amd or intel?

ahj
ahj
Joined: 25 Jul 10
Posts: 17
Credit: 4,331,992
RAC: 0

Yep, the hyperthreading

Message 80932 in response to message 80930

Yep, the hyperthreading doesn't scale like that my friend. It is still 10%, not 40%.

Conversely, the X6 DOES have 50% more physical cores than the i7, so in that case, you are virtually guaranteed at LEAST a net 40% increase in WU production compared to an i7.

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3,522
Credit: 689,257,729
RAC: 217,898

RE: Yep, the hyperthreading

Message 80933 in response to message 80932

Quote:

Yep, the hyperthreading doesn't scale like that my friend. It is still 10%, not 40%.

Conversely, the X6 DOES have 50% more physical cores than the i7, so in that case, you are virtually guaranteed at LEAST a net 40% increase in WU production compared to an i7.

Well, my Core 2 Q8200 has 100% more cores (4 vs 2) than my i5 (hyperthreaded), and it's not even a bit more productive CPU-wise at all...

Please, let's do the numbers correctly. Pick real runtimes from existing machines and then we can compare the projected best case 24/7 output for each machine.

CU
HB

ExtraTerrestrial Apes
ExtraTerrestria...
Joined: 10 Nov 04
Posts: 770
Credit: 539,790,843
RAC: 147,896

Guys.. didn't we just work

Message 80934 in response to message 80932

Guys.. didn't we just work out the numbers a couple of posts before? At Einstein an AMD 6-core can not match an i7 4-core if both are running stock. OC makes it worse for AMD since the Intel has more clock speed to gain. Sure, it's 50% more cores - but they're crunching much less per clock.

Also Alex's got the power consumption numbers swapped: the AMD 6-core is 125W TDP, whereas socket 1156 i7 are all 95W TDP.

Furthermore you can't assume Nehalems HT to scale the same as for your atom. Sure, Nehalem is more efficient, but it's also got many more execution units to feed and wider decoders etc. It's a totally different architecture and thus HT behaves differently.
and, yes, HT can reduce performance. But this is not the norm. We've had this discussion recently over at Seti.Germany and a total of one example has been brought up, where HT actually hurt (Prime Grid LLRs). So if all you do is running that then go ahead and switch HT off, but otherwise: run a project mix, leave it on and enjoy the higher throughput virtually for free.

And a Bulldozer X8 chip does have 8 integer cores, so ABC should rock. But it's only got 2 FPUs for each integer core pair. That X8 has half the raw FP power of a 4-core Sandy Bridge. I wouldn't bet on Bulldozer to be a great general purpose BOINC'er.

MrS

Scanning for our furry friends since Jan 2002

ahj
ahj
Joined: 25 Jul 10
Posts: 17
Credit: 4,331,992
RAC: 0

AMD's are that bad for

Message 80935 in response to message 80933

AMD's are that bad for E@H?

I didn't realise; the other 6-7 DC projects I've been involved in over the years seem to be fairly evenly matched between the two CPU makers.

Bikeman; I'm not understanding correctly. Are you saying that your C2Q is about the same as an i5? But with or without hyperthreading?

ExtraTerrestrial Apes; The latest revision of the 1055T is 95w, and while its distribution is limited, it's out there.

Also, I understand your point about the Atom's hyperthreading; as was stated in my OP. Of course you can't compare the Atom to Nehalem, you're looking at a crippled in-order CPU to a full fledged out-of-order CPU with immense bandwidth, cache and floating point power.

I only drew the 10% figure because, from a lot of previous DC projects, that seems to be the median improvement when using hyperthreading.

