CPU price/performance graph

[HotFoot]

The difference in price between an adequate overclockable Intel-compatible motherboard and AMD-compatible motherboard should be about $20, at most $50, but that's pushing it. This is an order of magnitude less than the extra cost of the QFX motherboard, or even of the price of many of the processors on the list. Other factors are greater, such as if you don't live in the U.S. and newegg doesn't deliver to you, what price do you actually have to pay? Data points are not actually points. They are estimates and if I were doing this scientifically I would have error bars in the X and Y axes directions.

The other negating factor is the cost of memory. If you're going with the Intel, then you can get 95% of your maximum performance with DDR2-533 memory. Currently, the cheapest memory with at least those specs is this DDR2-667 (not that I'm personally recommending the brand). It currently goes for $64.99. The cheapest DDR2-800 RAM I can find is $84.99 for the same 2X512MB size, and this is exactly the $20 difference you'd have saved buying the AM2 board. It's hard for me to recommend getting only 1GB of RAM in either case. The difference for 2GB would obviously be $40.

The above reasoning provides two reasons for not including the difference in price between a budget Intel and AMD motherboard:

1) The difference is an order of magnitude smaller than the magnitude of the data itself.

2) The difference in price that does exist is at least partially negated by the difference in price of the memory suitable for best performance in either system.

I hope this satisfies your concern. I am sure you're not alone. I thought hard about it.

 

[flasher702]

Typo: You have the e6400 as a $22 chip.

 

[HotFoot]

Oops, my bad. I'd caught that number when the graphs didn't look right, but I must have missed updating the table properly. Sharp eye!

 

[fishboi]

This graph again says that the E6600 is the best bang for your buck when it comes to gaming performance. It looks like the derivate point of the price/performance function.

General performance is something different. Nice jump in the Quads. Thanks for the effort.

 

[halbhh]

Without the blow-up graph (of the lower quadrant), it is not possible to read many of the processors on the graphs. They overlap each other and are not legible on a 1280x1024 19" LCD (the most common type of monitor). The result is only a general vague impression for many of the chips, and a clear picture for some of them. For instance, the e4300 is superimposed with another chip I belive. This is my guess, since I don't see it. Several AMD chips are illegible also.

This is why I suggested an expansion of the bottom half, not for any other reason.

 

[HotFoot]

I will be putting the expanded section in the next update. Right now it is just a matter that updates are very frequent, and the more images I show, the more I have to work just to post updates. Since updates shouldn't be as frequent from now on, I will be able to be much more thorough without consuming too much of my time.

Are you able to open the .fig files in the .zip download?

 

[halbhh]

Yes, I could download and unzip. I don't have matlab.

I have a question. Reading reviews like this one (note this link is to the conclusion section of a cpu gaming review with the 8800 gpu):

http://www.legionhardware.com/document.php?id=618&p=6

It's clear that for many of the most popular games the C2duos don't drastically outperform the A64 X2s at the same price points, as this review (link given from an Intel fan, no less) demonstrates with top graphics cards.

This begs the question of whether your gaming benchmarks are representative of gaming. It isn't statistically proper to select only games that maximize a difference (although interesting). Instead the set of games should be representative of the most widely played games. Perhaps if you average in 2 of the most popular games it would be more representative.

 

[HotFoot]

I currently am using Quake and FarCry, since I was able to find data on those. I can add in F.E.A.R., which is also quite popular and I already have a lot of data for it, but in general that game is very GPU bound and doesn't show as much benefit with faster CPUs. The F.E.A.R. benchmark, then, doesn't demonstrate the strength/value of a CPU.

However, benchmarks are rather concocted in the first place. Most CPU gaming benchmarks are done at a lower resolution (like 1024x768) in order to avoid being GPU bound. On the one hand, GPU-bound benchmarks are of poor value when trying to determine the speed of a CPU. On the other hand, the value to the consumer is how the system performs at the resolutions they'll actually use. If I were building a gaming system for myself, I'd probably get either the e4300 or e6300 on Intel side or any of the <$200 chips from AMD and spend my money on a good graphics card.

So perhaps for my next update I should put in the F.E.A.R. benchmark, as this one demonstrates the uselessness in buying more CPU power than games need. I really do look forward to multi-threaded gaming benchmarks down the road!

I would like to put up a poll where you can vote on what combination of benchmarks are used for both gaming and general scores. However, the poll would need to change over time, and that's not an option in a forum. For now, I'll just ask for a second to the idea of putting F.E.A.R. benchmark results into the gaming index. I'm assuming halbhh is in favour.

 

[halbhh]

I currently am using Quake and FarCry, since I was able to find data on those. I can add in F.E.A.R., which is also quite popular and I already have a lot of data for it, but in general that game is very GPU bound and doesn't show as much benefit with faster CPUs. The F.E.A.R. benchmark, then, doesn't demonstrate the strength/value of a CPU.

