lower-end T-Bred use 1.5V.

Oh, I knew that. It's not released yet, which threw me off. Don't mind me.

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

 

[eden]

Didn't have your coffee yet?

--
Meow

 

[SammyBoy]

<b>SammyBoy says</b>:
I'll give you a simple explaination for the poor OC ability of the T-Bred.

The K7 is on its last leg. It's an old core, soon to go the way of the K6 and K5.

It doesn't overclock well cause the core is too small for thermal transfer to occur at the level that it did with the Palomino.

The red vs green debate can be solved easily as well. Just like any other core that has come into existance, its safe to say that a few revisions have occured since the T-bred was initially built. Remember AXIA vs. AVIA (or whatever the "inferior" oc'er was)? The red CPUs might represent the initial batches the were run that operate at their speced speed. Just what they are supposed to do. There is nothing "wrong" with them, they are just a less refined version of the T-Bred core. T-Breds, regardless of reports, must have been running off the fab lines since January or earlier, so that AMD would have enough to fill the OEM channel and have some leftovers for the retail channels. But, the flip side is that there have been mobile versions since April or so, which run at lower clockspeeds, and theoretcially could be where the initial batches went.

AMD doesn't give a damn about the K7 anymore. It's budget, and they only have 2, maybe 3 speed bumps for the T-Bred and Barton before the K8 appears. After that, the K7 will be relegated to the budget/mainstream market, and after a few more months, will be retired completely when the clockspeed ceiling is reached for the budget line. A budget K8 will replace it, probably around the time of the .09 micron intro, and the wonderful K7 and Socket A will vanish from the minds of performance freaks. Why would AMD invest more capital and R&D into a dying core when they could be spending those man-hours revising the K8 into some uber-CPU, capable of ripping holes in the fabric of time? It makes no sense, either from an engineering perspective, or a sales/marketing perspective.

It's dying, prepare the memorial service, and give it a decent burial instead of saying what it should be or what it will be when you have a different colored packaging. Just give it up, guys. Spend your time working more hours to buy everything needed to get a Hammer or Prescott system.

I'm reposting because it was apparently overlooked. The point is, even a 1700+ at 1.85v (+.35v over stock) will only get to maybe a 2000+ speed, as Anandtech shows <A HREF="http://www.anandtech.com/cpu/showdoc.html?i=1635&p=14" target="_new">here</A>. The 1800+ made it to 1813MHz and the 1900+ got to 1875MHz. Yes the packaging reveals that the cores are possibly older, but it could also mean that AMD used the rest of the old packaging on the T-breds and only very recently exauhsted the supply and move to the new packaging. Seems to me that the ceiling of the K7, in it's current form, is about 1.9GHz to 1.95GHz. Not quite the 2GHz, but close. It overclocks like crap, regardless of the die shrink, because there is nothing left, MHz-wise, in the K7.

-SammyBoy

 

[eden]

Once again there is something deeply wrong here, and it worsens the matter with the XP1700+.

1)It is 1.5V. The XP1700 0.18m is 0.20V higher.
2)The 0.18m version went up to XP2100 speeds on aircooling, ask Matisaro's system speccs.
3)At 1.5V, it should do WAY more than what Mat's OC 0.18m does. Especially considering the 1.47GHZ is not the end zone at all.

So once again, more and more weird things appear. How can a chip 0.2V less not OC as good as the 0.18m, seeing as it's a step not downbinned, and has been converted to 0.13m? Makes ABSOLUTLY no logical sense any more than the XP2200+.

--
Meow

 

[LED]

Why cant a palomino hit 2ghz? Why cant a P3 hit 2ghz? You can find out why the T-bred OC's like crap and in the end you will still be left w/ a proc that OC's like crap.

I sold my sig for $50.

 

[lhgpoobaa]

u realise that downbinning has alot to do with overclocking... why do u think that 1.6A and 1.8A's are so good?

im sure the tbred XP1700+ @ 1.5v will be pretty stellar too when it arrives shortly.

give the tbred core time to mature.

<font color=green>Hamsters</font color=green> <font color=red>don't</font color=red> <font color=green>play</font color=green> <font color=red>Soccer.</font color=red>

 

[LED]

NW's w/ higher multipliers can be OCd too, not as good percentage wise as the 1.6/1.8, but they OC....So whats your argument? 2.5s are downbinned too? I dont get what youre trying to say. You can probably OC the 1700 tbred to where the 2200 tbred OCs to. Still nothing special. The AXP hit its limit.....

I sold my sig for $50.

 

[lhgpoobaa]

i just think u should wait for a slightly larger statistical sample than 1 batch and a few cpu's before making grand claims.


<font color=green>Hamsters</font color=green> <font color=red>don't</font color=red> <font color=green>play</font color=green> <font color=red>Soccer.</font color=red>

 

[Matisaro]

But still not fact? "I just dont think so" proved to be better than your educated guess....

At the time the "i dont think so" had no predicative weight.



The Cash Left In My Pocket,The BEST Benchmark

 

[Matisaro]

2)The 0.18m version went up to XP2100 speeds on aircooling, ask Matisaro's system speccs.
3)At 1.5V, it should do WAY more than what Mat's OC 0.18m does. Especially considering the 1.47GHZ is not the end zone at all.

