Amazing Bang-for-Buck? AMD Athlon II X4 631

[SickBeast]

I don't think I'd call ~17-18% faster "barely faster".

Also, you can't compare directly the CPU voltage of AMD and Intel on the same process when it comes to degradation. AMD CPUs are quite a bit more tolerant of voltage, and this should do 1.45V just fine for three or more years if it has a good cooler to keep temps low.

44/38=1.16, therefore 16% faster based on clockspeed alone. The L3 cache of the Phenom II will eat up a good chunk of that differential.

1.45v is a lot for a 45nm Phenom II. I don't think it's a wise move for something on a smaller process.

Like I said, we don't even know if that 4.4ghz CPU is stable. Perhaps someone will hit 5ghz with one but I highly doubt it.

 

[LOL_Wut_Axel]

44/38=1.16, therefore 16% faster based on clockspeed alone. The L3 cache of the Phenom II will eat up a good chunk of that differential.

1.45v is a lot for a 45nm Phenom II. I don't think it's a wise move for something on a smaller process.

Like I said, we don't even know if that 4.4ghz CPU is stable. Perhaps someone will hit 5ghz with one but I highly doubt it.

As I said in another thread, the L3 cache difference of the Athlon II was noticeable in gaming by around 10%. In file compression, +- 5%. In everything else, from 1-0%. So even if this was a normal Athlon II, at the very least it'd be a bit faster and on most scenarios noticeably faster. Llano has, again, double the L2 cache as the Athlon II to mitigate this difference. It also has comparable frequency scaling to the Phenom II.

Also, I don't know if you've actually used a Phenom II. 1.45V is about the average overvolt for 24/7 use, and 1.5V starts to push it. Since it's just a die shrink when it comes to that it should change very little.

 

[SickBeast]

I was under the impression that as the manufacturing process gets finer, the amount of voltage you can pump through the thing becomes less and less.

I was not aware of the L2 cache being doubled. Thank you for pointing that out.

In any event, 16% is not such a big deal to me.

Remember, the i5 2500K can easily hit 4.5ghz and will crush this new FM1 chip. It will only compete against the cheap i3s and stuff.

 

[Hypertag]

I was under the impression that as the manufacturing process gets finer, the amount of voltage you can pump through the thing becomes less and less.

This is correct, and will be extremely important to remember whenever Ivy Bridge is released

 

[LOL_Wut_Axel]

I was under the impression that as the manufacturing process gets finer, the amount of voltage you can pump through the thing becomes less and less.

I was not aware of the L2 cache being doubled. Thank you for pointing that out.

In any event, 16% is not such a big deal to me.

Remember, the i5 2500K can easily hit 4.5ghz and will crush this new FM1 chip. It will only compete against the cheap i3s and stuff.

Yes, but 1.45V should still be fine. The higher you go on voltage, the higher degradation gets. On 45nm for a Phenom II at 1.5V I'd estimate it'd be fine for 2 years or so before you need to raise voltage again, and 1.45V fine for over three years. On 32nm it's worse, but shouldn't degrade as bad as a Phenom II at 1.5V. I'm saying 1.45V as the max recommended for long-term, around three years. For five years 1.4V seems like a better bet. For Sandy Bridge we don't have exact numbers to compare, but 1.35-1.38V or so for three years seems about right.

If 16% isn't that big of a difference then by the same logic going with faster CPUs like the i5-2400 doesn't make that much sense, right?

Also, a 2500K is officially at $220, and this is a meager $90.

 

[Keysplayr]

Yes, but 1.45V should still be fine. The higher you go on voltage, the higher degradation gets. On 45nm for a Phenom II at 1.5V I'd estimate it'd be fine for 2 years or so before you need to raise voltage again, and 1.45V fine for over three years. On 32nm it's worse, but shouldn't degrade as bad as a Phenom II at 1.5V. I'm saying 1.45V as the max recommended for long-term, around three years. For five years 1.4V seems like a better bet. For Sandy Bridge we don't have exact numbers to compare, but 1.35-1.38V or so for three years seems about right.

If 16% isn't that big of a difference then by the same logic going with faster CPUs like the i5-2400 doesn't make that much sense, right?

Also, a 2500K is officially at $220, and this is a meager $90.

What are you basing all these estimates on? Especially when no two chips are exactly identical and the very reason binning is done. I just don't get how you can just randomly say that 1.5v on a PhenomII will last you 2 years before you have to........ etc. etc. etc.

 

[SickBeast]

What are you basing all these estimates on? Especially when no two chips are exactly identical and the very reason binning is done. I just don't get how you can just randomly say that 1.5v on a PhenomII will last you 2 years before you have to........ etc. etc. etc.

Overclocking is a bit of a science. A CPU is sorta like an incandescent light bulb. They will burn out over time. If you give them more voltage, they will burn out faster.

I don't know *how* people come up with the maximum safe voltage, yet they do, and it tends to work out ok. I have yet to hear about CPUs spontaneously combusting during their useful life cycle, overclocked or not. The only way people kill their CPUs is with a ridiculous amount of voltage, which varies depending on the CPU.

 

[LOL_Wut_Axel]

What are you basing all these estimates on? Especially when no two chips are exactly identical and the very reason binning is done. I just don't get how you can just randomly say that 1.5v on a PhenomII will last you 2 years before you have to........ raise voltage to maintain the overclock.