However, benchmarks are rather concocted in the first place. Most CPU gaming benchmarks are done at a lower resolution (like 1024x768) in order to avoid being GPU bound. On the one hand, GPU-bound benchmarks are of poor value when trying to determine the speed of a CPU. On the other hand, the value to the consumer is how the system performs at the resolutions they'll actually use. If I were building a gaming system for myself, I'd probably get either the e4300 or e6300 on Intel side or any of the <$200 chips from AMD and spend my money on a good graphics card.

So perhaps for my next update I should put in the F.E.A.R. benchmark, as this one demonstrates the uselessness in buying more CPU power than games need. I really do look forward to multi-threaded gaming benchmarks down the road!

I would like to put up a poll where you can vote on what combination of benchmarks are used for both gaming and general scores. However, the poll would need to change over time, and that's not an option in a forum. For now, I'll just ask for a second to the idea of putting F.E.A.R. benchmark results into the gaming index. I'm assuming halbhh is in favour.

While it makes good sense to look at least 1 or 2 games that are cpu demanding, it makes no sense to look at resolutions that are below the standard for new computers. For this reason you should not use any benches below 1280x1024 (which even 17" LCD do).

I doubt you'll find even a single person interested in upgrading or building that would go for a lower resolution. Instead, there is a different question:

Is even 1280x1024 enough!

Actually, 1600x1200 will soon be the standard, so it's the most useful resolution for a benchmark for someone building, since they want their build to be valid for more than only a few months.

I know that most of the reviews I've read included a 1600x1200 data point, and many had even higher resolutions.

But 1600x1200 is a nice point, since it would be the most common I bet in new computers withing a few months.

btw, it'd be nice if you didn't have to enter data for each graph. Software should be better than that!

 

[HotFoot]

I re-checked, and the F.E.A.R. benchmarks (taken from THG since HW.info doesn't have them) are at a resolution of 1280x920. The other configurations are most likely at lower resolutions.

While it would be great to have a standard 1920x1200 resolution for any modern benchmarks, data availability has to be considered. I'm not running the benchmarks myself, and I have to scavenge what I can from the internet. Take close look at the "Mapping" worksheet and you'll see where the real work comes in.

By the way, I've just updated all the graphs, put back the zoomed-view, and put up a new .zip file. In the updated spreadsheet you'll see that I have started to organise the mapping process a little better (towards the bottom of the worksheet). Also, I am now taking advantage of more redundant data and I use averaging to improve the confidence in the interpolated points. On the "Sheet 1" worksheet, the numbers in blue italics are the interpolation points that have received the more sophisticated treatment.

 

[halbhh]

I'm used to seeing game benches for 1600x1200 in the reviews I read (quite a few!), and I actually pay the most attention to that resolution, since it is the one I will have soon enough that I'd want to be ready for it.

I think 1280 x anything is reasonable though.

On the whole, let me say: great work.

 

[jamiepotter]

Wow. This is really impressive work.

Another nail in the AMD 4x4's coffin, I'm afraid to say, is that they require more poiwer and thus generate more system heat. Probably a negligible effect in terms of P/P, but it is well to remind yourself of this if you start to think there might be a Intel bias at play.

I wonder if the new Intel chips will make much of a difference? There seems to be little point, if you're OCing, going for anything higher than an E4300, from these graphs.

 

[halbhh]

No one here on the forum has suggested AMD's 4x4 is a good choice so far as I know anytime recently, exactly because of cost and power-use and performance vs the quad intel, so that's kinda a non-issue, old news, etc.

But recently I have pointed out a new thing. With Barcelona expected to match and better the 2duo arch, 4x4 actually has a certain interest....you can have 8 cores of the best new arch!

That's kinda a new view of the 4x4, and so what you say is both old and beside the new point, actually.

Agree about the 4300 though, as I mentioned further up.

 

[jamiepotter]

OK, calm down...

 

[halbhh]

hmmm...if I make a post about another nail in the coffin of P4 netburst arch, I'd expect someone to point out it's kinda old news.

 

[jamiepotter]

OK. So P4 came out in...? And 4x4 came out...?

A bit of an overreaction, I'd say, especially since other people had already been discussing the 4x4s anyway.

My point about E4300 wasn't simply to echo your earlier point (which I had read, because I make a point of reading the entire thread), but to raise the further question as to whether the new Intel chips would change this very much (e.g. the E6320).

Once, again, calm down...

 

[halbhh]

I'm so calm I'm yawning...but on this forum even 3 months ago is actually very old news, see?

