I think what would be neat is to compare several generations of processors in order to see how things have changed and reveal where things are stagnating.
Imagine plotting a PentiumIII vs. Core 2 Quad vs. i7-E vs. Xeon, etc... all performing some fundamental test and scaling the graphs appropriately to give a sense of how far we've come and to extrapolate where we'll most likely be by lets's say 2025 if things continue along the same course. Would also be interesting to reveal where the biggest bottleneck is in terms of future computing. Is it the power consumed? Motherboard/Interconnect technology? Data Storage? Graphics not communicating effectively back to CPU? Programming paradigms, etc... What is holding us back from making an exponential leap? Greater integration? Harnessing of molecules? Nano-photonics? What's the most promising future direction...
Tomshardware also did a 25 chip test for the last decade of computing... though I don't think the anadtech storage locker is big enough to house 25 full desktop builds with dives, memory and all. It's pretty normal graph though, basically GFLOP+GPU GFLOP= performance, Windows is actually pretty good at masking the memory and transfer issues when dealing with games.
Thank you, I'll look into that. Just curious because I've had a PentiumIII 450 that seemed pretty potent and now it's all these years later and it's like what have we really gained besides a few more seconds. I think the big story here is frequency.. Anand really needs to report on why frequencies have stagnated and what we can do to get them climbing again. We really need to break 8GHz within the next five years or something has gone terribly wrong. We've been treated to story after story for the past fifteen years about breakthrough after breakthrough and yet it never seems to trickle down into tangible product. Billions of transistors are nice but frequency across the chip is ultimately more important imo.
Can you specify if the system cost excludes a discrete GPU for chips with integrated GPU and how much is the discrete GPU when included? It doesn't make much sense how the ranking shifts when you look at system costs, if you factored in a flat system cost.. The ranking should not change , just the differences should be muted so can you explain what you did or verify that you didn't messed up the math? As it is right now it seems just wrong when compared to just the CPU ranking. Anyway , using a flat system cost is also not at all a good way to look at things, total costs tend to be proportional with the CPU cost. We do aim for a balanced system in w/e budget so comparing just the chips is closer to reality than comparing with a flat cost for the system. It also seems that the GPU perf per $ AMD offers over Intel is much higher than the CPU perf per $ Intel offers over AMD. Maybe AMD should shift a bit of die area (or thermal budget) from GPU to CPU although that is unlikely before the new cores. When i started reading kinda hoped i'll see some die size per price and die size per perf numbers too, maybe next time.
So at this point i assume that in perf per system costs you went for a flat system cost without even adding the cost of the CPU. So take a 300$ CPU but factor in 350$ system cost without adding the 300$ for the CPU and that made that ranking defy logic.
yeah that's what you did, note how similar the ranking for pure CPU perf without price and the one for CPU perf per system price are. The LGA 2011 shifts but all else stays so you just don't factor in the CPU cost at all.( again even factoring that in is not a good way to do it and just CPU/price is a better) Sorry for the triple posting but non edit.
Actually, the relative ranking can shift quite a bit with the entire system cost. It's easy enough to see why:
Say CPU 1 has performance of 1.0 and costs $100 with a base system cost of $400. That gives a CPU performance/price ratio of 1 (scaling by 100) and a system performance/price ratio of 0.25. Say CPU 2 has twice the performance and twice the cost, with a total system price of $500. The CPU performance/price is still 1, but the system performance/price is now 0.4.
Basically looking at the total system cost, unless the total scales linearly with price, faster CPUs will be relatively better vs. just looking at the cost of the CPU alone.
I am sorry, not sure if this is intentional or you really don't get it. if it is intentional that it shows how deeply corrupted Anadtech is. You are comparing CPU perf to the cost of bananas in Bnagladesh. You take a 320$ CPU of perf X and do perf per 350$ then you take a 100$ CPU of perf Y and do perf per 350$. You don't include the cost of the CPU at all, you just do perf per cost off all the other parts excluding CPU and that cost is rather random. This is why the only change from perf reanking that does not include price is a penalty for LGA2011. The result has NOTHING at all to do with cost I mean,you take a "system cost "of 450$ for a machine where just the CPU is 1k$ instead of factoring in 1450$. I really hope you correct and republish this because otherwise ,it's just too obvious
The only things obvious here are the following: 1) You're incapable of cost-benefit analysis if you can't even figure something as simple as the one in this article. 2) You're not clear about what you're trying to get across. What is your point? You just scaled the price of the other components based on the price of the CPU. It's entirely possible to have either a $300 i7-4790 or $60 Pentium G3220 while the rest of the components stay in the $350 range. He's isolating the CPU's performance. 3) You can't even be bothered to proofread. You come in here talking like someone from primary/elementary school.
Yes it is true that i can't be bothered to proof read a comment here when i am way busy at 6 am and after all English is my 3rd language so maybe ii even made some mistakes not just typos but at least my logic is above 3rd grade. You must be very pleased that you got to say cost benefit analysis but if you think this way is reasonable then you clearly need a dictionary.or go Google for it- if we are being rude and all.
With that said he might not have made the mistake i assumed he did but since the article doesn't provide clear numbers and details it seems way off. However the chart is still way off compared to reality and pretty sure it doesn't include a discrete GPU where there is no on die GPU. It s possible but very unlikely for a 100$ CPU to be paired with 16-32GB of RAM and a 700-1000W PSU and so on, while it is is very likely that a 1k CPU will have a lot more RAM and bigger PSU. A fixed cost for the other parts makes only sense if you try to distort the results. Any reasonable system guide provides a balanced system at a certain budget. If Anandtech was Intel's marketing team it would be ok , we are used with such tricks but it would be nice to a different behavior.
The performance relative to system cost was scaled as I stated in the article: the base cost of the system was estimated at $350 for all of the systems except for LGA2011, which was estimated at $450 (thanks to the more expensive motherboards). Add to that cost the price of the CPU, so the Extreme CPUs end up priced around $800-$1500 for the complete system while the budget CPUs end up priced around $400-$500 for the complete system.
I'm sorry if that wasn't entirely clear in the article, but this was intended as more of an introduction and overview than an in-depth look at every facet of CPUs. It's for people that say, "Hey, if I'm looking at a new system, what's the best bang for the buck and roughly how much faster is chip X compared to chip Y?" We already have plenty of lengthier content, so this sort of article is designed to cater to a wider audience.
As far as the inclusion of discrete GPUs, the data comes from CPU Bench so everything obviously has to have some sort of GPU. In the case of CPUs that don't have on-die graphics, I believe Ian uses an entry level GPU like an HD 7450 similar. Most of the CPU-centric tests don't hit the GPU much, so as long as we're not throwing in a high-end GPU the overall performance should be consistent. We do not have any performance figures for discrete GPUs in our graphics section of test -- at least not that I used (there are SLI and CF numbers, but not with all systems and that would be more of a look at GPU performance than CPUs).
