Printable View
the performance per core is lower than i7, and HT helps them lot in MT...Quote:
Originally Posted by Oliverda
thuban can be very competitive in encoding and rendering scenarios, but it may perform just like deneb in any app that doesnt scale with more than 4 cores...
in rendering/encoding the 3.2ghz hexa-core will compete with I7 965, it's going to be around that guy. doesn't have HT or higher core perf, but it has 2 real extra cores so it is going to even out in the end. It's going to be a sweet budget workstation CPU.
Well the Opteron Octocore is coming out soon at 263$ 2 ghz + 16M cache. I will buy that...
It's true.Quote:
Originally Posted by -Sweeper_
IMHO it will compete in multi-threaded (6 or more threads) application. The most applications which can use 4 threads can use 6 or more.Quote:
Originally Posted by -Sweeper_
Anyway not all applications like HT. Bench
Sweeper summed it up nicely. :yepp:Quote:
Originally Posted by Oliverda
Which are going to be utilized in few and very limited situations. In the same scenario, i7 has 4 extra virtual cores which are helping many heavily multithreaded apps. Combined with the more efficient architecture, I'm pretty sure i7 will win in almost all the general user software available up to date.Quote:
The Phenom II X6 will have +50% cores
I'm afraid you are misinterpreting performance with frequency. Nehalem has much more efficient architecture than K10h. The differences between these two have been covered like million times on this forum, so I'm not going to talk about that.Quote:
and higher clocks as well. It will win by force of numbers.;)
Exactly, Deneb/Thuban cores are around 25% slower than i7 cores in average, clock for clock. At least 9 out of 10 things can't utilize more than 4 cores. In such scenarios, an i7 will outperform K10h. In that 1 out of 10 things, Thuban might or might not win, depending on how HT can help i7's cores for a given software.Quote:
Originally Posted by Manicdan
I'm pretty sure that you're completely wrong but the time will prove right one of us.:)Quote:
Originally Posted by gOJDO
Don't be afraid because I'm not misinterpreting anything. I know both architectures well. Just try to understand that the X6 will have 50% more cores than Bloomfield/Lynnfield which can compensate the lower clock2clock performance in many applications. That's all.Quote:
Originally Posted by gOJDO
Just to mention some popular titles which can take advantage of the 6 or more threads/cores: Adobe After Effects, Adobe Premier Pro, Sony Vegas, Cyberlink converters, Finereader, 3ds max, Maya, Lightwave, etc...Quote:
Originally Posted by gOJDO
If an application can't take advantage of the more than four thread/cores than it can't take advantage of the HT neither.
Nice buy for sure. But what is a good mobo going to cost?Quote:
Originally Posted by vitchilo
Some PR-people in AMD should get fired. This news are exactly one week late. It should break out on Mars 15. !
and more cores wont help against higher IPC in those cases...Quote:
Originally Posted by Oliverda
Exactly but the higher clocks can help. Turbo and more MHz for the same price.;)Quote:
Originally Posted by -Sweeper_
http://prohardver.hu/dl/upc/2010-03/13063_turbo3d.jpg
http://prohardver.hu/dl/upc/2010-03/...st-498x350.jpg
Any multithreaded app can take advantage of the two extra cores in Thuban. The application has no clue about how many "threads" are available to it. That's controlled by the OS. The OS will schedule the threads to execute on certain cores.Quote:
Originally Posted by gOJDO
From these graphs it looks like the unused cores are going into cnq idle mode, i wonder if its a good thing for performance.
