AMD Breaks Overclocking Record With Bulldozer

MojoKid writes "AMD recently held a press event at their Austin headquarters, offering hands on time with the company's upcoming Bulldozer-based FX-line of processors. Many of the details disclosed are still under NDA embargo, but AMD is allowing a sneak peek today to go along with a claimed Guinness World Record announcement. A team of overclocking enthusiasts and AMD engineers had a sampling of early AMD FX processors running at around 5GHz with high-end air and water-cooling, in the 6GHz range with phase-change cooling, and well over 8GHz on liquid-nitrogen and liquid-helium setups. Voltages of over 1.9v were used as well for some of the more extreme tests. The team had access to dozens of early FX processors and methodically worked through a batch of chips until ultimately hitting a peak of 8.429GHz using liquid-helium, breaking the previous world record of 8.309GHz for modern processor frequency." Update: 09/13 13:54 GMT by T : Adds user Vigile: PC Perspective was there and took some photos and video of the event.

Comments

8ghz

[Ruin666]

8ghz sounds quite sexy. lets see how it goes with an old corsair h70 tho...

Re:8ghz

[fuzzyfuzzyfungus]

Is their any word yet on how much punch the bulldozer cores get for their MHz? If this is AMD's version of Netburst, 8GHz will likely end up being merely competitive. If it does well per-clock and hits 5Ghz on air...

Re:8ghz

[Dunbal]

Installed an H-80 in a build I just made for my dad. They don't make the H-70 anymore. The H-80 is kinda cute and even better. The manual sucks though.

Re:8ghz

[Anonymous Coward]

Sadly this was my thought as well. Over 9000GHz on liquid unobtainium cooling is fine and great, but I'd be more interested in max OC and actual performance using methods easily obtained by the masses (as well as feasibly used, since tanks of chilled mineral oil are not).

Re:8ghz

[santiagodraco]

Might as well ask how it would do with a Zalman CPU fan. Corsair is no yardstick in Water Cooling by ANY stretch of the imagination. What it is is decent consumer level WC, but we are talking enthusiast here which means real deal WC components.

Re:8ghz

[spire3661]

It most certainly IS a yardstick for water cooling, the mainstream yardstick. Praise fancy full water cooling all you want, you will never see market penetration like the new self contained WC systems rolling out now. They work, and work well enough to make you forget about a full water rig and all its hassles.

Re:8ghz

[hairyfeet]

But you see, here is the thing about that. Unless you are one of the rare (although admittedly more of them exist here than in most places) people that are slamming the living hell out of your CPU and need to scrape up every single last flop you can CPUs are long past good enough for pretty much any task the common man can think up.

I was a lifelong Intel+Nvidia man until bumpgate and the massive bribery and compiler rigging came out, then for the first time since the K6 I built an AMD machine for myself, and you know what? Frankly even without going for the top o' the line chips its insanely overpowered. Video editing/transcoding, DVD ripping, gaming, you name it my 925 quad takes it like a champ and keeps right on coming and the whole thing, with 3Tb of HDDs, 8Gb of RAM, dual burners, HD4850, and Win 7 HP X64, cost me barely $700 before MIR, around $640 after. And that was before the prices dropped, now you can probably knock a good $150 or more off that easily.

I am a hell of a lot harder on a machine than the average folks which come to my shop so for them the bang for the buck is even more insane. When you can hand them a triple or quad for $500 or less and still make a decent profit? That is just nuts! But I bet if you were to check your CPU usage long term a good 80%+ of the time your CPU is twiddling its thumbs waiting for you to find some work for it to do. the only ones I've found that might need the extra 20-30% speed boost going Intel might give you and are willing to pay the 200% higher markup for that speed are guys doing a shitload of compiling. like I said there are more of those type here, but for everyone else? You can get an insanely overpowered and loaded to the brim machine for dirt cheap by going AMD. The bang for the buck is so far in the AMD camp now it isn't even funny and that is before you count the kick ass GPUs compared to the shite on a crusty roll that Intel calls a GPU.

So if you are not one of those 4% that slam their CPUs and push them to the absolute limit why would you spend the extra money on Intel? Like I said monster quads for less than $500 and you can even buy quad laptops for that price if being mobile is your thing. Its just nuts!

Hmmmm......

[robthebloke]

... Although I'm sure it's no easy feat, I'd be kinda more interested if they couldn't break the record (thereby implying we may be at the end of moores law)

Re:Hmmmm......

