Super User is a question and answer site for computer enthusiasts and power users. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

My current computer is less than a year old and is showing signs that the motherboard has a defect or something. I am not 100% sure at the time but that is not in scope of the question.

Basically I am wanting to replace my current system with another system (CPU, Memory and Board).

Right now I have a socket 2011 board with an i7 3820 chip and 32GB of ram (ram was cheap!).

I am thinking about moving to an AMD FX 3850 Vishera chip.

The issue I am running into is trying to understand if and what I am loosing or gaining if I move to AMD from Intel or vise versa.

For example:

My Intel i7 3820 is a Quad Core CPU with Hyperthreading. To the best of my understanding hyperthreading is a thread execution optimization. Meaning the CPU decieds which order queued thread should be executed in to achieve the best performance. Is this right?

What trips me up is the fact that the CPU only has 4 cores but yet it registers in Windows / Linux as 8 cores. But it only has 4 CORES? How can this be?

The chip I thinking about buying is an 8 Core CPU with 8 physical cores but I am thinking if it is anything like the previous chip (The FX 8150) it will have 8 Integer processors and only 4 floating point processors shared between pairs of cores.

So how does one make sense of which chip is dare I say better? No that does not sound right. How does one figure out which chip would serve their purpose better?

Basically I am looking to use the machine for Beginner Java development, heavy on VMware Workstation (up to 5 machines at once), and various gaming titles ranging from hardcore FPS to minecraft.

Gaming aside because that focuses more on the GPU. Does 8 real cores really beat the hyperthreading on intel? From a VMware perspective does 8 real cores make a difference?

Can anyone lend me some insight to how or where one can find information to make a decision for something like this. I am more trying figure out the differences between the CPU. Do not consider motherboards or memory into this.

share|improve this question
If its a motherboard defect, swapping the motherboard might be an option – Journeyman Geek Jan 31 '13 at 7:03
I was thinking of that anyway. The board is under warranty. But I was just gonna sell it. – ianc1215 Jan 31 '13 at 7:13
I'd recommend a LGA 1155 based intel board then. I'd note that chat's the recommended place to talk about hardware here, assuming you want specific recommendations (which you do, right?) – Journeyman Geek Jan 31 '13 at 7:16
Well I am kinda of looking more for guidance not so much a "buy this" conversation. The biggest thing I am trying to understand is 2 things. How does the new AMD stack up to Intel Ivy Bridge and whether 8 "real" cores stomps hyperthreading when leverging things like virtualization very heavily. On a sidenote I am conserned with lifespan of the socket. I don't want to be stranded because they decide to go EOL on a socket next year. – ianc1215 Jan 31 '13 at 7:20

What trips me up is the fact that the CPU only has 4 cores but yet it registers in Windows / Linux as 8 cores. But it only has 4 CORES? How can this be?

It has 4 physical cores, each of which has two virtual cores. The operating system schedules threads on the virtual cores. The CPU schedules execution units inside the physical core to instructions from the two threads running in that physical core.

To put it another way, hyper-threading allows two threads to run in the same physical core. This can be a big win if the two threads are doing different things. For example, if one thread is doing lots of integer math, without hyper-threading, the floating point units in that core would be idle. With hyper-threading, the other core can use them. However, if the two threads are doing similar things, they'll just fight over the execution units and each will run at about half speed, yielding little to no gain.

Under realistic conditions, the current generation of hyper-threading adds about 15% to CPU performance if your software can generate enough threads to run two in each physical core. (Compared to the same physical cores with no hyper-threading.)

share|improve this answer
But does that 15% boost compare to have 2x the physical cores to do the lifting rather than some clever CPU scheduling? Asking the same question I cant seem to get over. 8 Real vs 4 Real + 4 Virtual – ianc1215 Jan 31 '13 at 8:05
@Solignis: I'm not sure I understand what you're asking. What I'm saying is that 4 cores with hyper-threading will give you about a 15% boost over 4 cores without, assuming you can actually run 8 threads (two per physical core). – David Schwartz Jan 31 '13 at 8:07
Sorry its late, I re-read what you wrote. I think I get it now. So you can't really compare an 8 core chip like the 8350 to a quad core hyper threaded cpu like the 3750k, right? – ianc1215 Jan 31 '13 at 8:24
Well, you can. But you have to think of the 3750k as a quad-core chip with a feature that makes code with lots of threads run a bit faster. However, the 3750k's single-core speed is much higher than the single-core speed of the AMD CPUs you're comparing it to. One big downside of the AMD CPUs is that if your code can only take advantage of 2, 3, or 4 cores, the extra cores on the 8-core don't help you much. The faster Intel cores win big in that case. (But if you can fully use all 8 cores, then the AMD CPU can win.) – David Schwartz Jan 31 '13 at 8:27
Ah! Ok I understand that point. But doesn't the OS schedule processes from the system and attempt the allocate the entire CPU? Or is that an incorrect understanding? – ianc1215 Jan 31 '13 at 8:29

The AnandTech review of Vishera has a ton of information on how these processors stack up.