Power consumption should be great tho.
thats something very nice intel introduced in 2008.:DQuote:
Originally Posted by Oliverda
even at those frequencies, it wouldnt beat an i7 930 @ 2.93ghz/3.06ghz
but I got your point... :up:
Yes, and what about the very nice IMC which introduced in 2003 by AMD and only in 2008 by Intel. IMHO better to not start a stupid argument about this.;)Quote:
Originally Posted by -Sweeper_
Any evidence or should I satisfied with the empty words?:)Quote:
Originally Posted by -Sweeper_
yes, pretty much all the benchmarks with bloomfield 3ghz vs deneb 4ghz :up:Quote:
Originally Posted by Oliverda
Let's just remind the resident intel fans doing the shift this evening @ XS that Bloomfield has approx. ~22% IPC advantage,on average(variety of workloads,mostly MT with a mix of single thread apps) over same clocked Deneb(SMT on,Turbo off). So no,a Thuban with Turbo mode for poorly threaded apps. and 50% more real cores (that also do turbo if utilization is below 6 cores) will compete more than fine with Bloom and Lynnfield lines.
All right. Looks like that I have to be satisfied with the empty words.:)Quote:
Originally Posted by -Sweeper_
There is no such thing like a "K10h". You can call it either K10, where "10" is a decimal ten, meaning the 10th in the "K" line. Alternatively, you can call it F10h, where the F is the abbreviation of Family and "10h" is a hexadecimal number which equals decimal 16, meaning the 16th family of AMD x86 CPU's (which followed Family F(h) (K8), where F(h) equals decimal 15). So, "K10h" is a stupid mixture, used by some people knowing nothing about hexadecimal numbers.Quote:
Originally Posted by gOJDO
If AMDs trubo work anything like intels Turbo, the windows scheduler will prevent any higher turbo modes then + x1 (same as for intel).Quote:
Originally Posted by Oliverda
Got love that the windows scheduler distributes loads over all cores....
Not as high as it could be.Quote:
Originally Posted by informal
dream on :p:Quote:
Originally Posted by -Sweeper_
3Ghz Bloomfield should be around 3.3-3.4 Ghz Deneb only
A 3GHz Bloomfield should be around a 3.3-3.4 Ghz Thuban.Quote:
Originally Posted by jcmarfilph
3GHz Bloomfield should be around a 3.0-3.1Ghz Thuban in highly multithreaded apps.Quote:
Originally Posted by accord99
Based on two benchmark results posted in this thread, 1055T can compete with i7 920 or even 930. That's 100Mhz difference only. :clap:Quote:
Originally Posted by accord99
Exactly.Quote:
Originally Posted by Oliverda
Thanks for correcting me. My bad, making a stupid mixture of K10 and F10h(or how AMD calls their K10/K10.5 in their tech docs). As for the hexadecimal numbers...:rofl: Do you think that an electrical engineer with an M.Sc degree, who has programmed in assembler at age of 15, knows nothing about hexadecimal numbers? :up:Quote:
Originally Posted by dess
:rofl:Quote:
Originally Posted by jcmarfilph
Of course, sooner or latter time will tell. :up:Quote:
Originally Posted by Oliverda
And that would be the case only when heavily(more than 4 cores) multithreaded software comes in to play. For the rest of the apps, which are not optimized to utilize more than 4 cores, and such are most of the apps available today, the extra 50% cores will be useless, while the higher IPC of Nehalem will yield higher performance.Quote:
Don't be afraid because I'm not misinterpreting anything. I know both architectures well. Just try to understand that the X6 will have 50% more cores than Bloomfield/Lynnfield which can compensate the lower clock2clock performance in many applications. That's all.
I agree, most of the apps you mentioned can take advantage of the 50% extra cores, but that won't yield 50% better performance. Most of these apps will yield no more than 15~20% better performance. Only the x264 and divX converters might yield 30+% better performance in average(and that depends of the type and quality). That won't be enough to counter an i7 clocked 5% lower.Quote:
Just to mention some popular titles which can take advantage of the 6 or more threads/cores: Adobe After Effects, Adobe Premier Pro, Sony Vegas, Cyberlink converters, Finereader, 3ds max, Maya, Lightwave, etc...
Exactly. And in such scenario, Nehalem dominates over same clocked Phenom II.Quote:
If an application can't take advantage of the more than four thread/cores than it can't take advantage of the HT neither.
I stand behind what I said. I hope that benches will popup in the following two weeks, so we can include some exact numbers in our discussion.