^^

 

[LOL_Wut_Axel]

Overclocking is a bit of a science. A CPU is sorta like an incandescent light bulb. They will burn out over time. If you give them more voltage, they will burn out faster.

I don't know *how* people come up with the maximum safe voltage, yet they do, and it tends to work out ok. I have yet to hear about CPUs spontaneously combusting during their useful life cycle, overclocked or not. The only way people kill their CPUs is with a ridiculous amount of voltage, which varies depending on the CPU.

True. Max "safe" voltage will depend, again, on how much you plan to keep it.

 

[SickBeast]

True. Max "safe" voltage will depend, again, on how much you plan to keep it.

In all my years posting on this forum, I have yet to hear about a CPU that died because it was overclocked and then used over a long duration of time. I'm sure it happens, but it seems to be quite rare, and by the time it does happen, the computer is so obsolete that no one even bothers to troubleshoot the issue. 99% of the CPUs of that type that were manufactured would probably be in a landfill site at that point.

 

[Vesku]

I can see your die-size argument for no L3 but based on current GF 32nm the power vs performance victory would probably go to a lower clocked Llano x4 with L3. I do think that a Black Edition Llano as it stands now with working GPU that can be disabled would be a nice addition for FM1 family.

 

[antisocialmunky]

I rather have more L2 cache and faster average successful memory access over potentially slower misses. TLBs have really high success rates these days so I'm tempted to say that the L2 will lower your average memory accesses compared to just having a crap ton of L3 and less L2.

http://www.tomshardware.com/reviews/athlon-l3-cache,2416-6.html

The disparity is quite small with and without the L3 cache for gaming so based on that I'd rather have the extra L2 and clock it higher for better performance over most applications since it seems to you rarely need to block for access to L3 or Ram.

I think the primary issue you'd run into is scheduling since you'd have to swap the data out and back on. You'd probably be able to do a L2<->L3 thread context switch swap faster than say L2<->RAM. This would probably hit Linux harder than Windows and its not a huge issue unless you're running a crap ton of applications at once (more high load processes than number of cores).

But that's just my understanding of it.

 

[cebalrai]

I actually started a thread last month speculating that the 631 would be an extraordinary overclocker:

http://forums.anandtech.com/showthread.php?t=2194686

I'm curious about some possible upside with this chip in terms of value though. It's clocked at a modest 2.7 ghz, however it's manufactured on a 32 nm process where half the chip isn't generating heat. Could this be the most overclockable Athlon ever? Assuming FM1 boards continue to drop in price is there some serious value potential here?

Sounds like I nailed it *geek brags all over the thread*. But I didn't forsee quite this much OC potential though.

 

[lOl_lol_lOl]

This undercuts the i3-2100'in price and provides massive advantages in multithreaded apps. They were both equal in cinebench, but this just blows it out of the water. How are gaming and ST benchmarks?

But of course, Intel is prepared. Welcome the unlocked Dual core Sandy:
http://www.nordichardware.com/news/...eparing-unlocked-core-i3-2120k-processor.html

Funny how the Athlon in its dying moments shines among the AMD lineup.

 

[sm625]

But of course, Intel is prepared. Welcome the unlocked Dual core Sandy:
http://www.nordichardware.com/news/...eparing-unlocked-core-i3-2120k-processor.html

That table they posted is at least 6 months old. I wonder what new info they have to justify the old rumor.

 

[LOL_Wut_Axel]

In all my years posting on this forum, I have yet to hear about a CPU that died because it was overclocked and then used over a long duration of time. I'm sure it happens, but it seems to be quite rare, and by the time it does happen, the computer is so obsolete that no one even bothers to troubleshoot the issue. 99&#37; of the CPUs of that type that were manufactured would probably be in a landfill site at that point.

I didn't say it'd die. I said it'd degrade faster. That's not the same thing.

It means that the more voltage you give to a CPU, the faster you'll need to raise its voltage again to keep your overclock, and that would make it degrade even faster while producing more heat and consuming more power. When you reach this point where you need to raise voltage again, the best thing to do is lower the voltage and overclock, but in many cases you need to do so by quite a bit.

Example: a Core i5-2500K at 1.5V to achieve 5.0GHz. One year from then, the overclock is no longer stable. You need to either raise voltage again to keep it or lower the voltage and overclock. In many cases degradation can be so bad that the chip isn't even stable afterwards at stock voltage and clocks, but that doesn't mean the chip died.

 

[LOL_Wut_Axel]

I can see your die-size argument for no L3 but based on current GF 32nm the power vs performance victory would probably go to a lower clocked Llano x4 with L3. I do think that a Black Edition Llano as it stands now with working GPU that can be disabled would be a nice addition for FM1 family.

But why would they? They already gave Llano 2x more L2 cache to mitigate that, so that L3 cache isn't needed in the first place. All it'd do is increase die size, heat, and manufacturing costs.

 

[Vesku]

Not worth the cost to AMD, but I'd prefer it over having to clock higher.

http://www.anandtech.com/bench/Product/121?vs=97

Assuming we are talking about OCing for maximum performance/dollar. All depends on power curves though, if Llano hits 4+GHz and draws equal or less power than an equivalent Phenom II then FM1 is a bit more interesting.