Personally, I'd sum it up this way: if you can stress all 8 cores equally, compared to all 4 cores of a comparable Intel processor, you can gain something from switching to Vishera. However, in the vast majority of situations, that won't be the case (gaming especially, but also software development), and Intel's superior singlethreaded performance wins. Don't get me wrong, there are advantages to AMD (price especially), but in most situations an FX8350 is definitely going to be slower than your i7-3820.

One example, compiling Firefox:

All credit to Anandtech

share|improve this answer
Yeah I do realize that Intel most of the time is going to take AMD to the cleaners. Personally it almost seems like a fact of life. What I am more wondering if for things like VMware / QEMU and I better with hyperthreading or more cores? Because one half of me says 8 "real" cores even if they share FPUs (Floating Point Units) is better than 4 Cores with 4 "fake" cores. But I know the rabbit hole goes much deeper. – ianc1215 Jan 31 '13 at 7:24
Hmm, that is a really good question. discusses HT in virtualization and gives a rough +0-40% performance estimate; it would be tough to compare that to a full 2x core count and lower ST performace... – Marcus Chan Jan 31 '13 at 7:36
That document is good but does not apply to me because I would be running VMware workstation 9, not ESX which is a custom linux install for baremetal servers. – ianc1215 Jan 31 '13 at 7:44
Thanks for that AnandTech read. It was very informative and gave me a good insight into what the AMD 8350 has for pros and cons. Also it gives me an idea of where it sits in the Intel hierarchy. – ianc1215 Jan 31 '13 at 7:57
@Solignis (I was just correcting Marcus' 'full 2x cores'...) it's definitely going to offer a greater improvement than Hyper-Threading - but, taking into account the single-core performance (remember, a percentage improvement depends on what the base single-core performance is), Intel's high-end offerings are simply more powerful for the vast majority of use cases. On the other hand running enough (busy, since idle VMs don't count) VMs, or compiling large enough projects, might max out the cores - but, I say plan for the most common case. – Bob Feb 1 '13 at 12:56

One of the issues you're going to face is that most modern systems are pretty much powerful enough that with most workloads, you're not going to be processor bound. A lot of things that might have been important a decade ago (amount of cache for example) are less important than they used to be. Nearly any modern machine should be able to handle most of what you throw at it. VT-X is nice for virtualisation (and nearly standard). VT-D allows some neat stuff (like allocating a physical PCI-E device to a VM), but isn't supported in VMware. In short, unless you push your system hard, there's not much that recommends a more powerful processor over one thats a slot or two below.

While its not the first thing that comes to mind performance per watt is a pretty important thing to look at. As things are, intel's processors are significantly more efficient when it comes to most workloads. My core i7 has a tdp of 75 watts, and i uses between 10-65 watts.

AMD's approach with the fx series is somewhere in between SMT/HT and having two seperate physical cores in a 'module'sharing a interface. The X86 is not as good at floating point calculations, so this might make sense (intel on the other hand, uses HT since they figure that not all the processor is being used at once). If you're going to stress out all the cores, with maximum usage (I donno, maybe for a render farm), AMD might win out. However, they've seemed to have focused on raw clockrate and those cores, while there's more of them, arn't quite that efficient.

Techreport did some benchmarking on games here and in most situations core i5s beat the AMD systems. The gaming when video coding charts should be the interesting one here - while the 8 core AMD did better than many of the other AMD processors (but not the Phenom II X4 and X6), both the quad core core i7s and the dual core core i5s did significantly better with this workload than the AMDs. I'd guess there's some more polishing needed.

Edit: I note that cost was brought up. While the highest end AMD processors are cheaper than their Intel equivilent (say, a 3820, and its contemporary, say a ivy bridge core I 7 3770), in many of these tests, a core i5 or less gives comparable performance to the AMD processor. If we're looking at cost, cost/performance is probably a better metric in most cases - Techreport's review of the 3820 has a pretty interesting chart on price-performance in both overall and gaming contexts that backs this up.

share|improve this answer
You make some excellent points about efficiency. But sadly the 800 pound gorilla is price. For the sake of this posting I tried to avoid the price argument but that is really hard to do. I mean even if AMD's latest chips are not as efficient as the latest Intel chips their price point is hard to beat. Something else that is affecting my decision as I read is socket lifespan. AM3+ seems to be sticking around for a little bit. While I am not sure what will happen to LGA1155 is the future. – ianc1215 Jan 31 '13 at 7:55
Oh yeah I should probably mention to I will end up overclocking so power efficiency is not really a concern since I will end messing with the power consumption anyway. – ianc1215 Jan 31 '13 at 7:59
added some information on price/performance. The techreport article should be pretty good reading, and backs up what everyone's said – Journeyman Geek Jan 31 '13 at 9:29

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .