Two 275's or two 252's?

[wagoogee]

I've recently purchased a TYAN s2895 dual Opteron mobo and need to decide which processors to buy. The system is used for gaming 50% of the time, browsing and some office stuff the other 50%.

So the question is; which is the better option to get me through the next 24mo. It's easy to make an argument for both. The 252's may be better for the next 6-12 mo, the 275's after that as games transition from single, to dual, to multi-core. People say that the transition will take a few years but I'm not so sure. AOE3 will support dual-core. Is this exception for the next 12mo or just the first of many. It's all speculation making the cpu question difficult to answer.

Thoughts anyone?

 

[Order]

I'd go for the 252s at the moment. Your applications don't really have much need for 4 processing cores. If you want to upgrade in the future you'll save some money .

 

[Moog]

Go with the 252s.

 

[big daddy fatsacks]

why are you spending thousands on a dually for gaming, browsing, and officin'?

 

[DougLite]

Even when games do become SMP aware, it will most likely be only for two cores, as consumer/gamer systems will still only have one CPU socket, with two cores. The only exception to this rule is the EE 840, which will have Hyper-threading for a total of four simultaneous threads (2 hardware and 2 logical).

With that said, and the knowledge that you aren't running workstation apps or doing content creation (if this is incorrect, please let us know), I'd go with an Opteron_1_52, then Ebay it when pricing on dual core chips becomes a little better and pick up an 875 when you get some multi-threaded games. Seriously, you don't need SMP right now, and you definitely could use the cash on something like a better video card, faster hard drives, etc.

 

[wagoogee]

Yah, I know I said "buy" but in truth, to answer big daddy's question about cost; I'm not spending a dime on these processors, they're being offered to me by my vendor to "do with as I please". If I had to pay for them then it'd be the 252's but since cost is not the issue the question isn't so easy for me to answer. Furthermore, I actually have all four processors in hand and have spent the entire weekend doing the "processor shuffle" dance. I gotta return two this week.

Douglite; excellent point about the number of sockets, I wasn't thinking in those terms. We're still a long way away from quad-core and the D 840 will sell in such low volume that no one's going to code specifically for it. The faster 252's really are attractive and the obvious choice but there's something really techy/sexy about seeing 4 cpu windows pop up. And no, no workstation apps or content creation stuff.

 

[visaris]

I don't know about your work load, but I have a Tyan K8WE and two Opteron 265's, and I find myself maxing out the box on a regular basis.

I find that getting 100% CPU usage on all four cores is really pretty easy to do. I think you should find a dual or quad box to play with for an hour or so, and keep your eye on CPU usage to see if you need the extra cores.

I know I do, and since this costs you nothing, you shouldn't pass up the four core setup unless you know you aren't going to make use of them.

 

[Dan_D]

I don't know about your work load, but I have a Tyan K8WE and two Opteron 265's, and I find myself maxing out the box on a regular basis.

I find that getting 100% CPU usage on all four cores is really pretty easy to do. I think you should find a dual or quad box to play with for an hour or so, and keep your eye on CPU usage to see if you need the extra cores.

I know I do, and since this costs you nothing, you shouldn't pass up the four core setup unless you know you aren't going to make use of them.

What could you be doing to use that much of the system, especially with four cores? I am just curious. I mean if I intended to do that I could do it, but even multi-tasking with high end applications won't usually do that.

 

[USMC2Hard4U]

Thanks for making this thread, for I was just about to ask this question.

I originally wanted 1, Athlon X2 Processor. However, looking at it now, I was going to spend 1000$ on a Vapochill System to cool it so I could get a high OC and everything. Why not just get a whole nother CPU for the price.

I am in this situation too. What to get. 2- 252s or 2- 275's.

Sigh.

i knnow this has already been asked but... what is faster. A Dual Core CPU or 2 single core CPUS....

