Dual Core AMDs Not for Gamers...Why Not, *if You OCed Them?*

[DrCR]

I was just thinking about this a while back. Basically anyone knows a dual core AMD @ 2.0GHz isn't going to make your games (or at least, 99% of them) run any faster than the single core sibling @ 2.0GHz.

Now I was thinking...why just get a dual core and overclock it to the same levels you would overclock your single core anyway? The additional core will still just be on for the ride when your gaming, true, but it would be really sweet to have for things like video encoding. And it would also give you a little more future-proofness as well I would think.

Now sure, that dual core is going to be seriously cooking. But isn't that just where a setup like an Iwaki MD-20Z with a G4 or G5 Storm would really be in its element?

One things to consider I'm sure is that one of the cores might not like to OC as high as the other. So it would be even harder finding a dual core that's also an excellent OCer. Kind of like how the more memory you have on your video card, the more likely you are to get a module that doesn't want to go as high as the others.

Any thoughts? I would post this at one of the other forums I participate in that have a CPU or OC section, but I wanted a perspective from guys that know watercooling.

Thanks,
DrCR

__________

[tong]

well if the dual cores might not oc as high as single then pls explain the pentium hyperthreader (basicly 2 cores in 1) and how that thing could oc like crazzy.

[nikhsub1]

Dual cores won't be as good for games as the FX55. Dual core (SMP in general) is not optimized for gaming yet, plus you have some latency with the duals that you don't see in a single CPU setup. Once games are multithreaded, this won't be the case.

[Breach]

I think there is a misconseption about Dual core chips and that "2 is better than 1". These have twin cores yes, but unless the application is programmed to utilize both it will not do any good. And even if, it doenst use both cores like a single processor, it splits up the work load between them. Dual core chips are no differant than SMP configurations with 2 seperate processors.

Therefor, that is why it is not the "ultimate gamer" chip, you would still be better off with a solid single core processor.

[DrCR]

OK, so unoptimized architecture and more latency.

Hum, I wonder how much of a difference both of those would make. I've got to find some benchs with a dual-core and F55 OCed to the same frequency. Unfortuantly though, I don't think any such benches exist yet.

Thanks.

[DrCR]

Originally Posted by Der Brewling

I think there is a misconseption about Dual core chips and that "2 is better than 1". These have twin cores yes, but unless the application is programmed to utilize both it will not do any good. And even if, it doenst use both cores like a single processor, it splits up the work load between them. Dual core chips are no differant than SMP configurations with 2 seperate processors.

Therefor, that is why it is not the "ultimate gamer" chip, you would still be better off with a solid single core processor.

Yes, yes, I already mentioned that for gaming and other single tasks, the second core wouldn't do a thing. That's pretty obvious. The code just isn't designed to utilize an additional (or multiple) "CPUs". But unless the two issues nikhsub1 presented are fairly significant (which might be the case), a dual core OCed clocked at the same speed of an OCed FX55 would perform basically the same -- in games and other non multi-CPU/core optimized softwares.

Now even if those two issues are only trivial, the OCed dual-core would still, by definition, not be the "ultimate" gaming solution because of those two issues. But it would be awfully close (again, assuming trivial in this case) and would give the user a lot more power for things like video encoding (which I do on a fairly regular basis), rendering, and such, and prove to be a getter choice with the future in mind. More “futureproof”. I guess this last benefit isn't an issue for those who layout 2k+ every 6mths or year for a new rig though.

Remember, were talking about an OCed dual core. Let's assume it has the same max OC potential as the FX55…. My question is: Just how inferior would the dual core be in this scenario? How significant are those afore mentioned issues? Any additional cons?

Thanks again.