ExtraTerrestrial Apes
ExtraTerrestria...
Joined: 10 Nov 04
Posts: 770
Credit: 539,790,843
RAC: 147,896

Cannibal Corpse, that's quite

Message 80936 in response to message 80925

Cannibal Corpse, that's quite an extreme build you're planning there. Apparently you want to clock it quite high, which does make it costly. May I ask what you want to do with it? Games and general purpose BOINC or mainly Einstein?
Personally I wouldn't pay that much for a PC. I'd go with solid, quite air cooling and upgrade more often if I feel the need to do so. For you this point may be reached beginning of next year when Sandy Bridge hits, or maybe next fall when we should see a new high end platform with 4 memory channels and Sandy Bridge CPUs with more than 4 cores (don't know if it's going to be 6, 8 or both - the bandwidth for 8 is certainly there).

Regarding your actual components:

Quote:
Intel Core i7 875K Unlocked 2.93GHz LGA 1156 Boxed Processor( intel turbo to 3.6) $249.99


Very nice.

Quote:
Asus P7P55D-E Premium LGA 1156 P55 ATX Motherboard $271.97


It's probably a good board.. but almost 300$?! I'd go for the cheapest Gigabyte or Asus with 4 DIMM slots and USB3, or whatever over feature you may require. Over here that's <100€. Especially since you've got an unlocked multiplier, so you don't need to push the board very hard to push the CPU.

Quote:
A-Data 4GB (2 x 2GB) DDR3-2000 (PC3-16000) X Series Dual Channel Memory Kit $149.99 x2 $300.00


For me I found that buying high end RAM is not really worth it. I'd rather go for DDR3-1600 CL7, which should be ~100$ / 4 GB. And makes it much easier to extract that performance, motherboard wise.

Quote:
BFG Technologies EX1200 1200W ATX 12V / EPS 12V Modular Power Supply $219.99


Totally overkill and I couldn't find a more precise efficiency number than "80+". Which means it probably does not even have 80+ Bronze (otherwise they'd have advertised it as such?). You really want an 80+ Gold unit, which saves you quite some money over time. Take a look at the Seasonic X Series and Enermax 87+ Modu. Both reach 91 - 92% efficiency under typical loads. 600 - 800 W will be fine, depending on your GPU choices.

Quote:
Antec Dark Fleet 35 (DF-35) Mid-Tower ATX Case $114.99


That's obviously up to you. But boy, is this thing ugly! I'd prefer the P183 or Mini P180 over this any day.

Quote:
Logisys 3.5" Aluminum Dual-Fan Hard Drive Cooler $15.99
Western Digital VelociRaptor WD3000GLFSRTL 300GB 10,000 RPM SATA 3.0Gb/s 3.5" Internal Hard Drive $199.99..optional
Western Digital Caviar Black WD1002FAEX 1TB 7,200 RPM SATA 6.0Gb/s 3.5" Internal Hard Drive - OEM $99.99


The Raptor is nice and the Black is nice. However, I'd go for a 60, 90 or 120 GB Sandforce SSD plus a 2 TB Green ('Cuda LP or WD Green). This gives you more performance, more capacity, less noise (no cooler needed) and is possibly cheaper, depending on how large you'd want your SSD to be. This wouldn't work so well with a 200 GB Steam folder, though.

Quote:
PNY GeForce GTX 465 1024MB GDDR5 PCIe 2.0 x16 Video Card $219.99 after rebate savings [details] $249.99


What do you want to use it for? If you don't need double precision then the GTX460 is much more economical and barely slower. And if you need double precision - both cards suck at this anyway (compared to ATI).

If you do want to spend the 2.5k$ and want to maximize BOINC CPU-throughput you might want to consider a 6-core Gulftown.

MrS

Scanning for our furry friends since Jan 2002

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3,522
Credit: 689,257,729
RAC: 217,898

RE: AMD's are that bad for

Message 80937 in response to message 80935

Quote:
AMD's are that bad for E@H?

They perform rather poorly especially with the ABP2 (CPU-variant). For the other search, S5GC1, AMDs are kind of ok. After the next round of optimization, hopefully the AMDs can close the gap a bit.

Quote:

Bikeman; I'm not understanding correctly. Are you saying that your C2Q is about the same as an i5? But with or without hyperthreading?

