HWBOT Research Lab - AMD Thuban Voltage and Temperature Scaling - Part 4

  • Uncategorized
  • 9
  • HWBOT


Conclusive thoughts

First of all it’s important to acknowledge the fact that the Thuban overclocking results in this article are preliminary and possibly not conform what you will reach at home, be it either in a positive or a negative sense. From the Deneb C3 launch we have seen that the press samples reached a bit higher frequencies than the normal retail samples, so one could assume that’s the case with this sample as well. However, it’s important to know that this is not a press sample (not send to me as press) and that over time normal retail samples get better than the first series of processors.

What I’m getting from the system at the moment reminds me a lot of the very first Phenom II results we saw roughly a year ago. The potential is there, but at the moment there still is a problem with the multi-core scaling beyond 1.7V, which makes it look kind of disappointing at the extreme level (or maybe we just expected too much?). The contrary is true, however. The overclockability at low to medium voltage levels is beyond what I could have hoped for when I first started testing the Thuban, especially if you consider that these are six fully functional cores. Up until today, I haven’t tested each core separately (maybe something for the future), but the chances of having one freakishly high-clocking core for single threaded applications did elevate in comparison to the Deneb. On the right side I’ve included a screenshot of the LN2 overclocking result: 5.8GHz through 3DMark06 at 1.75V.

People who’ve read through the entire article may have noticed that I’ve repeated several times that the Thuban looks like a highly-binned Deneb. The most interesting part of this hypothesis is allows us to make the speculation that the lower-end Thuban products may be better for overclocking than the 1090T. Since all Thuban products receive the same power consumption rating, the high-end models are likely to be the least leaky ones. The lower the frequency, the lower the product rating, but the maximum tolerated power consumption remains equal; so the more high-end the product, the less chips qualify. To make it clear, what I’m trying to say here is: if AMD releases a mainstream Black Edition Thuban, it might be interesting to check it out. And, more importantly, let’s all cross our fingers that AMD has found the time to put aside some of the very leaky chips and is preparing another Liquid Helium overclocking attempt!

I believe it’s also important to repeat that the mainboards used in the test sessions were not fully ready yet. During the different testing, I experienced some issues regarding high Northbridge and memory clocks, which will be the determining factor for performance in the end. Since there are now six cores demanding data from the memory, you will see quite a significant performance boost when increasing the Northbridge frequency. Luckily, it seems that on air cooling, the Thuban is ready for high NB clocks (3GHz 3D stable) and rumors go that the memory frequency will hit unseen heights; higher than the DDR3-1900 CL7 3D stable I’ve seen thus far.

Although I haven’t included any performance numbers in this article (browse the web – you’ll find enough material), maybe I should elaborate on the topic anyway. Just like at the launch of the Phenom II series it’s pretty clear that the Thuban is not capable of producing one world record after another. The highest scores will still be in the hands of an Intel Gulftown-based system and Thuban will not change that in the next couple of months. However, similar to the previous AMD parts we have to recognize the fact that an AMD system is just a lot of fun to bench with: no coldbug, no coldbootbug and decent performance. Also, and this is something I rarely find important, the prize it retails at is simply amazing in comparison to the Intel six core counter product.

If you like this article, or if not, please let us know in the forums. Your feedback will play a big role in deciding if we invest more time in this kind of articles or not.

Thank you for reading.


Page 1 – Introduction
Page 2 – AMD Thuban Voltage Scaling
Page 3 – AMD Thuban Temperature Scaling
Page 4 – Conclusive thoughts


9

France anvil says:

Interesting article. Any expectation on the clocks under Liquid Helium ?

Belgium Massman says:

I noticed that the temps were dropping by ~10°C at 5.8G 06, so it seems that the dragon f1 isn't capable of holding the load. The CPU still scales with temperature a lot, so I'd say at least 200 to 300MHz more on the same chip when going LHe. Leaky chips would be even better ...

France anvil says:

I read this morning that Pt1t reached 6.5 on the 1090T he tested. http://valid.canardpc.com/show_oc.php?id=1149052 I don't know if this chip is simply a Golden Sample or not, but this promises a lot for the future.

Belgium Massman says:

Well ... no. It's one thing to get a CPU-Z validation, but something totally different from having the six cores completely stable through a 3D benchmark. This chip can only do 5.5GHz Wprime 1024M iirc

France anvil says:

Yes you're right, didn't pay attention to that ;)

Chile X-treme_Rpm says:

Fine! amd is take advantage on your processors but intel has the leader

Hong Kong imamage says:

nice sharing !! I got similar result with my 1055T/1090T on air Thanks !

Argentina Warrior_oF_Byte says:

Great article Massman!It has a lot of info very helpfull. Thanks!

Poland Yotomeczek says:

Excellent analyze :) Great job!

Please log in or register to comment.