[jman1310]

i think the memory bandwith is split evenly between the cores

and that would be very BAD for performance




***read this about about intel solution - don't know about AMD***

[bobkoure]

How exactly is a dual core different than multiple processors?
There are speed problems in inter-thread communication when the threads are spread cross-processor.
Apps like photoshop don't have a problem with this because many of the tasks are not particularly interactive and usually run to completion. To the extent games are not like this, multiple processors (and multi cores?) they will not be helped.
Anyone know about game-app architecture? I know a bit about windows and multi-processor interaction, all-be-it pretty dated (WinNT drivers for multi-processor machines - windows architecture hasn't changed much since then but the msoft tools/APIs may have).

Speaking of multiple cores, I did some work in the mid 80's with something called a "transputer", which had lots of processors (and you could add banks more) all tied together with what was essentially a store-and-forward network (so sidestepped the whole shared memory issue). Great little machine. Anyone else work on one? I think the FBI is still using one to scan for fingerprint matches (they were really, really good for parallel tasks). Wonder how many processors they have netted together now - I have a hazy recollection of there being an addressing limit of 65536... Oh - did some hypercube and "butterfly" work about then too - anyone else? As I remember the hypercube was an intel project, so maybe it'll get re-incarnated (and maybe that's what lead to hyper-threading, I dunno...).

[pHaestus]

I don't think you CAN assume that the dual core will have the max OC potential. Look at the obstacles:

1) The memory controller of A64s is already pretty limiting to overclocks. There's lots of fiddling to get the last bit of oc stable, and it always seems to be with memory timings and settings for me. This is going to be worse if the same controller is dealing with 2 cores. And there's a finite amount of communication between the mem controller and the CPU that can happen; is this bandwidth so far above CPU requirements that we'll never hit it?

2) Good lord the heat. 190 is the speced amount for dual core 2.2GHz chips. I use a retail hs on my Winchester 3000+ (1.8GHz) at 2.5GHz and it stays under 50C. I dont think I'd be able to do the same with a dual core 1.8 chip.

3) Price. If you're dissipating 190W at stock dual 3500+ speeds, then what kind of PSU will you need? What kind of extra precautions will mobos have to take to keep onboard parts near the socket from overheating? How stable will the voltage regulation and supply be? This is gonna be expensive, and it is probably why they are introducing these chips to the server market first (Tyan can put the extra quality into their opteron boards because they are charging big bucks for them).

Think about the following scenario: If you are a gamer would you rather have a dual 2.25Ghz 940 setup with a 6600GT or a sincle 939 chip at 2.5Ghz with a 6800GT? The answer is pretty clear, and I think this is the real decision people will have to make on the AMD side. For a fixed computer price, is it better to spend more $$ on multitasking at the expense of raw performance in single apps, or to spend more $$ on upgrading other components and stick with single CPUs?

Or should you just say screw both options, stick with current PC, and get an xbox 360?

[andy497]

I wouldn't rule out that games will start using multiple threads. There are lots things you could do in seperate threads e.g. physics, ai, sound processing. Running those in parallel could help a lot, but of course you have to add in the administration costs and latency for resources, threads waiting for each other, etc. Also parallel programming is really hard. It's no good if threads end up waiting for resources or output from other threads more than they're working. Even if you design with the best foresight and intentions, you may still end up with a lot of this in the end.

On the other hand, a dual core means your folding@home will still be maxed on one core while you're playing CS:Source, which is pretty cool.

[Joe]

I am an SMP fiend, I havent had a non SMP rig as my main box in nearly 11 years now.

The simple fact is that expect games to use one processor, but being able to do other stuff at the same time is OHH so nice. I can rip DVD's, wile playing BF:Vietnam with no slowdown in the game. I normally do like 12 things at once on the box, and really never have lag caused by any application taking over the resources on the box. While the peak speeds arent as high as systems with a single cpu clocked faster, it just never slows down or lags as many single processor boxes do when you are doing one intense task.

Its like the difference between peak HP in a car and peak Torque. SMP = Lots of Torque, Single high speed processor = lots of HP.

and as its said, Hyperthreading is NOT 2 cpu's. Its just using areas of the CPU that werent in use for the process running to alow other threads to execute.

