1090T or 950 for Scientific Computing

[exotic444]

It all depends on your budget really. TBH you'd be much better off with a I7-970, hex core, hyper-threading, cheaper than the 980x and would be better than the AMD. It's got the same 12mb cache as the 980x. So if you spent bit more money on the 970 and clock it to 4Ghz you would be much better off IMO. Feel free to correct me if im wrong though.

PS: I know it costs an arm and a leg but it is still cheaper than the 980x and I rekon it would be worth it in the long run.

 

[klein_orz]

I don't know how you'll speed up MCMC. Have you considered integrated nested Laplace approximations? I haven't but it's meant to be faster or more general.

I may discuss the details with you later, as I am still working on my MCMC classification tree at this moment. Once my algorithm is done, I will try to optimize it in one way or another, there has been already some paper talking about parallelized bayesian computation, e.g. Parallel Bayesian Computation
by Darren J Wilkinson. TBH, I have no idea atm, but I think I will have some thoughts later.

 

[klein_orz]

It all depends on your budget really. TBH you'd be much better off with a I7-970, hex core, hyper-threading, cheaper than the 980x and would be better than the AMD. It's got the same 12mb cache as the 980x. So if you spent bit more money on the 970 and clock it to 4Ghz you would be much better off IMO. Feel free to correct me if im wrong though.

PS: I know it costs an arm and a leg but it is still cheaper than the 980x and I rekon it would be worth it in the long run.

970 is obviously out of my budget atm...

 

[exotic444]

970 is obviously out of my budget atm...

Alright mate lol I cant mind read lol you never told me your budget so I was suggesting or maybe you could save?? if you need the computer now then dont worry, but was just trying to do my bit on the fourm and help

 

[NathWraith]

Why would the fact that the i7 is 4core/8thread instantly mean that it will provide better performance?

Hyperthreading does not provide a 100% improvement in performance for the i7 cores. If you disable hyperthreading on the 950 it will not perform at 50% compared to it being on. The i7 appears in windows task manager as having 8 cores, but these are 8 virtual cores; the processor only actually has 4 true cores & the hyperthreading allows the cores to perform more instructions per cycle which is why performance of each core is increased & also why the software only has to be able to scale to 4 cores for the i7 to operate at full load rather than 8. If the i7 had 8 true cores without hyperthreading then the software would have to be able to scale to 8 cores for the processor to operate at full load.

From the comparisons we did on this forum at the same clock speed a single i7 core with hyperthreading had higher performance than a single core from the phenom x6 in single threaded applications due to the fact that hyperthreading would allow the i7 core to carry out more instructions per clock. If the software was able to make full use of 6 cores (eg cinebench, handbrake) then at the same clock speed the phenom x 6 would perform the same or fractionally better than the i7.

In the case of the 1090T & 950 at default they do not have the same clock speed. The 1090T has a higher clock speed so it will provide higher performance if the software can make full use of the 6 cores. If the software can only scale to 4 or less cores then the i7 will probably provide better performance than the 1090T even with the lower default clock speed & if both processors are overclocked to the same speed then they will perform about the same.

The i7 is still faster.

 

[bifday2k]

No argument really, for scientific type stuff, Intel is way ahead of amd atm, not saying that AMD is bad at all, just intel are better at it.

@Someone
Going by F@H which is a scientific computing, an I7 with hyperthreading on, will spank an X6, as the client software will use all 8 threads, just as if they where all cores. So the cinebench/handbrake thing really dos'nt come into it here, as scientific software is for a start far better written to handle threads. So the X6 at the same clock speed as an I7 will not perform the same or slightly better, it will infact perform far worse.

 

[braveheart]

Intel using Hyperthearing as AMD using Hypertransport which running both way so I would go for X6 1090T that Im still using... also cheaper to buy and save some money in your pocket rather than blow all the money away.

 

[CAT-THE-FIFTH]

Are you using Linux??

This webpage maybe useful:

http://global.phoronix-test-suite.com/index.php?k=profile&u=wolle-901-27320-27846

The Phoronix forums maybe more appropriate for your question if you are using Linux:

http://www.phoronix.com/forums/

 

[Big.Wayne]

You make no sense.

An i7 950 would be superior in both instances.

Hyperthreading is: 1core/2threads

So a 950 is 4core/8threads

And a AMD X6 is 6core/6 threads

Beat me to it

Hey guys,

can you clarify if you are suggesting that HyperThreading is as effective as a real processor core?

Why would the fact that the i7 is 4core/8thread instantly mean that it will provide better performance?

Hey Someone,

I'm wondering the same thing? . . . as far as I know once you get some really serious load on an individual core the effectiveness of HyperThreading is greatly diminished . . . pulling a figure out of thin air I suspect it would take four HT threads to equal a real Core?

