Phenom II x6 crunching performance?

[jcool]

Does anyone have specific numbers yet? I'd be interested in potential PPD (HCC & HFCC projects), power draw, overclockability (4Ghz would be nice) and so on.

Thuban is already listed at various distros across Europe and the US. I read some good things in the AMD section, apparently this new E0 stepping is better than other current AMD offerings (it better be, the PII 940BE I bought a while back just to see whether they are any good was, well - not convincing, to say the least )

Anyway, if the price is right (knowing AMD it will be), I'd like to get one of those chips to bolster AMD presence in my little farm.

One of my LianLi V350's is still vacant, so I'd like to use that. Any good (decently priced) AM3 µATX boards than can OC and hold a Thuban? I got loads of DDR3, a few HDDs, PSUS, etc.

[sierra_bound]

One thing to keep in mind. The Phenoms have 6 cores but only 6 threads. They are priced well below retail Westmeres. But the 1090T will be priced about the same as the i7-930 which as you know has 8-threads of crunching power.

[jcool]

Yes, I just talked to Jaco in IRC. We estimate around 30k PPD @ 4Ghz for a Thuban - equal to a Bloomfield at the same frequency.
Now if the 1035T (smallest Thuban model) is at around 200€ (lower than an i7 920), and I can get this for around 70 bucks net (=half the price any x58 board costs), and IF Thuban doesn't cosnume more power than a 920, it seems like a good idea.

Of course, ES Westmeres are still better... that Dual X5650 is insane. It does almost 75000 PPD at 300W from the wall. If my whole farm would consist of these machines, I could do half a million PPD with only 2KW/h from the wall - less power than I am using right now.

[informal]

Jcool,it will have roughly 50% better perf. in terms of PPD than Deneb at the same clock(or maybe add max. 5% more over that,due to small speed tweaks between Thuban and C2 Deneb:think better IMC,a bit lower cache latencies etc.)

[jcool]

I know that
Interesting point (for me) is mainly power consumption - will it eat 50% more than a Deneb (if so, not interested) or less? If less, how much less?

[sierra_bound]

I have not seen results yet for the 1035T. The 1055T seems to clock very well. And AMD boards are generally a lot cheaper than Intel ones.

[Sparky]

I would guess they are going to try and keep it in the power envelopes. No bigger than 140W envelope I'll bet. So while it may use some more than Deneb, I wouldn't expect another half again on the power usage. That's something both AMD and intel usually try to avoid (well, the P-D really didn't, and it showed lol).

At one point I was considering selling off parts of my primary desktop (case, watercooling, board) and buying a uATX case and board for it instead. Then I realized that didn't make much sense... but the board I was looking at was basically the uATX version of my current MSI 790FX-GD70.

[informal]

It will "eat" roughly the same power as same clocked Deneb.These bad boys have some massive process node improvements,so much that AMD managed to make hex core at the same clock and same TDP specification as they did with their previous quad core chip working at the same clock . Thuban @ 3.2Ghz(with Turbo core ability) is rated @ 125W max TDP while Deneb C2 in the form of 955BE is a quad rated @ 125W max. TDP. This almost equals what intel did with a whole process node transition going from Bloomfield to Westemere(130W 3.3Ghz 45nm QC -> 130W 3.3Ghz 32nm 6C).

[jcool]

If true, that would be impressive.
On Intel side, the Westmeres actually draw less power than Bloomfield/Gainestowns at the same clock and voltage. Despite their identical TDP ratings.
Yup, I have measured 5 or so different Hexas, ranging from A0 to current B1 - even the A0 wasn't as power-hungry as my w3520.

Can't wait to see prices on the 1035/1055T - I guess I'll decide then what to buy.

[informal]

Well you are the expert on the WCG crunching abilities,you will surely decide what is the best perf./$/watt

[Particle]

30K may be a little optimistic. With 12 K10 cores at 2.4 GHz I get about 25K/day.

[charged3800z24]

30K may be a little optimistic. With 12 K10 cores at 2.4 GHz I get about 25K/day.

True, but there is a nice clock speed bump also. Will be interesting to see. I hope to get one and find out.

[tbone8ty]

mmmm X6 has freshly Zambonied silicon :p

[lkiller123]

I think it will be 50% (or even more) increase in points. I hope that it will OC the same way.

[jcool]

Well - if you can get such a chip, please do post BOINC results

[PoppaGeek]

I plan on getting a 1090T first of May. Will post when I do.

[v0dka]

A quick calculation based on 30% extra performance from HT. A quad i7 has 5,2 "real" cores and Thuban has 6. That should roughly eliminate the IPC difference that exists between the two. So assuming that Thuban indeed clocks to 4Ghz and beyond on low volts it will be a good cruncher with about the same perf / watt as 45nm i7. Probably even better perf / watt as i7 with HT is quite powerhungry. Not to mention very hot.

Im not sure about the price, but I assume Thuban is inexpensive as well as the board. All in all I would say Thuban is slightly more attractive and efficient than 45nm i7 which is exactly what we need to get prices down on 32nm i7.

I hope to see a lot of Thubans on our team the coming six months or so.

edit: holy crap seems that everyone is watching Thuban now