[Anonymous Coward]

Moore's law has to do with number of transistors on a chip, it has absolutely nothing whatsoever to do with processor frequency.

Re:Hmmmm......

[fuzzyfuzzyfungus]

Given that Moore's law deals with transistor density, rather than transistor switching speed, I assume that that question will be settled in the fab, long before the overclockers ever get their hands on the goods...

Re:Hmmmm......

[I.M.O.G.]

In regards to frequency scaling for Moore's Law, that came to an end in 2004 essentially. It's one topic wikipedia has right if you want to read the details. These days, moore's law only holds true for transistor density, which is why everything is multicore, power efficiency, and integrating more features on chip - there are extra transistors they can fit on the chip and they are finding more things to use that die area for.

Re:Hmmmm......

[UnknowingFool]

Like integrating the GPU. Granted you can't play l33t games at uber settings but for casual games it's acceptable. Given the graphic demands of average consumers focusing on video is the better choice for integrated GPUs.

Re:Hmmmm......

[julesh]

Moore's law has to do with number of transistors on a chip, it has absolutely nothing whatsoever to do with processor frequency.

"Absolutely" nothing may be overstating the point somewhat: as transistor density increases, the size of the transistors decreases (duh). Smaller transistors require less charge to build up in order to switch, therefore can switch faster at the same voltage as a larger transistor. Admittedly, the maximum voltage they can handle tends to drop, but at least so far the tendency has been smaller transistors -> faster switching -> higher processor frequency.

No New Egg Warranty

[Dareth]

I bet these liquid-helium cooling kits do not come with a warranty from New Egg!

Re:No New Egg Warranty

[Anonymous Coward]

At least when there's quench, you talk real funny for a while, so its not a total waste.

Signal propagation limits

[Anonymous Coward]

Due to the speed of light, isn't there a limitation on how fast you can clock a CPU die? As I know, eventually you will run into problems maintaining synchronicity.

Re:Signal propagation limits

[Anonymous Coward]

Everyone knows light moves faster when it's cold!! That's why they cool them!!

Re:Signal propagation limits

[Lumpy]

I though the Police solved the Synchronicity problem in 1983....

https://secure.wikimedia.org/wikipedia/en/wiki/Synchronicity_(album) - a wiki article about their research.

Re:Signal propagation limits

[ShakaUVM]

>>Due to the speed of light, isn't there a limitation on how fast you can clock a CPU die? As I know, eventually you will run into problems maintaining synchronicity.

A nanosecond = one foot. (Take that, metric system!)

You can calculate the theoretical maximum thereby, by just plugging in clock rate and die size.

Re:Signal propagation limits

[fuzzyfuzzyfungus]

Luckily, the design limit, depending on how clever you are and what sort of work is being done, can be slightly less dire than it appears:

It isn't necessarily a problem if you have multiple signals "in flight" on the same line; but it is a problem is signals that are supposed to arrive simultaneously on different lines stop doing so as increasing clock speeds ratchet up the requirements for what "simultaneous" means.

You can already see it happening with external busses: check out the traces between the CPU socket and the DIMM sockets on any remotely recent motherboard. Some of them are about as straight as routing allows, some are densely squiggled, because they all have to have almost exactly the same delay.

As clock rates get faster still, the acceptable deviation in delay between "identical" lines gets smaller and smaller and the penalty(in clock cycles) for accessing a piece of silicon some distance away(ie. RAM connected to a different CPU in a multi-socket system) rises.

Re:Signal propagation limits

[tepples]

Their solution won a Grammy, but the public won't be able to make full use of it until 2078.

Re:Signal propagation limits

[Anonymous Coward]

Second is a metric system unit...

Re:Signal propagation limits

[ShakaUVM]

>>Second is a metric system unit...

Yeah, but a foot isn't. That's the funny thing. A foot is a better natural unit for distance than a meter.

We really ought to be using kilofeet and millifeet rather than meters. =)

Re:Signal propagation limits

[blueg3]

The speed of light is a limitation. Like a sibling commenter says, the speed of light is one foot per nanosecond. Propagation time through transistors is slower than this, but it's a good estimation tool. However, you don't have to have an operation go all the way through the system in a single clock cycle. Hence, pipelining.

Re:Signal propagation limits

[hackertourist]

A foot is a better natural unit for distance than a meter.