Would be interesting to see some data on this?

The i7 is still faster.

 

[bifday2k]

Hey guys,

can you clarify if you are suggesting that HyperThreading is as effective as a real processor core?


Hey Someone,

I'm wondering the same thing? . . . as far as I know once you get some really serious load on an individual core the effectiveness of HyperThreading is greatly diminished . . . pulling a figure out of thin air I suspect it would take four HT threads to equal a real Core?

Would be interesting to see some data on this?

If the software used properally handles hyperthreading then yes wayne, both myself and nath do F@H, and i can 100% say that both I7 and X6 clocked to the same speed, the I7 really does beat the X6 very comfortabley. It was the only reason i did'nt wait for the X6's to come, when i made my comp, as i primarily do F@H, with the odd gaming now and then.

Yes real cores do matter, but as iv said it really all depends on how the software is written, now i can't really speak about the stuff the OP uses, that stuff seems quite intense, i can only speak from experience with F@H.

From all the reading up, it looks like on average an X6 clocked to 4ghz has a TPF (time per frame) of 40+ mins for a big adv work unit.
Compare that to a TPF of 30+ mins for a I7 920 clocked to 4ghz and you can clearly see that if the program is coded right for hyperthreading, as far as the client knows, there is actully 8 cores.
Note each work unit has 100 frames, so times those times by 100 for the time it takes to complete, these bigadv units are immense, and are generally meant for multi core/multi cpu set ups.
Standford who set up F@H won't even let X6 users do these bigadvs, as they feel they are not powerfull enough to complete the units in the given time frame. By that i mean, yes u can fold them, but u won't get the corresponding points.
But even in the normal smaller A3 work units the storey is the same an I7 with Hypertheading on, will still beat a X6.

Now as i said i can't speak about the software the OP uses, or plans to use, as that looks very specific to what he is doing, or coding himself.

So if the software he is using can fully utilize hyperthreading then an I7 is by far better. Now if its coded to utilize cores more, then an X6 would be better to use.

As i said in that cinebench thread, one version was better for hyperthreading and the I7 was winning, once people started testing with the latest version that was coded better for more cores, the X6 started winning.

So to properly answer the OP, its all down to how you code the software, if you are coding it yourself, or how its already coded if its all ready made.

Simple answer:
Coded to utilize cores = Use the X6
Coded to utilize hyperthreading = Use the I7

 

[Someone]

Bifday2k thanks for the detailed reply. Is there anywhere we can look up the results for different processors as it would be interesting to see. I may look to run F@H to see what the results are as it just doesnt seem right to me but I am happy to be proven wrong.

 

[bifday2k]

Sadly its not easy to find detailed comparasions when it comes to F@H for cpu's takes a lot of forum reading on many diff forums really to see what peoples TPF's and PPD's (points per day) are, and it can vary spec to spec really.

The only time iv seen a review of this kinda stuff was on GPU's when someone took the time to test them like that, but as far as i know, no ones done a cpu round up yet.

That and not all units are the same, so can get tricky comparing, but if u wanted to try and test, feel free, best to clock ur cpu down to 4ghz and i can allways see how my past units fare up to any you do. If you are gonna test, get the latest SMP client from stanford, as bigadv set up can be a pain, the other is far easier, especially if only testing.

 

[arknor]

Simple answer:
Coded to utilize cores = Use the X6
Coded to utilize hyperthreading = Use the I7

simeply reply.

you know not what you are talking about, you do not understand what hyperthreading is.

nothing can be coded to utilize hyperthreading because it is invisible even to the OS.....

hyperthreading uses 2 threads on one core using any otherwise wasted cpu cycles at most under perfect condition a core might seee 30% more performance but this is an an ideal world with perfect conditions.

the reasonan i7 @ 4ghz beats a x6 @ 4hz in folding at home is because nearly every @home client is optimised for intel if you get one of the user made amd optimised builds the x6 will obviously fair a lot better.

i7 has more efficient cores than x6 so its generally better to have an i7 cpu but in any task that can manage 100% cpu usage on an i6 will be slightly faster than an i7.

easyrider was trying to claim i7 kills x6 in rendering and video encoding so he came with his lets benchmark so i can laugh at amd users threads.

first try was x264 video encoding benchmark.... amd was faster.
second try he came with handbrake ..... amd was faster.

he quickly got the thread closed and hoped it would die.

he then came with his cinebench 10 benchmark thread.... intel were faster , people quickly realised he was making everyone use an old version.

someone decided to bench cinebench 11.5 suprise suprise amd were faster.

the problem is not many programs can manage to fully utilise 6 cores and even when they do manage to use 6 cores amd is only 5-10% faster clock for clock.

anyway if your scientific computing program can use CUDA than a nvidia GPU will leave any CPU for dust

 

[bifday2k]

simeply reply.

you know not what you are talking about, you do not understand what hyperthreading is.

