A few questions about processors.
If I search Intel for info I see the following:
Number of cores: 6
Box: no
Bus type: DMI
etc. What does "Box" mean?

How do I know which chipset (thus type MOBO) will fit? Not with all processors they mention the chipset I notice. Do I only have to look for the socket to fit?

This one is available in the Netherlands: Intel Core i7-4770K - 3.5GHz - Socket 1150 - Unlocked. What does that "Unlocked" mean? It is not explained on the site.

Finally I see that in the Netherlands AMD processors are very cheap compared with Intel, could be a factor 3 or 4. Am I right that AMD does not have HT?
Will AMD processors work flawless with nVidia GPU's?
Thanks.

---

Greetings from TJ

Box means its in a box (retail), rather than a tray (OEM product). Box usually means the CPU is in a box with a heatsink and fan.

Unlocked means you can overclock the CPU.

AMD does not use HT.

Generally speaking AMD processors do not perform as well as Intel CPU's but AMD's tend to be less expensive. While you could pay around 400Euros for an Intel CPU you could also pay 60Euros for a lesser Intel CPU or get an AMD processor.

You will have to research each processor to make sure it fits the board you want. NVidia GPUs run fine on systems with AMD processor or Intel processors.

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Well I have one AMD system with AMD CPU and AMD GPU. I use it only for my study and I must say it runs nice, quit and boots fast. I didn't find any problems with it.

I didn't know that it is so important to tell that is is in a box :)
I will use a Zalman CPU cooler.

I found a AMD 8 core black edition 4GHz (125W) for €179 and a comparable (however they can not be compared) Intel 3.5GHz (77W) would cost €305.

I will go for Intel then, first seek a processor and then the MOBO.
However remains my question about the socket and the chipset unanswered.
If a processor needs a 1150 socket and the MOBO has a 1150 socket, then it will fit. But plays the chipset any role? On several retail sites and on Intel's I do not always find info about the chipset. That is way I am asking, as parts that don't work together are easily bought.

---

Greetings from TJ

Any LGA1150 CPU will work on any LGA1150 motherboard.

Note that the CPU's bring limitations of their own; an i3 won't support PCIE3, but an i5 or i7 will. If a CPU only clocks to 2GHz it will reduce the GPU performance for here, but only slightly.

My next system will be another Linux rig with 2 GPU's. It will replace an existing (more power hungry) system (another 17% Electric price hike just about to kick in here thanks to privatization) and will be built for purpose (GPU crunching). As I don't intend to use it to crunch for any CPU projects, I don't want to spend £350 on a CPU so I will get the cheapest dual Intel CPU I can (£50). I know I will lose some performance running at PCIE2.0 @X8 and supported by a lesser CPU, but it won't be much and it's certainly not worth paying an extra £300+running costs for a couple of percent improvement.

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Finally I see that in the Netherlands AMD processors are very cheap compared with Intel, could be a factor 3 or 4. Am I right that AMD does not have HT?
Will AMD processors work flawless with nVidia GPU's?

AMD processors work perfectly with NVidia GPUs. I'm running 9 machines with one NVidia GPU and one ATI/AMD GPU in each. Intel is a little faster in most CPU projects but AMD is faster in some. Much of the reason you're seeing faster benchmarks in CPU reviews is that the top AMD processors have more cores and most single programs don't use that many cores at once. From the Guru 3D review:

Concluding then. I'll keep saying this, personally I would have preferred a faster per core performing AMD quad-core processor rather then an eight-core processor with reduced nice per core performance. However we do have to be clear here, we have been working with the FX 8350 processor for a while now and it simply is a great experience overall. Your system is very fast, feels snazzy and responsive. The Achilles heel simply remain single threaded applications. The problem here is that it effects game performance quite a bit, especially with high-end dedicated graphics cards and that's why in it's current form the FX series simply is not that popular amongst the gaming community.

However when multi-threading kicks in wheter that is a game or application ... that loss is turned around in a gain.

http://www.guru3d.com/articles_pages/amd_fx_8350_processor_review,21.html

Of course DC crunching uses all the cores so multicore usage is not a problem. As a couple reviews mention: with the money you save on the AMD processor you can afford a better GPU and thus end up overall with a faster system.