No it isn't. It just seems natural to you since you're used to it. To everybody not still wedded to archaic units, meters are perfectly natural and usable.

Re:Signal propagation limits

[julesh]

Sound recording whose copyright holder is European. Copyright expires in Europe after 70 years. Copyright protection in US under Berne Convention is only valid as long as the work is under copyright protection in its country of origin, so also expires in the US after 70 years. (IANAL; this is not legal advice.)

Re:Signal propagation limits

[ShakaUVM]

>>No it isn't. It just seems natural to you since you're used to it. To everybody not still wedded to archaic units, meters are perfectly natural and usable.

Natural in the scientific sense, not the psychological sense. In other words, Kelvin are a better unit for temperature than Celsius, since you have to convert Celsius to Kelvin in order to do any thermodynamic calculations. A nano-lightsecond (aka a foot) makes a lot more sense to work with than the rather arbitrary meter unit. Call it a nls.

In fact, this thread shows exactly why that is true. If people used feet/nano-lightseconds instead of meters, the OP would have been able to much more easily calculate how far travels in a 1GHz processor (i.e. one foot) or a 8Ghz processor (1/8th of a foot or .125nls) and how that compares to the 17mm die size. Ah, see, but wait - we're using the non-natural mm measurement, so we have to convert to nls! 17mm = 0.0558nls. So we're still below the theoretical maximum by about a factor of two, though path length issues are obviously going to set a hard limit before that.

Re:Signal propagation limits

[Anonymous Coward]

A nanosecond = one foot. (Take that, metric system!)

You can calculate the theoretical maximum thereby, by just plugging in clock rate and die size.

Well, not really. In modern processes signal propagation time isn't the limit, it's charging and discharging capacitors. Long wires have too much parasitic capacitance (and relatively high resistance since they have to be very thin these days) so RC delay is very important. Games can be played with selectively using beefier drive transistors for long wires, but this costs die area and power.

I do a lot of work with FPGAs. They're not a great way of analyzing ASIC behavior because FPGA signal routing has all kinds of interruptions to make it a reprogrammable routing matrix, so there's always lots of hidden buffers and muxes and whatnot going on. But to give you an idea... it can easily take 10ns to get a signal from one edge of a large FPGA die to the other. 10ns ~= 120 inches, and a large FPGA is probably about 1" wide, so we're talking about a factor of >100x. Take off a factor of 10 for a SWAG at the differences between FPGA and ASIC routing, and you're still looking at 10x the wire delay predicted by the 1ns=1ft maxim.

Wire delay has been a real problem in chip design for years now. For example, it's known that most or all members of the Pentium IV family had pipeline stages whose sole purpose was to move data from one part of the chip to another. No computation, just move bits from point A to point B. That was one of the PIV's flaws: trying to go for extreme clock speeds on those process nodes forced them to do tricks like this, and the tricks burned lots of power and made the design less efficient in extracting real-world performance from that high clock speed. (Deep pipelines are the friend of clock speed and the enemy of performance efficiency.)

Re:Signal propagation limits

[ShakaUVM]

>>Well, not really. In modern processes signal propagation time isn't the limit, it's charging and discharging capacitors.

I think you missed the "theoretical limit" part of the calculation. =)

It's a fun back of the envelope calculation. 1GHz = 1 foot. 8GHz, therefore, is 1/8th of a foot.

All about who has the biggest

[bongey]

http://www.youtube.com/watch?v=GiWQZhUmmRw

Distraction.

[Zeek40]

Is AMD hoping this impractical PR stunt will distract us from the fact that their chips have been lagging behind Intel's for the past two years?

Re:Distraction.

[pinkj]

To almost a tenth of the price? I'll take it. I don't need the fastest. Just fast enough for my needs.

Re:Distraction.

[jedidiah]

I was going to say something similar but you beat me to it.

Re:Distraction.

[Anonymous Coward]

Then why did you post instead of just modding him up?

Re:Distraction.

[Albert Sandberg]

Then why did you post instead of just modding him up?

You need mod points for that stunt.

Re:Distraction.

[Anonymous Coward]

If only you'd bought Intel (bing bong bing bong) TM.

Re:Distraction.

[fuzzyfuzzyfungus]

I doubt it, most new architecture releases have some sort of little overclocker party, whether sponsored by the CPU manufacturer, or a bunch of gamer-gear motherboard, power, and cooling widget outfits...

The OEMs who ship the bulk probably don't even notice them, and I'm guessing that the market for ludicrously overclocked servers is pretty much nonexistent.

Re:Distraction.

[Anonymous Coward]

Maybe at 8.4GHz they finally have better single-core performance than Intel.

Re:Distraction.

[Anonymous Coward]

I'm guessing that the market for ludicrously overclocked servers is pretty much nonexistent.

I think that no one in is right mind would risk overclock a server, the failure risk is not acceptable.

Of course that can be appealing to some but to risk downtime for some GHz... its just crazy.

Re:Distraction.

[LWATCDR]

Okay excluding the L33T gamers, super heavy CAD users, HD video producers, Movie F/X houses, and research labs no one needs the power of Intel's high end desktop chips. Frankly some of those users like the video and science users do most of their stuff on the server CPUs anyway.

Impractical... Well yes today but I thought the same thing when I heard about Intel running a cpu at 100 mhz with ln2 cooling back in the 80s. Of course that was when we thought an 8 Mhz system was fast. They where getting 5 mhz on air so that is of interest to the normal enthusiast community. What this really shows is that the new core is both fast and has a lot of head room built in. The old record was held by an Intel CedarMill core. It does look as if the Bulldozer will be interesting.
And please cut the fanboy crap. AMD makes good CPUs in their price points. They don't play with Intel at the high end in the desktop but then I don't know many people that pay $1000+ for a desktop CPU.

Re:Distraction.

[fuzzyfuzzyfungus]

I could imagine that the "stack-em deep, sell-em cheap" fault-tolerant cloud guys considering the possibility(at least during the mature phases of a given chip's lifecycle, where the likelihood that a chip has been binned down purely for market segmentation, not for failing validation, is much greater); but the energy costs would keep them from doing much more than, possibly, cranking a lower speed grade chip up a tier or two, since energy use tends to spike pretty badly once you get outside of a given design's marketed speeds.

Cranking it up straight into nutty overclocker territory would be counterproductive(if your overclock involves custom modifications to the motherboard VRM, you are probably blowing the datacenter's energy budget...); but being able to trade a warranty for a free model number or two bump might be worth it, depending on the maturity of the chip.

Re:Distraction.

[wye43]

Because your "needs" are clearly defined and measurable, right?
You need your Notepad to open up between precisely 200 and 400 ms. In no way you could use a blazzing 15ms loading time, or you could settle for a lousy 450ms. RIGHT?

Not that processor speed is the most influential on loading time of a program, but you get the picture. Most of the users don't have a precise measurement for what they "need". And my most I mean pretty much mean ALL personal users, including myself.

Please stop with "this is WAY too much processing power, its overkill" lines. Its getting boring, and we all know you crave it too, you just down-rate what you can't have. Its hypocrisy. Any stupid little user on this planet could have a very nice use for a 1 Petahertz processor, or more.

Re:Distraction.

[Runaway1956]

Impractical PR stunts have their uses. They help to focus attention on what is possible today, and what will become commonplace in years to come. 8 Ghz may remain impractical for another two years, or ten years. But, eventually, people will come to expect that, and more. Someone has to do the early experimenting!

Re:Distraction.

[Noughmad]

Okay excluding the L33T gamers, super heavy CAD users, HD video producers, Movie F/X houses, and research labs no one needs the power of Intel's high end desktop chips.

I'm none of the above, but I don't like waiting more than a minute for a program to compile. And I compile often.

Re:Distraction.

[LWATCDR]

Get an SSD and more ram.
Even then unless you are writing giant modules which are you are not supposed to do you will not need to wait more than a minute to compile. After all you only compile the module you are working on.
I use a Core2 and my compile times are well under a minute. Now if I do a clean and build it is longer but no one does that often.

Re:Distraction.

[rubycodez]

Intel and AMD are both massively lagging behind in the state of the art. PPC whoops their ass

Re:Distraction.

[Anonymous Coward]

Okay excluding the L33T gamers, super heavy CAD users, HD video producers, Movie F/X houses, and research labs

...what have the Romans ever done for us?

Re:Distraction.

[Ferzerp]

And someone did. Back in the NetBurst era, 8Ghz was attained. Must mean NetBurst was a great architecture.... :snicker:

Re:Distraction.

[Anonymous Coward]

Not only that, but there's also Intel 'Product Differentiation'. Things like hardware virtualization extensions only work on Intels higher-end or server models while on AMD every cpu has the same basic features. If you don't care about performance-per-watt, AMD is still the best value for money.

Re:Distraction.

[nedlohs]

Do you often launch into rants that have no relation at all to the post you are replying to.

Said post didn't mention "too much processing power" or "overkill" or the requirement of any other person than themselves.

In fact all they said is that they'd take the one with 50% of the performance for 10% of the price since it seems good enough. Obviously faster would be better, but that poster happens get more utility spending the money on something else.

Why would you get so worked up about someone elses simple economic choices?

Re:Distraction.

[Shinobi]

"Okay excluding the L33T gamers, super heavy CAD users, HD video producers, Movie F/X houses, and research labs no one needs the power of Intel's high end desktop chips. Frankly some of those users like the video and science users do most of their stuff on the server CPUs anyway."

Bull fucking shit.

Many non-geek amateurs/home users definitely want as much CPU performance as possible. In fact, as someone else wrote in a comment on this site a few months back, many software dev geeks need less CPU/RAM/disk than many home users plugging away at their hobbies these days. Many home users do video editing, music production, heavy duty graphics work, just as a hobby. There are people who don't give a fuck about computers other than as a tool, but it helps them do their other hobbies more efficiently.

As for the price of the CPU? When I upgraded my gaming/hobby rig, I went from Athlon 64 to Intel Sandy Bridge i5-2500... Because for the games and hobbies I do, it outperformed the at the time more expensive AMD X6 1055T, and I needed a new mobo+new RAM anyway. And Sandy Bridge is Intel's low/mid end. Ivy Bridge will be the top end. Bulldozer is competing with the now 9 months on the market mid end from Intel.....

Re:Distraction.

[MikeBabcock]

Would you care to do the math on FLOPS per dollar? Because if you care about money at all, AMD has been winning the performance game for a while now.

Re:Distraction.

[Calos]

This is modded insightful? Sheesh.

A TENTH of the price? Yeah, maybe if you compare the most expensive Intel part to the least expensive AMD part, and completely ignore all those other Intel SKUs. I'm not sure if you're being dishonest, or are just really that obtuse.

I don't see any pricepoint where AMD offers a better value than Intel. They do have parts that are just plain cheaper - a quick search of Newegg indicates that only AMD has a new chip for under $40, an older Sempron. Above that, you can find similarly priced Intel and AMD chips right on up. Have you taken the time to look? Have you taken the time to compare? Or are you just a fanboy spreading FUD, or a non-fanboy who just hasn't changed their song in about 8 years? For reference, look at a chart like this one: http://techreport.com/articles.x/21208/18

A specific example from that chart: Intel's i5-2400, destroying the AMD lineup, yet priced the same as the X6 1075T and X4 980, and less than the X6 1100T, which it soundly trounces.

This is the case right down to the $100 range, too, about the lowest that chart goes.

I just don't understand how people can support AMD, except out of a root-for-the-underdog mentality, ignoring all else. Intel has design superiority, process capability superiority, technology superiority, manufacturing volume superiority. Given those advantages, you really think Intel is going to just leave a whole segment of the market to AMD? Sure, they're going to sell their parts at the highest prices they can first, but not everyone is buying $300 processors.

Re:Distraction.

[KingMotley]

They don't play with Intel at the high end in the desktop but then I don't know many people that pay $1000+ for a desktop CPU.

They don't play with Intel at the "high end" of $280 http://www.electronics-emporium.com/products/Intel-Core-i7%252d2600K-Processor-3.4GHz-8-MB-Cache-Socket-LGA1155-%252d%252d-1FO004ROHKFN06.html, which quite a few people do pay for.

Re:Distraction.

[MikeBabcock]

I believed not so long ago that CPUs were way over the speeds the average consumer needs, that is until I watched processor usage on a modern laptop doing live high resolution video chat.

For people who want to do face to face teleconferencing, high speed real-time video encoding at good resolutions and framerates will really eat up your CPU power. Its not an edge-case anymore either as more and more people expect to be able to do high quality video calls (and not the crap MSN pawned off on people for years).

Re:Distraction.

[Anonymous Coward]

Uh, the too much processing power lines probably come from (1) talking about processors with enough cores that some will always be idle or (2) talking about processors with such a high clocks speed that the processor is almost never the bottleneck. CPU-bound tasks are relatively rare these days, especially for your common web/e-mail/office user.

Re:Distraction.