My overclock is nothing special, locked 1700=@150fsb, if I unlocked it given my temprature I would expect around 1.9ghz, but then again my guesses arent spot on this week apparently ;-).

The Cash Left In My Pocket,The BEST Benchmark

 

[Matisaro]

I agree poobah, I advocate waiting and not using results from possibly questionable(red/green) chips before saying anything definitive.

The Cash Left In My Pocket,The BEST Benchmark

 

[texas_techie]

Im gonna stick with my prediction the T-bred is gonna suck at OC for a while. A die shrink doesnt automatically translate into higher overclocks. A possible reason is this (im no expert so be patient):
The electrons within the cpu have to arrive at their destination within a couple of nanoseconds of a spec. AMD shrunk the die, but didnt change anything architecturally (no improvements). So that means you have to retime everything, cuz all the electrons are hitting their destinations sooner (shorter distance to travel). So what happens when we overclock? We change the FSB or the clock multiplier. And a part of what the multiplier interacts with is the quartz clock that regulates the cpu.
AMD had issues with the die shrink. Part of it was timing issues (trust me on this one). So if ya go jacking with the clock, It upsets the timings. One way around this is ramping up the voltage. This literally forces the electrons to go where ya want them. However, if the tolarances are too narrow, then even a voltage hike wont help an overclock.
So therefore, I draw the conclusion the silicon cant tolerate overclocks because AMD hasnt figured out how to lay the traces efficiently/effectively (on the new process). Secondly, no matter how hard you try.. a core will only go so far. It doesnt matter if its .13, SOI and SSE 8,000,000... A core has a lifetime, period. I think we are starting to see the end of Palamino.
I do think with time AMD will get better at .13 and we will see faster T-breds. But i dont see it happening now.
My 2 cents. I could be wrong, its just a guess.

Benchmarks are like sex, everybody loves doing it, everybody thinks they are good at it.

 

[eden]

You might be right there...
However it still will not explain, how:
A 1.47GHZ CPU at 0.18M OCed as high as 1.73GHZ Aircooled at 1.75V
A 1.47GHZ CPU at 0.13m OCed as high as 1.53GHZ on same voltage of 1.5V, or 1.66GHZ max on 1.85V, yes 1.85V!

How can that be explained? It is NOT the end of the road for a low clocked one. There is no reasonable explanation like that for it.

--
Meow

 

[SammyBoy]

Actually, texas_techie explains it perfectly. Reread his post. You'll notice he explicitly mentions raising core voltages as not being a perfect solution, and combined with timing issues, all overclocking does is F things up more. Give the core time to mature. I'd say that by the time the Barton appears, the K7 will at least be able to overclock to 2500+, maybe 2600+. Beyond that, no. The core is too old, and AMD doesn't have the time or budget to invest in a dying technology. Remember, it's Hammer time.

-SammyBoy

 

[eden]

Ok yeah, I was more replying to his claim of the end of the core, but seeing as it is still 1.47GHZ.

Actually maybe Mat was right on the core layout being possibly bad, which makes texas' assumption also credible.
When you think about it, the Tbird 1.4GHZ was almost the end of the road back then, no? AMD instead made a special refinement and layout, which added a good ~28% higher clock speed under the same process. Now AMD has switched to 0.13m, they have made a new layout. When you think about it, they may have reset the clock on the Tbird days, and removed the special refinement the Palomino made, kind of like removing a patch off a program and trying to then push it further by making an updated version. So basically it's as if the 0.13m process should have began at 1.4GHZ, and ended around the 2GHZ line, and thus would explain why suddenly the refinement removed, the Tbred isn't doing much good without that "patch".

It might sound confusing but in the end it's to agree on Mat's claim. But yeah, I agree we should wait. It's AMD's first 0.13m process, Intel has had 2 chips under it, so some are really being hard asses on AMD now.

--
Meow

 

[SammyBoy]

Well, I think that AMD redid the layout as quickly as the could, without investing too much R&D, while still leaving room for an extra 256kb L2 to be thrown on later. I think Ray was right to say that the die looks more like it was laid out for future additions instead of optimal performance. They are cutting corners with the K7, a symptom of a dying technology that they need to stretch out a few more months.

One reason why the 1700+ OCs poorly might be the fact it isn't downbinned at all, and is in fact running at it fastest clockspeed without extreme methods (i.e voltage increase). Now they might be able to downbin, if yields are good enough, but maybe not when the review samples were made. I'd say give AMD a few months to iron out the kinks, but considering the Barton will probably appear around Aug/Sept, and the Hammer, I hope, in Oct/Nov, there really isn't much time for the T-bred. It's very true that the T-bred is a transition core, and nothing more. AMD is probably hoping to be able to get one, maybe two more speed increases out of it before the Barton, and one or two more after that before the Hammer appears. I have read rumors of Bartons already sampling, though no benches. I still think that I personally will be looking to see how the Hammer develops, and maybe pricing info. That will ultimately decide my next upgrade, though I don't want to, but my motherboard is seriously limiting my Ti4400 performance, both in terms of poor AGP drivers and support, and lack of support for the AXP. Damn it!

-SammyBoy