Everybody seems to forget one big glaring development, Jim Keller is back at AMD and there's going to be 1 maybe 2 more CPU upgrades for the socket AM3+.
Ya,ya,ya, just another AMD fanboy spouting off, I've already seen a leaked partially redrawn Steamroller core die shot (socket AM3+), so he's hard at work. That was the one and only reason that sold me on AMD years ago, is the upgrade path with their motherboards, they stick with the same socket for a long time (even AM3 CPU's work in the AM3+).

I'm not trying to start nothing here, I just get tired of AMD being treated like it's a dirty word. I think after what Intel just released, it's to big of a carrot for Jim not to jump on it and make up all the lost ground.

Edit: I mean Jim Keller, sorry about that.

The Haswell paradox: The best CPU in the world… unless you’re a PC enthusiast, By Sebastian Anthony

"Where the Core i7-3770K is happy to sit at 3.7GHz under full load at 90C, the Core i7-4770K throttles back to 3.5GHz within moments of starting Prime95."
K=pig on a poke!

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Honestly, I do not know much about Computer Science and if would not be for BOINC, I still would not read about it.

But I do know, that I am use all cores or threads (HT) for BOINC on all my computers all the time and if I do have a GPU in the computer, I will have one thread / core reserved feeding it.

So at the end, I am not very concerned about the single thread capabilities of one of the CPU makers, in my case all cores or threads are always 100% charged / occupied and one maker is a lot cheaper, helps keeping the price of CPUs in check, uses the same motherboard for various generations. Or do I miss something? Although I still do have one question mark: The TDP thing, it seems to me that there is a slightly difference of it’s definition between the two competitors, which does not help to compare this important specification of each CPU. But on the other hand is quite similar for more or less the same throughput, isn’t it?

Or do I miss something?

What you're missing is, that higher per-thread performance means higher overall performance. Especially for distributed computing projects, like BOINC, that utilize all available computing power, this means you'll get significantly more work done.

For crunching, I think AMD just can't beat Intel right now. The only AMDs I'd pick would be 8-cores, but again they're not real 8-core, they are 4-core with double integer units *, making for half-8-core, if such a term is valid. So, don't really know how they would fare against Intel 4-cores at similar frequencies.

* maybe other core parts as well

---

Or do I miss something?

What you're missing is, that higher per-thread performance means higher overall performance. Especially for distributed computing projects, like BOINC, that utilize all available computing power, this means you'll get significantly more work done.

For crunching, I think AMD just can't beat Intel right now. The only AMDs I'd pick would be 8-cores, but again they're not real 8-core, they are 4-core with double integer units *, making for half-8-core, if such a term is valid. So, don't really know how they would fare against Intel 4-cores at similar frequencies.

* maybe other core parts as well

An AMD 8 core will do more work than an 4 core Intel with the same clock speed.
Also keep in mind the Intel quad-cores with HT and thus run 8 cores, actually have 4 real cores. And the AMD with 8 cores have higher clock speeds than "compatible" Intels. I have an AMD 4 core and does not under perform with Intel, its temperature is lower. So that is a plus, as well as the lower price.
The only min point I have is that AMD's have higer TDP and theoretically use more power. This is off course an issue when running 24/7.

---

Greetings from TJ

For crunching, I think AMD just can't beat Intel right now. The only AMDs I'd pick would be 8-cores, but again they're not real 8-core, they are 4-core with double integer units *, making for half-8-core, if such a term is valid. So, don't really know how they would fare against Intel 4-cores at similar frequencies.

Oversimplified explanation: lets call it 8 integer cores and 4 floating point cores. AMD decided to focus on the much more common integer tasks and rely more on extensions to bolster floating point. This was a change from the Phenom X6 which had 6 powerful hardware floating point cores. In fact a case could be made that the Phenom X6 is still the best bang for the buck processor available. At some projects it's faster than ANY Intel i7 (my favorite CPU project Yoyo for instance). The 95w 1045T can be had for $80 from Microcenter or $90 from TigerDirect and works on the latest AM3+ motherboards. Think of all the extra cash you could use to buy a better GPU. For instance the price difference would more than move you up from a 650 Ti to a 660 Ti and still have enough money left over for a nice dinner, another 8GB of ram or a better power supply.