Im only going by what iv read up on, never claimed to be an expert did i?
And hyperthreading is quite simple to understand 2 threads per core, how hard is that to understand.
I never said i fully understand how that tech works, or even how some software can take advantage of it, personally id rather do more fun stuff then read up on that.

Also i was going by the standard commonaly available clients, made by Stanford, and you can't argue that what i said was correct about that.
My whole post was basicly about the coding of the software the OP was going to be using, and u said that user made folding clients the X6 runs better, my point proven then i guess, its simply all about how well coded the stuff is in the first place.

I know all about those threads btw, and i know once the latest was being used the amds where proven to be better, but again was'nt the latest cinebench better coded???

Ofcourse more real cores are better, im not trying to say they are not. Im not even suggesting that Intel is better for this, but like iv said now about 3/4 times it really all depends on how well the software is coded and if it can utilize multiple cores well.

 

[Big.Wayne]

Hello bifday2k,

No argument really, for scientific type stuff, Intel is way ahead of amd atm

Your opening statement is vague and sweeping . . .

If the software used properally handles hyperthreading then yes wayne, both myself and nath do F@H, and i can 100% say that both I7 and X6 clocked to the same speed, the I7 really does beat the X6 very comfortabley

Two things about your reply that are strange . . . the first is you have decided this thread is about F@H . . . the second is you have decided both processors are now clocked at the same speed?

klein_orz has made this thread about "Scientific Computing" and you may wish to read his posts #1, #16, #19, #22 . . . what he is describing goes right over my head, can you understand what he means?

I am indeed currently working on producing new algorithm based on Markov chain Monte Carlo

I am still working on my MCMC classification tree at this moment. Once my algorithm is done, I will try to optimize it in one way or another, there has been already some paper talking about parallelized bayesian computation, e.g. Parallel Bayesian Computation by Darren J Wilkinson

I don't see any mention of F@H or indeed overclocking?

Did I miss something?

[edit]

now i can't really speak about the stuff the OP uses, that stuff seems quite intense, i can only speak from experience with F@H

Ok thanks for the clock-for-clock feedback on i7 vs X6 running F@H . . . not sure what that is doing in this thread?

 

[Big.Wayne]

Simple answer:
Coded to utilize cores = Use the X6
Coded to utilize hyperthreading = Use the I7

you do not understand what hyperthreading is.

nothing can be coded to utilize hyperthreading because it is invisible even to the OS.....

Book him danno!

The i7 is still faster.

Is this statement relevant to klein_orz's thread or is it just a blanket statement?

 

[CAT-THE-FIFTH]

Like I said in post 28 is the OP going to use Linux to run their scientific applications?? The Phoronix benchmarks I linked to was done under Ubuntu 10.04.

 

[Big.Wayne]

Hey CAT-THE-FIFTH

Do you understand what it is that klein_orz is doing? . . . it's too complicated for me!

 

[setter]

I personaly think either are good, i do own an i7 rig, unfortunately at the time i built mine. Amd didnt have a comparable alternative, they do now on a raw cpu front, and at a good price. But for me as a gamer they unfortunately cannot offer me sli/crossfire support at dual x16 per pcie lane, something that only intel x58 can do at the minute. But i applaud amd for their work in making cpu's that really perform at a good price.

 

[bifday2k]

Wayne i brought up F@H as its a form of scientific computing, now i have no idea if its anything like what the OP is going to be doing or not and i'll openly admit i really have no clue what he is planning.

But F@H is relevant and i was trying to make a point, as iv said before im no expert and will happily admit that any time, i was just trying to put across what iv seen whilst using F@H.

Arknor wants to say i have no clue, then fine thats his right, and as i said in my reply back to him, im going by stuff iv read up on, and personal experience, and i think iv basicly been fairly clear that its basicly all about how the software he will be using is coded as to which processor would be best to use. Kinda like the same argument for anything really. Yes he came back and replied that User programmed clients run a lot better for AMD X6's, but is'nt that my whole point right there, its was coded to make more use of the AMD's cores.

Oh and just a side note to arknor if i turn HT off my Folding suffers drastically, if its on it improves, pretty good for something invisible would'nt you say lol. Also a few years back when HT first came about, where program makers not coding to take advantage of it? or was that just advertising crap? Or did i imagine that happening, nah, lol.

But anyways iv made my point, its about the damn coding of the software.