[Joe]

Oh and Dual Core and Dual processor's behave almost no different from what all the tests have shown. a Dual core rig is a true SMP rig, not the voodoo hyperthreading crap.

[pHaestus]

Yea but Joe you have to admit that for the $$ you were talking about to go dual opteron you could build 2 PCs: 1 for gaming and 1 for other stuff and just get a KVM.

[Joe]

Yep, but KVM's are ass I would rather have one powerful machine that can do the work of 2 boxes, in one smaller space, making less noise, and having to invest in only one high end disk IO setup. KVM's also arent cheap when you need one to switch USB devices, DVI links to monitors, and 5.1 sound. To be able to use all the features of 2 machines is expensive on its own.

[aaronspink]

Originally Posted by jman1310

i think the memory bandwith is split evenly between the cores

and that would be very BAD for performance




***read this about about intel solution - don't know about AMD***

Incorrect.

Aaron Spink
speaking for myself inc.

[aaronspink]

Originally Posted by nikhsub1

Dual cores won't be as good for games as the FX55. Dual core (SMP in general) is not optimized for gaming yet, plus you have some latency with the duals that you don't see in a single CPU setup. Once games are multithreaded, this won't be the case.

If the DC is running at roughly the same frequency of a single core counterpart, then the performance difference should be negligible.

The main limiter on frequency for all DC designs is thermals.


Aaron Spink
speaking for myself inc.

[aaronspink]

Originally Posted by pHaestus

I don't think you CAN assume that the dual core will have the max OC potential. Look at the obstacles:

1) The memory controller of A64s is already pretty limiting to overclocks. There's lots of fiddling to get the last bit of oc stable, and it always seems to be with memory timings and settings for me. This is going to be worse if the same controller is dealing with 2 cores. And there's a finite amount of communication between the mem controller and the CPU that can happen; is this bandwidth so far above CPU requirements that we'll never hit it?

Timing paths between the cores and the memory controller should be minimal. Most of the timing paths you are referencing sound like they are in the design of the memory controller itself.

3) Price. If you're dissipating 190W at stock dual 3500+ speeds, then what kind of PSU will you need? What kind of extra precautions will mobos have to take to keep onboard parts near the socket from overheating? How stable will the voltage regulation and supply be? This is gonna be expensive, and it is probably why they are introducing these chips to the server market first (Tyan can put the extra quality into their opteron boards because they are charging big bucks for them).

I highly doubt the 190W number. It is simply non-viable.

Think about the following scenario: If you are a gamer would you rather have a dual 2.25Ghz 940 setup with a 6600GT or a sincle 939 chip at 2.5Ghz with a 6800GT? The answer is pretty clear, and I think this is the real decision people will have to make on the AMD side. For a fixed computer price, is it better to spend more $$ on multitasking at the expense of raw performance in single apps, or to spend more $$ on upgrading other components and stick with single CPUs?

A lot of it is short term vs long term. If you upgrade on a yearly basis, then this is probably the correct recomendation, but if you upgrade on a semi-yearly basis, you will likely be better off with the dual core option as all the game physics packages become heavily multi-threaded. A good number of games coming out in 2006 will be able to take heavy advantage of multiple hardware contexts.

Aaron Spink
speaking for myself inc.

[nikhsub1]

I have heard some talk of being able to disable one core in a multicore CPU to allow less heat and better single threaded app. performance. This likely won't happen for a while, but that would be a very nice option IMO.

[mastermind2004]

AMD is claiming that their Dual core offerings will follow the same limits as existing single core. I believe that means 90-110W max.

[JamesAvery22]

http://www.anandtech.com/cpuchipsets...spx?i=2397&p=2

Pentium D

"Any communication between the two cores has to be done over the external FSB, and obviously, core-to-core communication over an external bus is slow. "

AMD Dual Core

"Instead of having all communication between the cores go over an external FSB, each core will put its request on the System Request Queue (SRQ) and when resources are available, the request will be sent to the appropriate execution core - all without leaving the confines of the CPU's die."