 

[wagoogee]

USMC2Hard4U asked "what is faster"

Since I play mostly games, with light multi-tasking otherwise, the twin 252's are faster, and will be for quite awhile. But the harder question is; which is more future proof (out to 24mo). The 275's may in fact be the better choice in the 12mo+ out timeframe. It all depends on how quickly developers adjust/adapt to the new muti-core world.

 

[Dan_D]

Thanks for making this thread, for I was just about to ask this question.

I originally wanted 1, Athlon X2 Processor. However, looking at it now, I was going to spend 1000$ on a Vapochill System to cool it so I could get a high OC and everything. Why not just get a whole nother CPU for the price.

I am in this situation too. What to get. 2- 252s or 2- 275's.

Sigh.

i knnow this has already been asked but... what is faster. A Dual Core CPU or 2 single core CPUS....

In the case of X2 vs. Opteron I'd say Opteron. It has a dedicated HT link for each processor and if you get the right motherboard, each CPU gets its own ram and a NUMA aware OS, (Windows XP 64bit Edition/Server 2003 64bit Edition) could bennefit from more memory. Overall, I'd say that the Opteron is the winner. Although cost wise, the Opteron setup will cost more.

You need a case, if yours doesn't fit, you'll need an EPS12v PSU, Registered ECC ram, and a motherboard that will fit your needs. Overall, the Opteron setup probably would edge out the X2, but the cost is very high.

I would know. I am building a dual Opteron rig right now.

 

[USMC2Hard4U]

I think I am going to get the Dual 2.6's and call it a day. I was really intrested in dual core X2's however, I think 2 Single Core Optys will be good for a long time yet. Espically if I can get some good low latency ram for them, and run NUMA.

Of Corse I plan to get the K8WE. Thier really is no other choice for top end performance. And this will eliminate my questions about SCSI PCIe Card too.

So I think I will have an all around better platform here.

What is some really good ram here. i have never ever looked into Regestered ECC Ram at all. I guess Overclocking is out of the question, so I need really really low latency ram. Do they make DDR400 Regesterd ECC ram at 2-2-2 1T?

Thanks

 

[visaris]

OCZ makes some of the fastest ECC Reg I've seen.

 

[/dev/null]

Even when games do become SMP aware, it will most likely be only for two cores, as consumer/gamer systems will still only have one CPU socket, with two cores. The only exception to this rule is the EE 840, which will have Hyper-threading for a total of four simultaneous threads (2 hardware and 2 logical).

With that said, and the knowledge that you aren't running workstation apps or doing content creation (if this is incorrect, please let us know), I'd go with an Opteron_1_52, then Ebay it when pricing on dual core chips becomes a little better and pick up an 875 when you get some multi-threaded games. Seriously, you don't need SMP right now, and you definitely could use the cash on something like a better video card, faster hard drives, etc.

Have you EVER seen a 152 for sale? I haven't. If so, where? AFAIK these only exist on paper.

Rob

 

[USMC2Hard4U]

Question for you who have many years in computer with multiple different platforms and now work with a dual opteron system.

Weather you want to beleive it, endorse it, love it or not, Hyperthreading in P4 systems make the overall expirance smoother. This is not something to debate, its a fact. I loved my P4C for it. Weather it took advantage in apps or not is negligable, but overall windows performance was their. Will I get the same / Better expirance with dual Opterons? Or was it just smooth because I could execute 2 streams of data on 1 CPU?

Thanks, this is a very important question for me.

 

[Vertigo Acid]

I went from a dual 2600+ to dual HT 3.2ghz xeons, and in windows there is no discernable difference. Going to from 1gb to 2gb was a bigger performance boost than going to 4 logical CPUs. I figure that if dual MPs can multitask as good as dual Xeons, then opties should blow them out of the water

 

[Martin Cracauer]

Even when games do become SMP aware, it will most likely be only for two cores, as consumer/gamer systems will still only have one CPU socket, with two cores. The only exception to this rule is the EE 840, which will have Hyper-threading for a total of four simultaneous threads (2 hardware and 2 logical).