The Phenom X6 is also slightly faster on Einstein, my main project.

---

Greetings from TJ

"8 integer cores and 4 floating point cores" - Now there's a good description, and explanation as to why the Phenom X6 processors outperform the latest 8-core AMD processors for floating point apps.

Which of the Yoyo apps does the 1045T perform best at?

"The Phenom X6 is also slightly faster on Einstein, my main project".
- Faster than what, an 8core AMD or an i7 (and which one)?

Some CPU projects are very dependent on memory speeds.

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

"8 integer cores and 4 floating point cores" - Now there's a good description, and explanation as to why the Phenom X6 processors outperform the latest 8-core AMD processors for floating point apps.

Which of the Yoyo apps does the 1045T perform best at?

"The Phenom X6 is also slightly faster on Einstein, my main project".

I run yoyo ecm but even projectwide the X6 is fastest:

http://www.rechenkraft.net/yoyo//top_hosts.php

The 2 opterons at the top are multi cpu servers and the intel in 3rd place has multiple cpus (24 cores). 4th (and fastest single cpu machine) is my lowly 1035T running 5 cores on Yoyo. Next is a 6/12 core HT 3930K that just popped up a few places. From there down on the first page (discounting the multi CPU servers) it's mostly AMD even though there are far more Intels running the project.. Even the 8120 does well. The sandy & ivy bridges have done far better than the earlier Intels and some are hovering near the top, but considering the cost. If you look at the all time list it's even more telling, 14 of the top 20 single CPU machines are Phenoms. It will be interesting though when a few of the new 8350 8 cores work their way up the RAC list. I would guess they might be the new Yoyo speed champs.

Both.

---

Greetings from TJ

OK, I spent some time looking at BOINCstats CPU break-downs to get as clear a picture of CPU performance as I can. First of all, I looked at not only Yoyo, but also WCG and SETI, as Yoyo is really a niche project with only ~3700 hosts vs ~216000 of WCG and ~207000 of SETI. Yoyo and WCG run apps of various kinds, so they should reflect overall CPU performance adequately well. SETI is more specific, but should also help.

Then I had some difficulty choosing the column(s) to sort CPUs by, as I wanted to take out historical, populative and usage (how many hours per day of BOINC execution) factors. Total credit is populative, historical and usage-dependent, Average credit is populative and usage-dependent, Credit per CPU is historical and usage-dependent, Average credit per CPU is usage-dependent. That leaves us with Average credit per CPU second, which sounds like independent of these three factors.

For Yoyo, here are the first 20 CPUs(link):

1 Cell Broadband Engine [Model 0 ]
2 Cell Broadband Engine
3 AMD Athlon(tm) II X2 265 Processor
4 PS3Cell Broadband Engine
5 Intel(R) Core(tm) i5-2405S CPU @ 2.50GHz
6 AMD Phenom(tm) II X2 521 Processor
7 Intel(R) Core(tm) i5-2550K CPU @ 3.40GHz
8 Genuine Intel(R) CPU @ 2.90GHz
9 Intel(R) Core(tm)2 Extreme CPU X9775 @ 3.20GHz
10 Intel(R) Core(tm) i5-3570K CPU @ 3.40GHz
11 Intel(R) Xeon(R) CPU E3-1275 V2 @ 3.50GHz
12 Intel(R) Core(tm) i5-2500K CPU @ 3.30GHz
13 AMD Phenom(tm) II X4 B50 Processor
14 Intel(R) Xeon(R) CPU E5205 @ 1.86GHz
15 Intel(R) Core(tm) i5-2500 CPU @ 3.30GHz
16 Intel(R) Xeon(R) CPU E31220 @ 3.10GHz
17 AMD Phenom(tm) II N620 Dual-Core Processor
18 AMD Phenom(tm) II X4 B95 Processor
19 Intel(R) Core(tm) i5-3470S CPU @ 2.90GHz
20 Intel(R) Core(tm) i5-3570 CPU @ 3.40GHz