Anyways... is 2 > 1? Depends on what you do. If all you do is play games then no. If you fold? Yes.

[jman1310]

Originally Posted by jman1310

i think the memory bandwith is split evenly between the cores

and that would be very BAD for performance

Originally Posted by aaronspink

Incorrect.

Aaron Spink
speaking for myself inc.

"The one limitation that both AMD and Intel have is bandwidth. In order to maintain compatibility with present day Socket-940 and Socket-939 motherboards, AMD could not increase the pincount of their dual core processors. The benefit is that AMD's dual core CPUs will work in almost all Socket-940 and Socket-939 motherboards (more on this later), but the downside is that the memory bus remains unchanged at 128-bits wide and supports a maximum memory speed of DDR400. So, while single core Athlon 64 and Opteron CPUs get a full 6.4GB/s of memory bandwidth, today's dual core CPUs are given the same memory bandwidth to share among two cores instead of one. " quoted from anandtech


seems like i was right, what's incorrect?

[Butcher]

If one core is not using bandwidth then the other gets it all (I assume).
However SMP setups do have additional latencies and stalls due to cache coherency issues. These should not be ignored. Basically if core A has a piece of memory cached and core B wants to access it, then it has to be flushed to memory, that's a lot slower than just keeping it in the cache. how much this occurs depends on the apps being run.

Originally Posted by bobkoure

Anyone know about game-app architecture? I know a bit about windows and multi-processor interaction, all-be-it pretty dated (WinNT drivers for multi-processor machines - windows architecture hasn't changed much since then but the msoft tools/APIs may have).

I write games for a living so I guess you could say I know games architecture.
Exploiting SMP is certainly possible in games, but it's not easy by any means. Everything has to run quickly with low latency in general which makes multiple threads difficult. Games are already a juggling act between the CPU and GPU, adding another CPU only adds to the complexity. To an extent though I expect to see things like phyics and similar simulation tasks pushed off to extra cores. Since both XB360 and PS3 are parallel machines this sort of thing will become the norm for games. It won't be long before the PC market follows suit. the main limiting factor is the cost of doing it - games are expensive, adding to the complexity will further increase costs, which means it may get done in a half-assed manner. That could end up worse than not doing anything at all.

[DrCR]

Hum, so IF you got a dual-core sample with a great memory controler that didn't mind breakneck OCing, and IF you could take the heat (i.e. G4/G5, MD-20Z, two 2x120+ heatercores, etc.), and IF you had the PSU and mother board to do the job...Then yes, a hardcore gamer use a dual-core rig with only negligible side effects. Correct?

Price? Well, true, that's an issue. If you crunch the numbers, it could make more since to get a 3500+ box and wait for cheaper dual-cores... but that's no fun lol. Besides, some batches of the cheapest dual-cores could prove to be OCers, and then you would get scenarios similar to most every CPU family of yore- cheaper lower rated chips that OC to the same max as the higher rated, more expensive chips.

I will admit I'm optimistic, perhaps too optimistic, on some batches of cheaer dual-cores OCing well or at least decently. I guess that's just because that's what AMD has been giving me/us for quite some time now. At any rate, the more I think of it, the more I like the idea of being able to use my computer for something genuinely profitable while wasting my time gaming. Hey, I'll wouldn't mind loosing a little OC just for that: guiltless gaming lol.

DrCR

_________

[pHaestus]

I expect we'll see things like much more involved system protection running on one core as well: antispyware, firewall, antivirus, system diagnostics, etc with minimal performance hit. That will be something that both the chip mfgrs and MS can easily sell to the average consumer.

[eander315]

"well if the dual cores might not oc as high as single then pls explain the pentium hyperthreader (basicly 2 cores in 1) and how that thing could oc like crazzy."


Hyperthreaded P4 CPUs do not have dual cores, and have nothing to do with dual cores. Thus they overclock the same as any other P4 processor.