Finally, regarding what sort of system a person would build, of course most people buying a high-end CPU aren't going to use low cost parts elsewhere, but if we then scale the supposed cost of a system by what parts it might use -- "No one would use a 128GB SSD with an i7-4960X; let's equip that system with two 512GB SSDs in RAID 0!" -- we run into a completely different problem. We have to use a baseline cost somewhere if we are to estimate performance/price for a CPU, and I'd argue strongly that using the minimum reasonable cost is a far better choice than going with a higher spec system. If we did the latter, who would actually use a 512GB SSD and a GTX Titan Z with an A6 APU or Pentium CPU? The reverse (using a high-end CPU with a moderate SSD) is far more likely -- there are compute workstations that I've used that do exactly that.
FWIW, Jarred, I thought your value analysis was clear and completely agree that you did things in a reasonable way (not to say that there aren't other reasonable ways, but you can only do so much in a short form article). Obviously people have to make their own personal decisions based on the systems they're looking at, but even as someone who has been building his own computers for almost 20 years now, I still found this a good read and will definitely point newer builders here for easy reference. Thanks!
I agree. It seemed obvious why the minimum total build price for each cpu was considered, and why it would result in the dramatic reordering of relative value - I don't think Jarred really needed to spell it out, but maybe something got lost in translation for jjj.
I think Symphony is right, your logic is flawed. The very idea that anyone could be comparing the chips means that you can isolate other system costs. If they were to prepare a chart with totally different system configurations for each system it would be totally useless.
Anandtech is sponsored by AMD so I don't understand why you think they're favouring Intel. The truth here is that AMDs chips are not competitive in CPU performance
Actually, Anandtech is sponsored mostly by Third World Press, and that's why they use so much descriptive poetry in their reviews. It also explains the heavy bias against E-Readers, since Third World Press wants to covertly steer electronic customers toward their papial regime rather than the e-readial consumption mechanism.
yeesssss YEESSSS.... Jarred is CORRUPT.. and DRUNK with POWWWWERR! The next stop? THE WORLD!!! Mmmmuuuuahahahahaha!!! And you caught him jjj, the readers are forever in your debt!
I thought it was a great article and very informative. I really liked the charts that showed direct comparisons of system performance relative to the i3 (cpu and graphics) I mean, we already knew that Intel had a cpu advantage and AMD had the graphical advantage, but it is really nice to see just how much of a difference there is between multiple processors (and even intel to intel/amd to amd comparison) all in one chart. Thank you for the write up, and please let me know if you're in need of a side kick on your quest for world domination, as every super villain needs one.
Anandtech is corrupt ,the question is how much. It's not all that obvious but it's there. Just accepting to publish sponsored posts is corruption and compromises it's integrity (just because everybody is doing it doesn't mean it's acceptable). Today we have seen benchmarks for a Snapdragon 805 device and ofc the devices it was compared to were hand picked to exclude some key competitors. Often Anandtech looks at a certain technology but only talks with one provider of that technology about it, ignoring the competition and basically just advertising that one provider.. Sometimes certain benchmarks are ignored because that particular product doesn't do well there. But mostly it's about what the product reviewed is compared against and small tricks to make it look good. And when was the last time a review here didn't had a conclusion that was overall positive and when was the last time certain companies had a negative conclusion to a review? How about Asus, any Asus product gets it's own news item and there are lots of Asus reviews while lots of lots of products that are far more relevant don't even get mentioned.Intel is similar and when AMD started paying for it's own section it started getting far more extensive coverage. I doubt it's about money , it's most likely more about getting access but that doesn't make it ok. So yes , without any doubt Anandtech is not objective and it's been going downhill for some years now, why it's less and less reliable , that's hard to say for sure.
I think you need to draw us a flowchart for all the payola Anand is taking, at this point I can't figure what poor beleaguered company you're trying to defend from Anandtech's sheer corruption but I'm sure there's one... Clearly it'll make all the sense in the world once you spell it out for the rest of the blinded readership.
There is actually truth in what jjj says here. It is "the Way of the World" these days. To call it corruption is too harsh though.
I mean, a whole new section on "Wearables", with giveaways etc.? This is just marketing push, pure and simple.
Unfortunately, it is a game that realistically has to be played, because I don't see the readership reaching into their pockets to fund independent journalism.
As for this article, in my experience, with every component in computing, you tend to get say 75-90% of the performance starting at around 30-50% of the cost of the "ultimate" must-have part. When putting a system together, that can quickly add up to a lot of money.
Obviously, when considering each individual component, if you hold the rest of the system price constant, then the marginal cost of upgrading that single part doesn't seem very significant.
More often than not, it is best to focus on what you want to do with the system, and consider the trade-offs in terms of "what-I-want/need to do" / price. Often it comes down to features and market segmentation rather than performance, e.g. AES-NI is a great CPU feature that really helps if you want to do software encryption on your storage. However, when you realise Intel wants to charge you $50 for that, and the federal government will just waterboard you if they want your password, or take you out in a drone strike, you figure it may not be worth that much to you. VT-D and ECC RAM are other features that cost little to produce, but cost a fortune in the marketplace (from Intel), mostly as increased platform cost. Would I like ECC RAM? Sure. Do I want to pay for it?
Sadly, my data is mostly of little value, even those supposedly priceless "family photos".
In addition to total system cost vs just the price of the processor, how does AMD stack up vs Intel in terms of long-term cost of ownership? The AMD parts look more power-hungry. I suppose most people's systems spend most of their time sitting around doing not-very-much-at-all, so it's probably based on local power costs and idle power usage, but power consumption over however many years you have the chip could be interesting.
The question to ask now for long term cost with AMD is: will they even be around in 2 years? Intel is a pretty safe bet, but when it comes time for updated drivers for Windows 9...will team Red still be supporting...anything?
that question has been asked already for the past 10 years... so go figure.
Is Intel a safe bet? they are so marketing dominant that they can decide wathever they want what to support and not, they decide what OEM are allowed to bring to market and what not. Consumers are the one to blame here and are the ones to suffer. THis article once again shows the vlaue proposition which should apply for a huge amount of buyers. But many are fooled by the sales guys in the mega shopping markets or the djingle and brand name :)
On the drivers side, look at the GPU drivers from Intel and AMD and you know which one has lots of work to be done...
Outside of the AM3+ and LGA2011 platforms, unless you're running CPU intensive workloads for a significant amount of time, the power costs won't be all that different. But AM3+ will generally use the most power at idle (on the quad-module parts). Cost of repairs due to unreliable parts will mostly come from failed motherboards and other components in my experience, and I couldn't say whether AMD is any better or worse than Intel in that area.
The dollar symbol in the first graph makes it very confusing. Your graphs would also be easier to interpret if you included the value which corresponded to each product such as 1.00 for the i3-4330.
Any chance we could get a couple of overclocked processers on the charts too to see where they really fit (I assume most people who read this are going to be enthusiasts who will have k processors overclocked somewhat. Maybe an i5-2500k, i7-3770 and i7-4790?
We'd need to have updated figures from the relevant parts, and unfortunately we don't have that for many of the older CPUs. Basically, I think Devil's Canyon and the Pentium AE are the only CPUs with a good set of performance data.
CPUs were eliminated based on those that didn't have enough recent benchmark results. The 8320 and 8350 are very similar in performance, but the 8320 had quite a few more results so I used it and left off the 8350. I did include a few CPUs that don't have very many results, mostly because they're important parts -- the i5-2500K and i7-3770K for example only have results in about half of the benchmarks IIRC. Ideally we'd have all results for all CPUs, but that would require a lot of time just to get Bench updated with those numbers.
Can you also show the Performance/Watt bars (they could be derived by the Performance bars since for each one you indicate the Watt peak). Even more interesting would be (but requires a actual power measurement) to show the Performance/Energy bars where the Energy should be the exact energy that is spent to get the given performance bar ! Thanks for the article !
Unfortunately we don't have clear power use numbers for the various systems, particularly for the performance workloads being tested. CPU Bench has Idle and Load for some of the chips, but what we'd really want is a viable "average power" or "power used across all benchmarks" and we don't have that.
Can you please include results for the lower-cost Core i5s, such as the 4460? I would like to know if they offer an even better price/performance ratio than the 4690 or 4590.
Unfortunately I'm limited to what we have tested here, and most of the i5 and i3 parts are missing -- or for those we do have, they only have a small selection of results. I'll most likely provide an update about once a quarter for this information, or perhaps a few weeks after any major CPU/APU launch.
What this shows is that AMD really needs to step up their game. They need to take their current Piledriver design, take out the "second processor" half, (meaning, the second set of integer pipelines and the logic that makes them look like two separate cores) and put out 4 to 8 of those cores per processor. They might catch up to Intel then, at least partially. This idea of sharing the FP unit and decoders just didn't work.
It could work if they have invested in compiler/driver (and generally software), but I think right now they are in panic mode, and we cannot expect anything serious coming out of them until 2016 (plus delays).
Well, at least, it seems that AMD realised that their Bulldozer concept with shared FPU did not play out well. Because their last Bulldozer-derived pure CPU (with no GPU) was (is) Piledriver-based FX (same for server Opterons), so that only APUs go beyond Piledriver (as current Steamroller-based Kaveri).
More than that, there are *rumours*, that the last generation of construction microarch, Excavator, won't even make it to the desktop APUs - Excavator-based Carrizzo supposedly will be released only for laptops (and dense servers).
Another *rumours* are out there that beyond Excavator AMD will release some new non-modular CPU microarch, presumably in 2016, so that the whole lineup of microarchs with shared FPUs will be on the market only from 2011 (Bulldozer) till 2015 (Excavator).
But take these *rumours* with the grain of salt - this is just what is written on the Web, but the validity of this info unconfirmed.
Could we have a similar article for laptop chips in the near future? The splits between Haswell/Baytrail and Kaveri/Beema should make the CPU table a bit more interesting; and you've got a wider spread on Intel IGP performance as well. I'm not sure if the spread on amd's mobile IGPs is wider than for desktops or not.
I'd love to be able to provide that, but unfortunately we don't have quite the same level of data -- and we're missing many of the low-end CPUs/APUs. Another issue is that laptops are reviewed as entire systems, so the presence or lack of an SSD can really skew certain results, not to mention the amount of RAM, inclusion of a dGPU (and which GPU is used), etc.
It would be good to have a power/battery life graph as well, but again: the OEM of the laptop can really mess things up -- I have literally seen two laptops with the same CPU and chipset where one is nearly twice as power efficient as the other, simply because of poor optimizations and/or component choices by one OEM (and good choices by the other OEM).
Anyway, it's definitely something I want to look at, but I'm not sure yet how useful we can make it given the current set of information in Mobile Bench. Tablets and smartphones are another area we want to look at with this sort of article, and they have some of the same problems as laptops, but I think they tend to be a little more consistent on the battery life at least.
Would it ever be possible to add a new section to the mobile bench to allow people to search for specific CPUs/GPUs? The data might not be 100% accurate comparison between best case scenarios between parts, due to thermal constraints, but it would still be helpful. Even having certain parts performing lower than you'd expect to see is useful because you know that thermals play a significant role for that part. I know there have been many, many times when I wish I could sort through all the mobile bench information by CPU or GPU to get roughly appropriate comparisons.
I like the system cost price ratio. Adding $100 to the LGA2011 made the price per performance is realistic. I did a quick check cheapest LGA2011 motherboard is $200, Z97 is $90. And if you spend $30 for memory on Z97, then it becomes $60 for LGA2011. That totals $140 price difference, $100 is probably close enough because you probably could find lower numbers than I did.
For memory, you don't have to populate all 4 channels on an LGA2011; and unless you go with a 6/8 core model it's not going to matter outside of benchmarks. Lastly if you're trying to match specs as much as possible while populating the extra slots 4 half capacity dimms are a lot closer in price than a 2x multiplier ($70 vs $85 for 8gb)
Excellent piece. I've been looking into the A8-7600 for a family HTPC that could do a little bit more, at 45W in a passively cooled case. Being in Europe this particular APU is a little bit overpriced, even compared to the A10-7850k, due to a different pricing structure relative to the US, so the graphic performance/price numbers don't look quite as good. Having waited 8 months for its release, I can still wait a few more until the pricing is more favourable.
Jarred, your unit in the first system cost table is wrong. It should not contain the $ as unit but be unitless as it is either normalised or in (benchmark points) PER dollar.
Jarred, Core i5-4670K average CPU performance on the first chart is too low; it's inconsistent with Core i5-4690(K) which is only 100 MHz faster but way faster on your chart. I guess it has to do something with the fact that they were tested 1 year apart in time (different SW versions?)
Realistically, on your chart, if Core i5-4690(K) slightly outperform Core i7-2600K on average, then Core i7-4670K should almost reach Core i7-2600K - if this will be the case, then everything will be consistent, IMHO.
The problem is that the i5-4670K was not tested on as many benchmarks as the i5-4690K, and apparently the tests where it wasn't run tend to be more favorable. To be specific, there are 11 of the 29 CPUs/APUs that were only tested on about half of the benchmarks.
If you think Anandtech is corrupt and going downhill, stop reading it. I've found Anandtech to have an occasional bias. I think you could say the same for any review/reporting agency, profit or non-profit. If you perceive a bias it isn't exactly difficult to take it into account when reading the article.
I sure wish there was a way to block seeing annoying posters in the comments.
Jared, nicely done/ I've been hoping for something like this for a while. :)
One more thing I think you can usefully add: If you factor in running costs for the various CPUs over a 1-year period, what does it work out to then in therm of value for performance?
It's really frustrating that when you talk about price you don't include the added cost of running AMD vs Intel. They cost more to keep on, they draw more power. No, it's not a LOT of money, but $5-$15 every year adds up.
The computer industry being how it is today it's totally realistic to run those numbers out 10 years. At which point the small dollar amounts aren't so small anymore.
I'd really like to see cost of electricity, based US national average (which I believe is about 10c/KWH right now), worked into these "performance/dollar" calculations. It colors things even further in Intel's favor.
PS, looking at the architecture maps. Why does AMD's look like a toddler started breaking blocks to fit them together? Intel's is all symmetrical and clean. AMD's is sloppy and inefficient.
Probably because a toddler (me) did the blocks on the AMD die shot. Hahaha... Thanks. I'm going to go cry in a corner now. ;-)
As for the cost of running the systems over the course of years, well, that all depends on what sort of workloads you're running. With the systems powered on (and not really doing much other than web surfing/office work), AMD's APUs generally consume less power than Intel's CPUs. The FX-series CPUs are a different matter of course, and the same goes for LGA2011 Intel systems
While we don't have idle numbers for all of the systems (so I couldn't really provide data on this aspect of the CPUs/APUs), you can at least get a reasonable idea of power use from Legit Reviews' latest article on the 65W Kaveri parts: http://www.legitreviews.com/amd-kaveri-a8-7600-a10...
Assuming 12 hours of day of moderate use (so let's say 80% idle, 20% load), using their numbers you'd spend about $27 per year running the i7-4790K compared to $14-$17 running the Kaveri platforms. I don't think most people are really going to care about spending $1 extra per month for a faster PC, and of course anyone using a discrete GPU will increase the power cost of having the system on by another $0.50-$1.00 even if they never use the dGPU for gaming.
TL;DR: Power costs really only matter if you're running the systems at load most of the time (e.g. servers), and if that's the case performance often matters even more. That's my opinion at least. :-)
This is an excellent article. Thank you so much for writing it, Jarred. Could you please add a couple of scatter plots with price and performance axes? Most products usually fit close to one of two asymptotes and the sweet spot is at their intersection point.
I talked about this with Anand; basically, I find scatter plots to be more advanced and difficult to understand, and as this was targeting the "easy to grok" crowd I opted for simple bar charts. Still, if you want a scatter plot, here's the performance vs. system cost plot: http://images.anandtech.com/doci/8312/ScatterPlot-...
What's amazing to me is how little AMD has changed on the CPU performance curve for the past several generations of APUs. Granted, the graphics plot would look better, but still... Of course, there are plenty of use cases where having more than 2 CPU cores isn't all that important.
Jarred, true on the CPU part of AMD, but if you look at the power draw you can actually see that they did make a lot of improvements on power/perf while not able to have the most high tech fabs and process...
And statement again, how much CPU is really needed for day time tasks? have been reading for approx 1 hour some articles on anandtech, did i really needed my fancy i7 for that?
Surely though you have power saving features, etc., turned on, so while doing simple tasks your i7 is hanging back, not drawing power it doesn't need? In that state, an i7 isn't much different from the other CPUs. Efficient oc'ing then depends on using dynamic vcore, which is a bit more difficult, but well worth the learning curve (good piece about it on the ROG site).
Jarred, consider yourself awared numerous medals just for using the word, 'grok'. :D
I'm confused about why you would ONLY want to include the cost of discrete GPU for the FX Cpus. Some AM3 boards do come with it on board. You could've had both.
I could see why you would want to avoid comparing CPUs after overclocking to ensure this suitable for all audiences but I think a concession should've been made here if you are going to compare what's left of the FX-series landscape.
I included a $50 dGPU in the cost of both the AM3+ and LGA2011 platforms. That's mostly because I figure anyone buying an AM3+ or LGA2011 rig are going to use a discrete GPU. The AM3+ boards that have integrated graphics are basically very old and slow IGPs at this stage in time. And really, I don't expect most people to buy either platform these days, as they've both been around for some time and neither one is due for any additional updates that I know of.
My approach to the big picture would be to find a fair point of comparison. AMD APUs offer the (single thread) performance of Nehalem CPUs (the one before Sandy Bridge), so even though AMD builds are viable they are about 4 to 5 years less "future proof", but they have well balanced power efficiency, come with better graphics and a significantly smaller markup. Desktop Kaveri APUs basically match the CPU performance of a mainstream MOBILE Haswell CPU clocked at around 2 GHz.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
69 Comments
Back to Article
jeffrey - Tuesday, August 5, 2014 - link
Nice article Jarred! Reminds me of the Anandtech days of old.., very well done in concept and execution.RU482 - Tuesday, August 5, 2014 - link
I was just thinking the same thing. I like and appreciate occasional articles that can cater to the beginners and the old hats in the room!TETRONG - Wednesday, August 6, 2014 - link
I think what would be neat is to compare several generations of processors in order to see how things have changed and reveal where things are stagnating.Imagine plotting a PentiumIII vs. Core 2 Quad vs. i7-E vs. Xeon, etc... all performing some fundamental test and scaling the graphs appropriately to give a sense of how far we've come and to extrapolate where we'll most likely be by lets's say 2025 if things continue along the same course.
Would also be interesting to reveal where the biggest bottleneck is in terms of future computing. Is it the power consumed? Motherboard/Interconnect technology? Data Storage? Graphics not communicating effectively back to CPU? Programming paradigms, etc... What is holding us back from making an exponential leap?
Greater integration? Harnessing of molecules? Nano-photonics?
What's the most promising future direction...
mkozakewich - Wednesday, August 6, 2014 - link
They did something like that last year with the Mac Pro line.basroil - Wednesday, August 6, 2014 - link
Tomshardware also did a 25 chip test for the last decade of computing... though I don't think the anadtech storage locker is big enough to house 25 full desktop builds with dives, memory and all. It's pretty normal graph though, basically GFLOP+GPU GFLOP= performance, Windows is actually pretty good at masking the memory and transfer issues when dealing with games.TETRONG - Thursday, August 7, 2014 - link
Thank you, I'll look into that.Just curious because I've had a PentiumIII 450 that seemed pretty potent and now it's all these years later and it's like what have we really gained besides a few more seconds.
I think the big story here is frequency.. Anand really needs to report on why frequencies have stagnated and what we can do to get them climbing again. We really need to break 8GHz within the next five years or something has gone terribly wrong.
We've been treated to story after story for the past fifteen years about breakthrough after breakthrough and yet it never seems to trickle down into tangible product.
Billions of transistors are nice but frequency across the chip is ultimately more important imo.
jjj - Tuesday, August 5, 2014 - link
Can you specify if the system cost excludes a discrete GPU for chips with integrated GPU and how much is the discrete GPU when included?It doesn't make much sense how the ranking shifts when you look at system costs, if you factored in a flat system cost.. The ranking should not change , just the differences should be muted so can you explain what you did or verify that you didn't messed up the math? As it is right now it seems just wrong when compared to just the CPU ranking.
Anyway , using a flat system cost is also not at all a good way to look at things, total costs tend to be proportional with the CPU cost. We do aim for a balanced system in w/e budget so comparing just the chips is closer to reality than comparing with a flat cost for the system.
It also seems that the GPU perf per $ AMD offers over Intel is much higher than the CPU perf per $ Intel offers over AMD. Maybe AMD should shift a bit of die area (or thermal budget) from GPU to CPU although that is unlikely before the new cores.
When i started reading kinda hoped i'll see some die size per price and die size per perf numbers too, maybe next time.
jjj - Tuesday, August 5, 2014 - link
So at this point i assume that in perf per system costs you went for a flat system cost without even adding the cost of the CPU.So take a 300$ CPU but factor in 350$ system cost without adding the 300$ for the CPU and that made that ranking defy logic.
jjj - Tuesday, August 5, 2014 - link
yeah that's what you did, note how similar the ranking for pure CPU perf without price and the one for CPU perf per system price are. The LGA 2011 shifts but all else stays so you just don't factor in the CPU cost at all.( again even factoring that in is not a good way to do it and just CPU/price is a better)Sorry for the triple posting but non edit.
JarredWalton - Tuesday, August 5, 2014 - link
Actually, the relative ranking can shift quite a bit with the entire system cost. It's easy enough to see why:Say CPU 1 has performance of 1.0 and costs $100 with a base system cost of $400. That gives a CPU performance/price ratio of 1 (scaling by 100) and a system performance/price ratio of 0.25. Say CPU 2 has twice the performance and twice the cost, with a total system price of $500. The CPU performance/price is still 1, but the system performance/price is now 0.4.
Basically looking at the total system cost, unless the total scales linearly with price, faster CPUs will be relatively better vs. just looking at the cost of the CPU alone.
jjj - Tuesday, August 5, 2014 - link
I am sorry, not sure if this is intentional or you really don't get it. if it is intentional that it shows how deeply corrupted Anadtech is.You are comparing CPU perf to the cost of bananas in Bnagladesh.
You take a 320$ CPU of perf X and do perf per 350$ then you take a 100$ CPU of perf Y and do perf per 350$. You don't include the cost of the CPU at all, you just do perf per cost off all the other parts excluding CPU and that cost is rather random. This is why the only change from perf reanking that does not include price is a penalty for LGA2011.
The result has NOTHING at all to do with cost
I mean,you take a "system cost "of 450$ for a machine where just the CPU is 1k$ instead of factoring in 1450$.
I really hope you correct and republish this because otherwise ,it's just too obvious
SymphonyX7 - Wednesday, August 6, 2014 - link
The only things obvious here are the following:1) You're incapable of cost-benefit analysis if you can't even figure something as simple as the one in this article.
2) You're not clear about what you're trying to get across. What is your point? You just scaled the price of the other components based on the price of the CPU. It's entirely possible to have either a $300 i7-4790 or $60 Pentium G3220 while the rest of the components stay in the $350 range. He's isolating the CPU's performance.
3) You can't even be bothered to proofread. You come in here talking like someone from primary/elementary school.
jjj - Wednesday, August 6, 2014 - link
Yes it is true that i can't be bothered to proof read a comment here when i am way busy at 6 am and after all English is my 3rd language so maybe ii even made some mistakes not just typos but at least my logic is above 3rd grade.You must be very pleased that you got to say cost benefit analysis but if you think this way is reasonable then you clearly need a dictionary.or go Google for it- if we are being rude and all.
With that said he might not have made the mistake i assumed he did but since the article doesn't provide clear numbers and details it seems way off. However the chart is still way off compared to reality and pretty sure it doesn't include a discrete GPU where there is no on die GPU.
It s possible but very unlikely for a 100$ CPU to be paired with 16-32GB of RAM and a 700-1000W PSU and so on, while it is is very likely that a 1k CPU will have a lot more RAM and bigger PSU. A fixed cost for the other parts makes only sense if you try to distort the results. Any reasonable system guide provides a balanced system at a certain budget. If Anandtech was Intel's marketing team it would be ok , we are used with such tricks but it would be nice to a different behavior.
JarredWalton - Wednesday, August 6, 2014 - link
The performance relative to system cost was scaled as I stated in the article: the base cost of the system was estimated at $350 for all of the systems except for LGA2011, which was estimated at $450 (thanks to the more expensive motherboards). Add to that cost the price of the CPU, so the Extreme CPUs end up priced around $800-$1500 for the complete system while the budget CPUs end up priced around $400-$500 for the complete system.I'm sorry if that wasn't entirely clear in the article, but this was intended as more of an introduction and overview than an in-depth look at every facet of CPUs. It's for people that say, "Hey, if I'm looking at a new system, what's the best bang for the buck and roughly how much faster is chip X compared to chip Y?" We already have plenty of lengthier content, so this sort of article is designed to cater to a wider audience.
As far as the inclusion of discrete GPUs, the data comes from CPU Bench so everything obviously has to have some sort of GPU. In the case of CPUs that don't have on-die graphics, I believe Ian uses an entry level GPU like an HD 7450 similar. Most of the CPU-centric tests don't hit the GPU much, so as long as we're not throwing in a high-end GPU the overall performance should be consistent. We do not have any performance figures for discrete GPUs in our graphics section of test -- at least not that I used (there are SLI and CF numbers, but not with all systems and that would be more of a look at GPU performance than CPUs).
Finally, regarding what sort of system a person would build, of course most people buying a high-end CPU aren't going to use low cost parts elsewhere, but if we then scale the supposed cost of a system by what parts it might use -- "No one would use a 128GB SSD with an i7-4960X; let's equip that system with two 512GB SSDs in RAID 0!" -- we run into a completely different problem. We have to use a baseline cost somewhere if we are to estimate performance/price for a CPU, and I'd argue strongly that using the minimum reasonable cost is a far better choice than going with a higher spec system. If we did the latter, who would actually use a 512GB SSD and a GTX Titan Z with an A6 APU or Pentium CPU? The reverse (using a high-end CPU with a moderate SSD) is far more likely -- there are compute workstations that I've used that do exactly that.
Hope that helps clarify things.
teiglin - Wednesday, August 6, 2014 - link
FWIW, Jarred, I thought your value analysis was clear and completely agree that you did things in a reasonable way (not to say that there aren't other reasonable ways, but you can only do so much in a short form article). Obviously people have to make their own personal decisions based on the systems they're looking at, but even as someone who has been building his own computers for almost 20 years now, I still found this a good read and will definitely point newer builders here for easy reference. Thanks!dananski - Wednesday, August 6, 2014 - link
I agree. It seemed obvious why the minimum total build price for each cpu was considered, and why it would result in the dramatic reordering of relative value - I don't think Jarred really needed to spell it out, but maybe something got lost in translation for jjj.Cool article.
Flunk - Wednesday, August 6, 2014 - link
I think Symphony is right, your logic is flawed. The very idea that anyone could be comparing the chips means that you can isolate other system costs. If they were to prepare a chart with totally different system configurations for each system it would be totally useless.Anandtech is sponsored by AMD so I don't understand why you think they're favouring Intel. The truth here is that AMDs chips are not competitive in CPU performance
seapeople - Wednesday, August 6, 2014 - link
Actually, Anandtech is sponsored mostly by Third World Press, and that's why they use so much descriptive poetry in their reviews. It also explains the heavy bias against E-Readers, since Third World Press wants to covertly steer electronic customers toward their papial regime rather than the e-readial consumption mechanism.HanzNFranzen - Wednesday, August 6, 2014 - link
yeesssss YEESSSS.... Jarred is CORRUPT.. and DRUNK with POWWWWERR! The next stop? THE WORLD!!! Mmmmuuuuahahahahaha!!! And you caught him jjj, the readers are forever in your debt!I thought it was a great article and very informative. I really liked the charts that showed direct comparisons of system performance relative to the i3 (cpu and graphics) I mean, we already knew that Intel had a cpu advantage and AMD had the graphical advantage, but it is really nice to see just how much of a difference there is between multiple processors (and even intel to intel/amd to amd comparison) all in one chart. Thank you for the write up, and please let me know if you're in need of a side kick on your quest for world domination, as every super villain needs one.
jjj - Wednesday, August 6, 2014 - link
Anandtech is corrupt ,the question is how much. It's not all that obvious but it's there.Just accepting to publish sponsored posts is corruption and compromises it's integrity (just because everybody is doing it doesn't mean it's acceptable). Today we have seen benchmarks for a Snapdragon 805 device and ofc the devices it was compared to were hand picked to exclude some key competitors.
Often Anandtech looks at a certain technology but only talks with one provider of that technology about it, ignoring the competition and basically just advertising that one provider.. Sometimes certain benchmarks are ignored because that particular product doesn't do well there.
But mostly it's about what the product reviewed is compared against and small tricks to make it look good. And when was the last time a review here didn't had a conclusion that was overall positive and when was the last time certain companies had a negative conclusion to a review?
How about Asus, any Asus product gets it's own news item and there are lots of Asus reviews while lots of lots of products that are far more relevant don't even get mentioned.Intel is similar and when AMD started paying for it's own section it started getting far more extensive coverage.
I doubt it's about money , it's most likely more about getting access but that doesn't make it ok.
So yes , without any doubt Anandtech is not objective and it's been going downhill for some years now, why it's less and less reliable , that's hard to say for sure.
Impulses - Wednesday, August 6, 2014 - link
I think you need to draw us a flowchart for all the payola Anand is taking, at this point I can't figure what poor beleaguered company you're trying to defend from Anandtech's sheer corruption but I'm sure there's one... Clearly it'll make all the sense in the world once you spell it out for the rest of the blinded readership.Klug4Pres - Wednesday, August 6, 2014 - link
There is actually truth in what jjj says here. It is "the Way of the World" these days. To call it corruption is too harsh though.I mean, a whole new section on "Wearables", with giveaways etc.? This is just marketing push, pure and simple.
Unfortunately, it is a game that realistically has to be played, because I don't see the readership reaching into their pockets to fund independent journalism.
As for this article, in my experience, with every component in computing, you tend to get say 75-90% of the performance starting at around 30-50% of the cost of the "ultimate" must-have part. When putting a system together, that can quickly add up to a lot of money.
Obviously, when considering each individual component, if you hold the rest of the system price constant, then the marginal cost of upgrading that single part doesn't seem very significant.
More often than not, it is best to focus on what you want to do with the system, and consider the trade-offs in terms of "what-I-want/need to do" / price. Often it comes down to features and market segmentation rather than performance, e.g. AES-NI is a great CPU feature that really helps if you want to do software encryption on your storage. However, when you realise Intel wants to charge you $50 for that, and the federal government will just waterboard you if they want your password, or take you out in a drone strike, you figure it may not be worth that much to you. VT-D and ECC RAM are other features that cost little to produce, but cost a fortune in the marketplace (from Intel), mostly as increased platform cost. Would I like ECC RAM? Sure. Do I want to pay for it?
Sadly, my data is mostly of little value, even those supposedly priceless "family photos".
jibberegg - Tuesday, August 5, 2014 - link
Great, really accessible summary article! Enjoy the in-depth analysis of course, but the holistic view presented here in neat. Thanks!evilspoons - Tuesday, August 5, 2014 - link
In addition to total system cost vs just the price of the processor, how does AMD stack up vs Intel in terms of long-term cost of ownership? The AMD parts look more power-hungry. I suppose most people's systems spend most of their time sitting around doing not-very-much-at-all, so it's probably based on local power costs and idle power usage, but power consumption over however many years you have the chip could be interesting.takeship - Tuesday, August 5, 2014 - link
The question to ask now for long term cost with AMD is: will they even be around in 2 years? Intel is a pretty safe bet, but when it comes time for updated drivers for Windows 9...will team Red still be supporting...anything?bill5 - Wednesday, August 6, 2014 - link
Err, CPU's dont need drivers.nathanddrews - Wednesday, August 6, 2014 - link
Err, sorta:http://en.wikipedia.org/wiki/Bulldozer_(microarchi...
DarkXale - Sunday, August 10, 2014 - link
The kernel needs to 'understand' quirks of the CPU - but after that there is no maintenance involved.duploxxx - Wednesday, August 6, 2014 - link
that question has been asked already for the past 10 years... so go figure.Is Intel a safe bet? they are so marketing dominant that they can decide wathever they want what to support and not, they decide what OEM are allowed to bring to market and what not. Consumers are the one to blame here and are the ones to suffer. THis article once again shows the vlaue proposition which should apply for a huge amount of buyers. But many are fooled by the sales guys in the mega shopping markets or the djingle and brand name :)
On the drivers side, look at the GPU drivers from Intel and AMD and you know which one has lots of work to be done...
JarredWalton - Tuesday, August 5, 2014 - link
Outside of the AM3+ and LGA2011 platforms, unless you're running CPU intensive workloads for a significant amount of time, the power costs won't be all that different. But AM3+ will generally use the most power at idle (on the quad-module parts). Cost of repairs due to unreliable parts will mostly come from failed motherboards and other components in my experience, and I couldn't say whether AMD is any better or worse than Intel in that area.ridic987 - Wednesday, August 6, 2014 - link
http://images.anandtech.com/doci/8312/CPU-performa...http://images.anandtech.com/doci/8312/iGPU-perform...
The dollar symbol in the first graph makes it very confusing. Your graphs would also be easier to interpret if you included the value which corresponded to each product such as 1.00 for the i3-4330.
Awful - Wednesday, August 6, 2014 - link
Any chance we could get a couple of overclocked processers on the charts too to see where they really fit (I assume most people who read this are going to be enthusiasts who will have k processors overclocked somewhat. Maybe an i5-2500k, i7-3770 and i7-4790?Impulses - Wednesday, August 6, 2014 - link
I understand why that's not done often, but I agree it'd be neat.JarredWalton - Wednesday, August 6, 2014 - link
We'd need to have updated figures from the relevant parts, and unfortunately we don't have that for many of the older CPUs. Basically, I think Devil's Canyon and the Pentium AE are the only CPUs with a good set of performance data.bill5 - Wednesday, August 6, 2014 - link
why not a 8350 in this article? Kind of an odd selection of CPU's such as the secondary part the 8320.monstercameron - Wednesday, August 6, 2014 - link
from what I see this was supposed to be an easy read, not an in-depth analysis.JarredWalton - Wednesday, August 6, 2014 - link
CPUs were eliminated based on those that didn't have enough recent benchmark results. The 8320 and 8350 are very similar in performance, but the 8320 had quite a few more results so I used it and left off the 8350. I did include a few CPUs that don't have very many results, mostly because they're important parts -- the i5-2500K and i7-3770K for example only have results in about half of the benchmarks IIRC. Ideally we'd have all results for all CPUs, but that would require a lot of time just to get Bench updated with those numbers.robgiorgi - Wednesday, August 6, 2014 - link
Can you also show the Performance/Watt bars (they could be derived by the Performance bars since for each one you indicate the Watt peak). Even more interesting would be (but requires a actual power measurement) to show the Performance/Energy bars where the Energy should be the exact energy that is spent to get the given performance bar !Thanks for the article !
JarredWalton - Wednesday, August 6, 2014 - link
Unfortunately we don't have clear power use numbers for the various systems, particularly for the performance workloads being tested. CPU Bench has Idle and Load for some of the chips, but what we'd really want is a viable "average power" or "power used across all benchmarks" and we don't have that.Richard Wolf VI - Wednesday, August 6, 2014 - link
Can you please include results for the lower-cost Core i5s, such as the 4460? I would like to know if they offer an even better price/performance ratio than the 4690 or 4590.JarredWalton - Wednesday, August 6, 2014 - link
Unfortunately I'm limited to what we have tested here, and most of the i5 and i3 parts are missing -- or for those we do have, they only have a small selection of results. I'll most likely provide an update about once a quarter for this information, or perhaps a few weeks after any major CPU/APU launch.Richard Wolf VI - Friday, August 8, 2014 - link
No problem, thanks for the reply!dgingeri - Wednesday, August 6, 2014 - link
What this shows is that AMD really needs to step up their game. They need to take their current Piledriver design, take out the "second processor" half, (meaning, the second set of integer pipelines and the logic that makes them look like two separate cores) and put out 4 to 8 of those cores per processor. They might catch up to Intel then, at least partially. This idea of sharing the FP unit and decoders just didn't work.mmrezaie - Wednesday, August 6, 2014 - link
It could work if they have invested in compiler/driver (and generally software), but I think right now they are in panic mode, and we cannot expect anything serious coming out of them until 2016 (plus delays).TiGr1982 - Wednesday, August 6, 2014 - link
Well, at least, it seems that AMD realised that their Bulldozer concept with shared FPU did not play out well.Because their last Bulldozer-derived pure CPU (with no GPU) was (is) Piledriver-based FX (same for server Opterons), so that only APUs go beyond Piledriver (as current Steamroller-based Kaveri).
More than that, there are *rumours*, that the last generation of construction microarch, Excavator, won't even make it to the desktop APUs - Excavator-based Carrizzo supposedly will be released only for laptops (and dense servers).
Another *rumours* are out there that beyond Excavator AMD will release some new non-modular CPU microarch, presumably in 2016, so that the whole lineup of microarchs with shared FPUs will be on the market only from 2011 (Bulldozer) till 2015 (Excavator).
But take these *rumours* with the grain of salt - this is just what is written on the Web, but the validity of this info unconfirmed.
DanNeely - Wednesday, August 6, 2014 - link
Could we have a similar article for laptop chips in the near future? The splits between Haswell/Baytrail and Kaveri/Beema should make the CPU table a bit more interesting; and you've got a wider spread on Intel IGP performance as well. I'm not sure if the spread on amd's mobile IGPs is wider than for desktops or not.JarredWalton - Wednesday, August 6, 2014 - link
I'd love to be able to provide that, but unfortunately we don't have quite the same level of data -- and we're missing many of the low-end CPUs/APUs. Another issue is that laptops are reviewed as entire systems, so the presence or lack of an SSD can really skew certain results, not to mention the amount of RAM, inclusion of a dGPU (and which GPU is used), etc.It would be good to have a power/battery life graph as well, but again: the OEM of the laptop can really mess things up -- I have literally seen two laptops with the same CPU and chipset where one is nearly twice as power efficient as the other, simply because of poor optimizations and/or component choices by one OEM (and good choices by the other OEM).
Anyway, it's definitely something I want to look at, but I'm not sure yet how useful we can make it given the current set of information in Mobile Bench. Tablets and smartphones are another area we want to look at with this sort of article, and they have some of the same problems as laptops, but I think they tend to be a little more consistent on the battery life at least.
garadante - Wednesday, August 6, 2014 - link
Would it ever be possible to add a new section to the mobile bench to allow people to search for specific CPUs/GPUs? The data might not be 100% accurate comparison between best case scenarios between parts, due to thermal constraints, but it would still be helpful. Even having certain parts performing lower than you'd expect to see is useful because you know that thermals play a significant role for that part. I know there have been many, many times when I wish I could sort through all the mobile bench information by CPU or GPU to get roughly appropriate comparisons.BreezeDM - Wednesday, August 6, 2014 - link
I like the system cost price ratio. Adding $100 to the LGA2011 made the price per performance is realistic. I did a quick check cheapest LGA2011 motherboard is $200, Z97 is $90. And if you spend $30 for memory on Z97, then it becomes $60 for LGA2011. That totals $140 price difference, $100 is probably close enough because you probably could find lower numbers than I did.DanNeely - Wednesday, August 6, 2014 - link
For memory, you don't have to populate all 4 channels on an LGA2011; and unless you go with a 6/8 core model it's not going to matter outside of benchmarks. Lastly if you're trying to match specs as much as possible while populating the extra slots 4 half capacity dimms are a lot closer in price than a 2x multiplier ($70 vs $85 for 8gb)Gadgety - Wednesday, August 6, 2014 - link
Excellent piece. I've been looking into the A8-7600 for a family HTPC that could do a little bit more, at 45W in a passively cooled case. Being in Europe this particular APU is a little bit overpriced, even compared to the A10-7850k, due to a different pricing structure relative to the US, so the graphic performance/price numbers don't look quite as good. Having waited 8 months for its release, I can still wait a few more until the pricing is more favourable.ovigo - Wednesday, August 6, 2014 - link
Jarred, your unit in the first system cost table is wrong. It should not contain the $ as unit but be unitless as it is either normalised or in (benchmark points) PER dollar.TiGr1982 - Wednesday, August 6, 2014 - link
Jarred,Core i5-4670K average CPU performance on the first chart is too low; it's inconsistent with Core i5-4690(K) which is only 100 MHz faster but way faster on your chart.