I disagree. First of all hyperthreading is almost useless if you have SMP anyway.

But in general I expect applications (including games) to make use of as many processors as you have, not just two. Once the program is threaded it is usually trivial to make it use more threads.

The only exception is when two threads are used for a hard seperation, like one for game logic and one for the renderer. I don't think this will prove to be a good way to go for anyone, and John Carmak actually tried it around 1998 and was disappointed. No doubt many developers will try anyway.

 

[Dan_D]

Question for you who have many years in computer with multiple different platforms and now work with a dual opteron system.

Weather you want to beleive it, endorse it, love it or not, Hyperthreading in P4 systems make the overall expirance smoother. This is not something to debate, its a fact. I loved my P4C for it. Weather it took advantage in apps or not is negligable, but overall windows performance was their. Will I get the same / Better expirance with dual Opterons? Or was it just smooth because I could execute 2 streams of data on 1 CPU?

Thanks, this is a very important question for me.

2 CPU's will give you that smoothness. Its alot better than 1 CPU with HT. Even though HT does make for smoother Windows operation.

 

[wagoogee]

So let's answer my original question; for the next few years, given the uncertainty of dual/multi core games, which is the better 24mo investment, 2x275 or 2x252? I see two recommendations for the 252's and none for the 275's. Is that the general consensus?

 

[visaris]

So let's answer my original question; for the next few years, given the uncertainty of dual/multi core games, which is the better 24mo investment, 2x275 or 2x252? I see two recommendations for the 252's and none for the 275's. Is that the general consensus?

My vote is for the 275's. I have two 265's and love them to death.

 

[cell_491]

i have a question, do you have to fill both cpu sockets in order for the system to boot? because if not id say get one of these http://www.newegg.com/Product/Product.asp?Item=N82E16819103553

 

[Vertigo Acid]

So let's answer my original question; for the next few years, given the uncertainty of dual/multi core games, which is the better 24mo investment, 2x275 or 2x252? I see two recommendations for the 252's and none for the 275's. Is that the general consensus?

I vote 252s because games *will* not be utilizing 4 cores. It wouldn't make sense because very few people have 4 cpus, whether logical or physical. Dual core chips are over-valued in the market right now anyway, and are pushing down the single-core prices as a result.

 

[DougLite]

I disagree. First of all hyperthreading is almost useless if you have SMP anyway.

But in general I expect applications (including games) to make use of as many processors as you have, not just two. Once the program is threaded it is usually trivial to make it use more threads.

The only exception is when two threads are used for a hard seperation, like one for game logic and one for the renderer. I don't think this will prove to be a good way to go for anyone, and John Carmak actually tried it around 1998 and was disappointed. No doubt many developers will try anyway.

Why do you code for a 4 core system when it is expected that the vast majority of the installed base will remain on one socket systems with two way SMP? Intel and AMD are driving dual core, but there is no way in the foreseeable future that consumer system builders are going to design two socket systems for the Dell/Gateway market. The costs are way too high - dual core raises the cost of just the chip, whereas dual socket raises the cost of basically the entire system. This, in turn, makes it suicide for a delveoper to build a game or application around four way SMP when only two way will be the least bit mainstream in the foreseeable future. Why do game developers code for DirectX 7? Because the GeForce 4MX has a HUGE installed base. Only a handful of recent games out there have explicitly dropped support for the GF4 MX, depsite the fact that it is a rehashed GF2 design that is several years old. It would be much easier to code DX 9 only, but they wouldn't sell too many copies to a broad consumer base if it wouldn't run on a GF4 MX or Radeon 9200 (the cards you see on the shelf at Best Buy, etc). There are ~40,000 of us here on this forum. Dell, Gateway, HP, etc have an installed base of easily at least 4 million. Who are you going to code for if you want to sell games? Four way SMP may gain in popularity, but two core installations, as dual core goes mainstream, will easily dwarf the four way boutique and enterprise market. Mass market consumer support drives this industry, not the enthusiasts.