Here are the top 20 CPUs for WCG (link):

1 IntelQEMU Virtual CPU version 0.9.0
2 Intel(R) Core(tm)2 Quad CPU @ 3.20GHz
3 Intel(R) Core(tm) i5-2300 CPU @ 2.80GHz
4 AMDQEMU Virtual CPU version 0.9.1
5 Genuine Intel(R) CPU 0 @ 2.70GHz
6 Intel(R) Xeon(R) CPU X5270 @ 3.50GHz
7 Intel(R) Xeon(R) CPU X5492 @ 3.40GHz
8 Intel(R) Core(tm)2 Extreme CPU X9750 @ 3.16GHz
9 Intel(R) Core(tm) i7 CPU X 995 @ 3.60GHz
10 Intel(R) Xeon(R) CPU X3380 @ 3.16GHz
11 Intel(R) Xeon(R) CPU E3113 @ 3.00GHz
12 Intel(R) Xeon(R) CPU X5677 @ 3.47GHz
13 Intel(R) Core(tm)2 CPU T6600 @ 2.20GHz
14 Intel(R) Core(tm) i5 CPU S 750 @ 2.40GHz
15 Intel(R) Xeon(R) CPU X3370 @ 3.00GHz
16 Intel(R) Core(tm)2 Duo CPU E8435 @ 3.06GHz
17 Intel(R) Xeon(R) CPU X5272 @ 3.40GHz
18 Intel(R) Core(tm) CPU 750 @ 2.67GHz
19 Intel(R) Xeon(R) CPU X5482 @ 3.20GHz
20 Intel(R) Xeon(R) CPU X5470 @ 3.33GHz

Here's the list for SETI (link):

1 Intel(R) Core(tm) CPU 975 @ 3.33GHz
2 Intel(R) Core(tm) CPU 960 @ 3.20GHz
3 AMDAthlon 64 Dual Core 4200+
4 AMD Athlon(tm) X2 Dual Core Processor 6850e
5 Intel[EM64T Family 6 Model 23 Stepping 10]
6 [Intel64 Family 6 Model 23 Stepping 10]
7 AMD Athlon(tm) 64 X2 Dual Core CPU 4400+
8 Intel(R) Xeon(R) CPU E5240 @ 3.00GHz
9 AMD Sempron(tm) Dual Core Processor 4700
10 AMD Phenom(tm) Ultra X4 22000 Processor
11 Intel(R) Core(tm) CPU 860 @ 2.80GHz
12 Intel(R) Xeon(R) CPU L3110 @ 3.00GHz
13 Genuine Intel(R) CPU 000 @ 2.93GHz
14 AMD[x86 Family 15 Model 107 Stepping 2]
15 Quad-Core AMD Opteron(tm) Processor 2358 SE
16 Intel(R) Core(tm)2 CPU E8500 @ 3.16GHz
17 Intel(R) Xeon(R) CPU X5272 @ 3.40GHz
18 Genuine Intel(R) CPU 000 @ 3.20GHz
19 Intel(R) Core(tm) i7 CPU 965 @ 3.20GHz
20 Intel(R) Core(tm)2 Extreme CPU X9000 @ 2.80GHz

I haven't included the credit numbers, as credit is dependent on the project, so the values themselves wouldn't contribute to valid comparisons. Besides, the numbers are there in the BOINCstats site for everybody to see. Obviously, there are other factors that shape the lists and that we can't control: overclocking, memory speeds, heat-induced clock throttling, etc.

Here's what I make from these lists: SETI appears to utilize AMD CPUs well, although the top AMD, which is the ancient X2 4200+ (which I had myself before replacing with my i7-870!), comes at third place. WCG, with its massive population and wide application range, clearly shows Intel as the winner. Finally, Yoyo gives the win to the PS3 with its Cell 9(?)-core processor and seems to favor AMDs slightly.

So I say that my conclusion holds still: for crunching in general, Intel is better than AMD.