With hyperthreading, my 2 core i5-430M has almost exactly the same overall CPU E@H performance as my slightly higher clocked 4 core Core 2 Q8200, yes!!!

[EDIT]And needless to say, my (notebook) i5 is consuming far less electicity as the Quadcore Core 2...[/EDIT]

And I'm not alone. When you browse thru the Top-List of hosts here, you'll see several quad-core i7 competing with dual quad-core Core2-era Xeons. And the individual runtimes clearly indicate those run hyperthreaded.

I guess it's the faster memory interface that, among other things, make hyperthreading much more useful with the i5/i7 CPUs compared to the old days of FSB and the Pentium 4.

CU
HB .

ExtraTerrestrial Apes
ExtraTerrestria...
Joined: 10 Nov 04
Posts: 770
Credit: 539,790,843
RAC: 147,896

RE: I guess it's the faster

Message 80938 in response to message 80937

Quote:
I guess it's the faster memory interface that, among other things, make hyperthreading much more useful with the i5/i7 CPUs compared to the old days of FSB and the Pentium 4.

It's also the wider front end, i.e. ability to decode and handle more instructions and the improved cache system and buffers.

MrS

Scanning for our furry friends since Jan 2002

Mike Hewson
Mike Hewson
Moderator
Joined: 1 Dec 05
Posts: 6,537
Credit: 286,509,719
RAC: 92,819

RE: Yep, the hyperthreading

Message 80939 in response to message 80932

Quote:
Yep, the hyperthreading doesn't scale like that my friend. It is still 10%, not 40%.


Yeah. I'd always believed that hyperthreading was effectively a hardware mechanism to recover time otherwise lost with context swaps - task state segments, various flushes et al. So your HT gain is only in that area and not with regard to uninterrupted aspects of the thread. So if you had four cores running only four threads - forever, no swaps - that's all you get. Thus the base question becomes 4 Intel cores @ some_spec_per_core vs. 6 AMD cores @ some_other_spec_per_core. Then toss actual machine thread activity ( with differing swap costs Intel vs AMD ) on top of that ....

Cheers, Mike.

( edit ) Of course a given HT labelled processor may be blessed with other advantages, not especially related to HT. As noted, to find out - turn HT off. I'm amazed what one can stuff onto a single chip now-a-days. I remember the gleam in my eye opening the package containing the math coprocessor for a 386. What was virtual - enacted via emulation behind an interrupt call - became actual.

I have made this letter longer than usual because I lack the time to make it shorter ...

... and my other CPU is a Ryzen 5950X :-) Blaise Pascal

ExtraTerrestrial Apes
ExtraTerrestria...
Joined: 10 Nov 04
Posts: 770
Credit: 539,790,843
RAC: 147,896

Well, don't make it too

Message 80940 in response to message 80939

Well, don't make it too complicated. It's just c*a + c*b = c*(a+b) :)
And HT does more than just recover cycles otherwise lost: it also uses idle decoding and execution time slots. It's not trivial to extract 4+ independent instructions from a single thread each clock, for the entire length of the pipeline (20+ stages).

MrS

Scanning for our furry friends since Jan 2002

Mike Hewson
Mike Hewson
Moderator
Joined: 1 Dec 05
Posts: 6,537
Credit: 286,509,719
RAC: 92,819

RE: Well, don't make it too

Message 80941 in response to message 80940

Quote:
Well, don't make it too complicated. It's just c*a + c*b = c*(a+b) :)


Heaven forbid ! :-) Maybe the biggest difficulty with HT is the hype. :-)

Quote:
And HT does more than just recover cycles otherwise lost: it also uses idle decoding and execution time slots. It's not trivial to extract 4+ independent instructions from a single thread each clock, for the entire length of the pipeline (20+ stages).


Well, I didn't know that. Thanks. :-)

Cheers, Mike.

I have made this letter longer than usual because I lack the time to make it shorter ...

... and my other CPU is a Ryzen 5950X :-) Blaise Pascal

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.