 

[mkim797]

AMEN to what DougLite said. I am going with 4 core, but the reason behind that is that I run a database server and use workstation programs (Solidworks, Adobe creative suite), in addition to playing games. The way I see it, if I have to have the hardcore programs running on the darn server, why not use it to play games during server shutdown (I don't get ANY hits outside of business hours Pacific Time)? Plus everything that goes into the machine is a tax write-off

Go with a single CPU dual-core, or dual CPU single-core. The higher clock speed of the single core Opterons will make for a better experience on your part, and you won't need 4 cores running (unless you rip something like 3 seperate MP3s while playing games...but that is going to be a major hit on your game because they are sharing memory resources).

 

[visaris]

Why do you code for a 4 core system when it is expected that the vast majority of the installed base will remain on one socket systems with two way SMP?

Because it is really easy to do. I write software for living and most of the apps I write end up being multithreaded for the added performance. Some threading, such as network in one thread, and logic in another, cannot be spread to lots and lots of cores. Sure. But many kinds of threading can.

Let me give an example:
For the biology simulation I'm currently writing, there is a given population of "individuals". The simulation, based on the genetics of the "individuals", decides that certain events (mutation, reproduction, etc) are to take place with certain probabilities. This computation of probabilities takes a while, and can easily be split off to lots of work threads. Each individual's genetics need to looked at, processed, analyized, etc. So the pool of individuals is split off into threads. If there are 2 cores, and 100 individuals, the genetics of 50 of them are processed in one thread, and the genetics of the other 50 are done in the second thread. I would be down right retarded to not enable the pool of work to be split into thirds or forths, etc, just because the market doesn't currently have the parts out for it. I let the pool be slit into any number of threads (up to the number of individuals in the pool).

Your idea about the market controlling the software is correct, but shows you know nothing about multi-threaded software development (that's not an insult, very few people do). Any time you have lots of data blocks that can be processed independently, you can write one peice of code and have it work for any number of threads.

The AI for each "bot" in a game could be written to: At each time-step, examin surroundings, make decision. With this AI model (which is commonly used in lots of games), you can write code once and have it work with any number of cores/threads.

This idea that somehow no one will write code for more than two cores because all you can buy are single-socket dual-core parts is a little crazy one you figure out how threading actually works.

 

[DougLite]

Ok, I will accept that. I am certainly not a programmer, and my understanding of how code is executed, how programs flow, etc is very limited.

Just how much does complexity increase going from two CPU heavy threads to four? It was my understanding that introducing a second thread was somewhat complicated, as you can no longer allow the first thread to simply eat all of the CPU on the system, but I guess if the program is well coded, it would only use the CPU time necessary to keep each thread in sync.

Also, with your underlined statement, does this mean the program can adapt to generate a number of threads equal to the number of cores available, ie it would use two CPU heavy threads of 50 people each on a two core system but have four threads of 25 people each on a 4 core system? At least in my mind, it would be extremely foolish to code something to use all of the CPU time on a four core system, as this would cause performance to suffer on a two core system, right? Or am I totally in the dark?

Basically, my main question is this: For a programmer, is there a difference between writing a program that can scale to four cores if they are available versus having a program that is designed from the ground up for four cores?

 

[mkim797]

The AI for each "bot" in a game could be written to: At each time-step, examin surroundings, make decision. With this AI model (which is commonly used in lots of games), you can write code once and have it work with any number of cores/threads.

Your argument has a major fallacy. In programming multiple threads in a scientific model, synchorization doesn't have to happen in real-time. You apply the model, split the processing, and if one process ends up finishing before the other, it can simply wait for the other processes to catch up (there are exceptions, but in those cases the programs are programmed EXCLUSIVELY for that task). In games, however, if the AIs must be synchronized with the game (sound, graphics, etc). It is the synchronization that hasn't developed quite that well yet. Even advanced CAD programs are only mildly multi-threaded and lots of CAD workstations ARE multi-CPU workstations.