---

A few additional issues with those results.
For Yoyo, there are several apps and BOINCstats doesn't reveal which were being crunched by which processor.
The Cell processors are basically a cross between a CPU and a GPU, so they don't make for good comparisons. You certainly can't plug an NVidia GPU into one, so it's irrelevant for here.
The credit/h at WCG can vary by >10% depending on what CPU you have and what project you crunch for. Historically WCG has usually had 6 or 7 active projects. At present it's closer to two projects. Their apps and relative contributions to projects varies significantly.
The results are for 'logical cores'. So an i7 will be seen as having 8 processors, while an Athlon X2 265 will be seen as two processors. Each core of the Athlon may get 0.03 credits per second, but the processor gets 0.06 credits per second as a whole. An i7-3970X gets 0.019585 credits per thread. So 0.15668 credits/second for the entire processor (~2.5times more than the X2 265).

The only accurate way to measure performance is to base it on run time at reference/stock speeds per project. Then compile a list of results to show what the relative performances of each CPU (not core/thread) is like. After that you can look at performance/Watt, and then performance/system-Watt.

Ultimately if you want to crunch on a GPU, it's better to spend the money on the GPU(s) than the CPU.

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

For Yoyo, there are several apps and BOINCstats doesn't reveal which were being crunched by which processor.

We don't care about apps though, just the general performance.

The Cell processors are basically a cross between a CPU and a GPU, so they don't make for good comparisons. You certainly can't plug an NVidia GPU into one, so it's irrelevant for here.

It may be irrelevant for GPUGRID, but it's certainly relevant for judging processing performance.

The credit/h at WCG can vary by >10% depending on what CPU you have and what project you crunch for. Historically WCG has usually had 6 or 7 active projects. At present it's closer to two projects. Their apps and relative contributions to projects varies significantly.

Agreed, but the statistics do have historical significance: they are derived from processing all kinds of tasks that have ever existed in WCG, by all types of CPU that have come and gone in the lifetime of WCG. Obviously, not all combinations are recorded, since some CPUs didn't exist when some tasks were available and vice-versa, but that's why I've only considered the top 20, to isolate the best performers (which should include newer chips), or the most efficient task-CPU combinations.

The results are for 'logical cores'. So an i7 will be seen as having 8 processors, while an Athlon X2 265 will be seen as two processors. Each core of the Athlon may get 0.03 credits per second, but the processor gets 0.06 credits per second as a whole. An i7-3970X gets 0.019585 credits per thread. So 0.15668 credits/second for the entire processor (~2.5times more than the X2 265).

Are you sure that the reported Average credit per CPU second is per thread? If it is, then the lists have to change to reflect the actual whole-CPU performance.

The only accurate way to measure performance is to base it on run time at reference/stock speeds per project. Then compile a list of results to show what the relative performances of each CPU (not core/thread) is like. After that you can look at performance/Watt, and then performance/system-Watt.

In an ideal world, yes!

Ultimately if you want to crunch on a GPU, it's better to spend the money on the GPU(s) than the CPU.

Agreed, but I guess we're almost all CPU crunchers as well, so it's not an out-of-scope discussion.

---

Are you sure that the reported Average credit per CPU second is per thread? If it is, then the lists have to change to reflect the actual whole-CPU performance.

It looks fairly obvious to me, otherwise the chart is a complete nonsense.

For example,
3 AMD Athlon(tm) II X2 265 Processor 11 985,829.86 8.03 89,620.90 0.73 0.029980
...
21 AMD Athlon(tm) II X4 645 Processor 27 2,258,548.79 3,311.85 83,649.96 122.66 0.021578

We are probably looking at an X4 645 being 44% faster than an overclocked X2 265.

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Vagelis, sorry you went to all this work but these charts are obviously useless. They contain info from various apps over the years that are no longer used and had cell and GPU clients and apps that once had much higher credit awards. In some cases in the past cheating was commonplace and rampant. For instance, a core2 host in SETI or WCG running a GPU would score much higher than an i7 without a GPU. These charts include those cases. There is no useful information to be gained from these. Even a cursory glance reveals that they make no sense. Sorry...