I guess it has to do something with the fact that they were tested 1 year apart in time (different SW versions?)
Realistically, on your chart, if Core i5-4690(K) slightly outperform Core i7-2600K on average, then Core i7-4670K should almost reach Core i7-2600K - if this will be the case, then everything will be consistent, IMHO.
TiGr1982 - Wednesday, August 6, 2014 - link
To fix i5-4670K, IMHO, it suffices just to slightly scale down the result of i5-4690, according to their CPU frequencies.JarredWalton - Wednesday, August 6, 2014 - link
The problem is that the i5-4670K was not tested on as many benchmarks as the i5-4690K, and apparently the tests where it wasn't run tend to be more favorable. To be specific, there are 11 of the 29 CPUs/APUs that were only tested on about half of the benchmarks.TiGr1982 - Wednesday, August 6, 2014 - link
OK, thanks for the reply :)nekoken - Wednesday, August 6, 2014 - link
If you think Anandtech is corrupt and going downhill, stop reading it. I've found Anandtech to have an occasional bias. I think you could say the same for any review/reporting agency, profit or non-profit. If you perceive a bias it isn't exactly difficult to take it into account when reading the article.I sure wish there was a way to block seeing annoying posters in the comments.
mpbrede - Wednesday, August 6, 2014 - link
Jared, nicely done/ I've been hoping for something like this for a while. :)One more thing I think you can usefully add: If you factor in running costs for the various CPUs over a 1-year period, what does it work out to then in therm of value for performance?
Hrel - Wednesday, August 6, 2014 - link
It's really frustrating that when you talk about price you don't include the added cost of running AMD vs Intel. They cost more to keep on, they draw more power. No, it's not a LOT of money, but $5-$15 every year adds up.The computer industry being how it is today it's totally realistic to run those numbers out 10 years. At which point the small dollar amounts aren't so small anymore.
I'd really like to see cost of electricity, based US national average (which I believe is about 10c/KWH right now), worked into these "performance/dollar" calculations. It colors things even further in Intel's favor.
Hrel - Wednesday, August 6, 2014 - link
PS, looking at the architecture maps. Why does AMD's look like a toddler started breaking blocks to fit them together? Intel's is all symmetrical and clean. AMD's is sloppy and inefficient.JarredWalton - Wednesday, August 6, 2014 - link
Probably because a toddler (me) did the blocks on the AMD die shot. Hahaha... Thanks. I'm going to go cry in a corner now. ;-)As for the cost of running the systems over the course of years, well, that all depends on what sort of workloads you're running. With the systems powered on (and not really doing much other than web surfing/office work), AMD's APUs generally consume less power than Intel's CPUs. The FX-series CPUs are a different matter of course, and the same goes for LGA2011 Intel systems
While we don't have idle numbers for all of the systems (so I couldn't really provide data on this aspect of the CPUs/APUs), you can at least get a reasonable idea of power use from Legit Reviews' latest article on the 65W Kaveri parts:
http://www.legitreviews.com/amd-kaveri-a8-7600-a10...
Assuming 12 hours of day of moderate use (so let's say 80% idle, 20% load), using their numbers you'd spend about $27 per year running the i7-4790K compared to $14-$17 running the Kaveri platforms. I don't think most people are really going to care about spending $1 extra per month for a faster PC, and of course anyone using a discrete GPU will increase the power cost of having the system on by another $0.50-$1.00 even if they never use the dGPU for gaming.
TL;DR: Power costs really only matter if you're running the systems at load most of the time (e.g. servers), and if that's the case performance often matters even more. That's my opinion at least. :-)
et20 - Thursday, August 7, 2014 - link
This is an excellent article.Thank you so much for writing it, Jarred.
Could you please add a couple of scatter plots with price and performance axes?
Most products usually fit close to one of two asymptotes and the sweet spot is at their intersection point.
JarredWalton - Thursday, August 7, 2014 - link
I talked about this with Anand; basically, I find scatter plots to be more advanced and difficult to understand, and as this was targeting the "easy to grok" crowd I opted for simple bar charts. Still, if you want a scatter plot, here's the performance vs. system cost plot:http://images.anandtech.com/doci/8312/ScatterPlot-...
What's amazing to me is how little AMD has changed on the CPU performance curve for the past several generations of APUs. Granted, the graphics plot would look better, but still... Of course, there are plenty of use cases where having more than 2 CPU cores isn't all that important.
duploxxx - Monday, August 11, 2014 - link
Jarred, true on the CPU part of AMD, but if you look at the power draw you can actually see that they did make a lot of improvements on power/perf while not able to have the most high tech fabs and process...And statement again, how much CPU is really needed for day time tasks? have been reading for approx 1 hour some articles on anandtech, did i really needed my fancy i7 for that?
mapesdhs - Monday, August 11, 2014 - link
Surely though you have power saving features, etc., turned on, so while doing simple tasksyour i7 is hanging back, not drawing power it doesn't need? In that state, an i7 isn't much
different from the other CPUs. Efficient oc'ing then depends on using dynamic vcore, which
is a bit more difficult, but well worth the learning curve (good piece about it on the ROG site).
Jarred, consider yourself awared numerous medals just for using the word, 'grok'. :D
Ian.
mutantmagnet - Thursday, August 7, 2014 - link
I'm confused about why you would ONLY want to include the cost of discrete GPU for the FX Cpus. Some AM3 boards do come with it on board. You could've had both.I could see why you would want to avoid comparing CPUs after overclocking to ensure this suitable for all audiences but I think a concession should've been made here if you are going to compare what's left of the FX-series landscape.
JarredWalton - Thursday, August 7, 2014 - link
I included a $50 dGPU in the cost of both the AM3+ and LGA2011 platforms. That's mostly because I figure anyone buying an AM3+ or LGA2011 rig are going to use a discrete GPU. The AM3+ boards that have integrated graphics are basically very old and slow IGPs at this stage in time. And really, I don't expect most people to buy either platform these days, as they've both been around for some time and neither one is due for any additional updates that I know of.know of fence - Friday, August 8, 2014 - link
My approach to the big picture would be to find a fair point of comparison. AMD APUs offer the (single thread) performance of Nehalem CPUs (the one before Sandy Bridge), so even though AMD builds are viable they are about 4 to 5 years less "future proof", but they have well balanced power efficiency, come with better graphics and a significantly smaller markup. Desktop Kaveri APUs basically match the CPU performance of a mainstream MOBILE Haswell CPU clocked at around 2 GHz.http://www.anandtech.com/bench/product/191?vs=1200
nand - Saturday, August 16, 2014 - link
performance per watt ratio is not to be forgotten for a long term effect on electricity bills...