 

[wagoogee]

However,

I too have written multithreaded apps and once you solve the synchronization problem for two cores, if you've coded it correctly it's easy to scale it up to n-cores. The hard part is going from one to two, not from two to n.

With that said, it's entirely unclear to me whether the game developers will simply try to "cut in half" their code or do it right from the ground up. And let's not forget; AMD has stated that we'll see quad-core in 2006, right around the time we'd expect to see dual-core games hitting the market in force. So if I'm a developer, developing a major title for delivery in summer 06, knowing that by then quad-core may be out, do I develop my game for dual-core or n-core?

It's this question that makes my choice so difficult and why I'm looking 24mo out.

 

[visaris]

Also, with your underlined statement, does this mean the program can adapt to generate a number of threads equal to the number of cores available ... ?

That is true. There are function calls one can use to find out how many cores there are available for use in the system, and you can dynamically create and kill threads on the fly based on that information. You could also create for kill threads based on CPU load, or anything else you wanted to for that matter.

Basically, my main question is this: For a programmer, is there a difference between writing a program that can scale to four cores if they are available versus having a program that is designed from the ground up for four cores?

There is. I'm not trying to say that everything can be dynamically scaled the way I describe above, but many things can. If you code smart, (and your algorithm is parallel enough) you should be able to write an app that can dynamically use anywhere from 1 to about 8 threads just as easily as you could code an app that only uses 8 threads.

Your argument has a major fallacy. In programming multiple threads in a scientific model, synchorization doesn't have to happen in real-time. You apply the model, split the processing, and if one process ends up finishing before the other, it can simply wait for the other processes to catch up (there are exceptions, but in those cases the programs are programmed EXCLUSIVELY for that task). In games, however, if the AIs must be synchronized with the game (sound, graphics, etc). It is the synchronization that hasn't developed quite that well yet. Even advanced CAD programs are only mildly multi-threaded and lots of CAD workstations ARE multi-CPU workstations.

Ah, that's a great point! Thanks for pointing out a minor flaw in my thinking (I'll get to the minor part later). That is exactly what my simulations do. If three threads finish before the fourth, they wait. I do see that this places a limit on scaling, but you have to admit one thing (and this is the reason why I call it a "minor" flaw): even in a game, where three threads may be waiting for the last one to finish, you have to admit that most likly the total time it took for all four threads to finish is probably less than the amount of time it would have taken one thread to do all that work. If you can do a reasonable job splitting up the work, the gain from the extra cores will outweigh the loss incurred by thread syncronization.

With that said, it's entirely unclear to me whether the game developers will simply try to "cut in half" their code or do it right from the ground up.

That is exactly what the question is. Well, I guess half of it. How well will games really be able to scale? I'm certainly not a game dev. I'm just making an educated guess from my not-at-all game related programming experience. If you can answer that question then the next question to ask is exactly what you said: will the devs do it right?

 

[DougLite]

What Visaris said explains a lot. However, I can see game developers taking the expedient way out, hard coding two threads that chew up all the CPU time they can get, and calling it a day. This is what I'm afraid of. Obviously, the better developers will do a better job of coding and adapt their programs to the capabilities of the system it is running on, make efforts to keep the threads in sync, limit memory and CPU usage to what the program actually needs, etc, etc. I have noticed more and more games that, at least at first glance and in my limited understanding, are better coded. They actually free memory and only use the CPU time they need. However, many titles out there lock CPU use at 100% as soon as they load, have memory leaks, and other problems related to poor programming and/or quality control. When they have two cores to play with in mainstream boxes, one of two things will happen. They will get smart and run clean multi threaded code, or they will get even sloppier. I hope it's not the latter, and that's why I posted my concerns about multi-threading games and consumer applications.