 

[jamiepotter]

Good, but you seem to neglect the point about context. Other people had asked the OP to update his data to reflect the extra mobo costs involved with regards the 4x4s. Other people had reflected that there might be costs involved in buying Intel C2D boards as opposed to AM2 boards, so there was a worry about the consistency of the overall heuristic.

My comment was supposed to address this context. Admittedly, the 'nail in the coffin' remark does make it sound like I'm just slagging off 4x4s in their own right, but my point was really to say that, if anything, the OP was being too kind on them.

I hope that clears things up.

 

[halbhh]

small point, those "other people" were me.

btw, I'm only objecting to your trashing the 4x4 for (valid) old reasons, without addressing the new considerations I wrote plainly in the thread.

 

[halbhh]

This is still very relevant, since plenty of people in the forum apparently haven't read it.

Let us know when you do updates!

 

[HotFoot]

I've been away and haven't added any new CPUs or benchmarks to the mix. Look for a pricing update by Monday morning, but I will make a post when I do the update so you'll get a notification if you're watching the thread, and it'll keep alive on the forumz list.

 

[halbhh]

Does anyone know if Tom's Hardware has ever published another price/performance chart as they used to do weekly? It was often hard to find towards the end, so I looked some, but haven't found one since the one where the author said AMD should be used for many builds. I wonder if he crossed a line???

 

[HotFoot]

I haven't come across any updates to the THG price/performance graphs. They may have lost interest or they may be waiting for something new, like inclusion of the E6300 chip.

The OP has been updated with new graphs for today. The update includes new pricing, and the Gaming index now includes F.E.A.R. I can't remember if this was included in the last update, but in any case I'm using it now. Prices seem to have settled, with either no movement or small reductions across the board. The two chips that moved up violently last week are back down (5600+ and 6000+). The blip last week may have been a channel supply issue.

The current update hasn't yet included a new table. I'm working away from home on a little laptop screen and the table is larger than my screen at the moment, which makes screen capture more difficult. I'll fix that up once I'm back at home with a real monitor.

Considering the lack of substantial changes in pricing over the last two weeks, I am wondering if a weekly update is really useful. It may be better to reserve updates for when Intel/AMD change their tray prices or release new products. Looking at the pricing development between these times is really an exercise of monitoring channel stability.

On the other hand, weekly updates keep the thread current, and effectively advertise it's existence. I would hope we get a couple new viewers every day, and that newcomers will post their comments and suggestions.

 

[halbhh]

Have you noticed the odd discrepancy where the x2 6000 gaming mark fails to outperform the x2 5600? The 6000 has a higher frequency and similar cache.

I first noticed this discrepancy on the Tom's CPU charts last week. While it might indeed be true, my first thought is of course to wonder if there was some error or wrong reading transcription somewhere by some reviewer. If not it's certainly something I'm curious to understand.

Edit: the two processors have the same Windsor 90nm core and cache, with the 6000 at the higher frequency. So it's theorectically impossible for it to underperform the 5600, as Tom's CPU charts show on the Serious Sam bench.

The only thing I notice that might affect it is for some reason Tom's put the 6000 memory speed down to 750, while leaving the 5600 memory at the normal 800!

It's more than a little odd, if you ask me.

end edit.

I haven't recommended the 6000 to anyone, because it's price needs to be close to the e6600 in order to be competitive, IMO.

I would consider it competitive if within $20, because of the interesting and tantalizing upgrade possibilities for AM2 boards.

 

[halbhh]

I haven't come across any updates to the THG price/performance graphs. They may have lost interest or they may be waiting for something new, like inclusion of the E6300 chip.

The OP has been updated with new graphs for today. The update includes new pricing, and the Gaming index now includes F.E.A.R. I can't remember if this was included in the last update, but in any case I'm using it now. Prices seem to have settled, with either no movement or small reductions across the board. The two chips that moved up violently last week are back down (5600+ and 6000+). The blip last week may have been a channel supply issue.

The current update hasn't yet included a new table. I'm working away from home on a little laptop screen and the table is larger than my screen at the moment, which makes screen capture more difficult. I'll fix that up once I'm back at home with a real monitor.

Considering the lack of substantial changes in pricing over the last two weeks, I am wondering if a weekly update is really useful. It may be better to reserve updates for when Intel/AMD change their tray prices or release new products. Looking at the pricing development between these times is really an exercise of monitoring channel stability.

On the other hand, weekly updates keep the thread current, and effectively advertise it's existence. I would hope we get a couple new viewers every day, and that newcomers will post their comments and suggestions.

btw, prices are wrong (in the table at least): the x2 6000 is $439 at newegg, not the 5600.

the 5600 is $329.

This will be especially significant for the x2 5600.

finally, a smaller correction (upward) on the X2 5200, which is listed at $229.