[Calos]

Stop being so willfully obtuse. Intel sells parts all over the range, competing at nearly every pricepoint, and in general offering a better value proposition to boot (performance/price). For example:

http://techreport.com/articles.x/21208/18

So yeah, cut the fanboy crap, and the FUD about Intel being so expensive. You can get newer Intel parts just a hair above $100, and they're still good in that range. At least as good as, if not better than, similarly prices AMD chips.

Re:Distraction.

[GameboyRMH]

HFT servers have rack units with hot-swappable CPUs. They're considered consumable. They overclock them to high heaven and pop in new ones when they burn out. I wouldn't be surprised if they've come up with a magazine / chain loading system by now...

Re:Distraction.

[aztracker1]

At this point, if you are using an HDD over an SSD that's your most likely bottleneck. I rarely find compiles these days CPU bound on my Core i5 at work, or the i7 at home. Neither is at the highest end of the cpu pricing scale.

Re:Distraction.

[LighterShadeOfBlack]

What? He never said anything about "overkill" or anything remotely hypocritical. His argument was simply that he gets performance he finds satisfying at a proportionally superior price-point.

Reactionary much?

Re:Distraction.

[Shinobi]

Not quite true. It was true for some real-world workloads, while Intel won at some other real-world workloads. Then came Sandy Bridge and kicked AMD's ass all over. I went from being a 9 year long user of AMD over to Sandy Bridge when it was time to replace my old, aging Athlon 64 3500+, mobo and RAM, because for what I do, the i5-2500 beat the shit out of the X6 1055T(which was more expensive than the i5-2500 at the time too...)

And Sandy Bridge is low and mid end for Intel. High-end desktop/consumer performance comes with Ivy Bridge.

So the fact that AMD's high-end competes with the mid-end from Intel that's been on the market for 9 months now is rather worrying.

Re:Distraction.

[Anonymous Coward]

Okay, your post is a little bit of flamebait, but I'll bite.

I can purchase the A-3850 for about $40 less than the i5-2400. Certainly, the i5 will trounce the A-3850 up to the point where you want to process graphics. Once you start to process graphics, you can't really compare the Intel HD 3000 to the on-die Radeon 6550. If I want comparable graphics, I have to purchase $150 graphics card. I can also purchase the ASUS FM1-Pro motherboard for about $50 less than a similar ASUS board for the i5 (with comparable features). After this setup, my A-3850 will probably peak (with full graphics acceleration) at about 150 watts, while the Intel would likely run at 250 watts or more.

Taking into account power consumption, the AMD setup won't be 10% over 3 years, but it might be 30% the cost of the Intel setup. Unless I'm doing a massive amount of heavy lifting (like constantly transcoding HD video) then the Intel might be worth while. But, my PC is idle 95% of the time with only a few spikes of CPU usage.

Re:Distraction.

[Anonymous Coward]

(Still wye43, just too lazy to log on the phone)

"need"

The original post was not about economical choices, but having "what I need" as a performance limit. Perhaps we are each talking about different posts :)

On the other side, are you ranting about other people's rant often? :)

Re:Distraction.

[LWATCDR]

Well on a modern system that should be off-loaded to the GPU. Even then I would bet that a Sandy-Bridge i3 or i5 would be more than good enough for that task. You sure don't need a quad core i7 for that.

Re:Distraction.

[nedlohs]

So you also don't understand the meaning of need. OK.

Clearly his computer lets him do what he does on it, that's so obivous it is a tautology. Thus it meets his need.

If you want to be technical of course he doesn't "need" it at all. I promise take that computer away completely and he continues to live.

Re:Distraction.

[Anonymous Coward]

In a modern system, the user would be doing live chat, streaming a 1080p video, rendering his browser and desktop, and [optionally] encoding a video... all at the same time. And each of those tasks could be utilizing the GPU.

Re:Distraction.

[Rockoon]

I'll raise your "single sample" techreport FUD about the A8-3850 and go all-in with 122 samples (and counting) for the A8-3850, from real users, using real systems, that they really purchased, from real retailers....

Every single time you find a benchmark which significantly disagrees with the large sample set taken by PassMark on cpubenchmark.net, you have found a benchmark that is complete bullshit. Now that you know that techreport puts out bullshit benchmarks.. will you ever attempt to cite them again?

My guess is that yes, you will.. as long as techreport continues to favor Intel... you will blindly swallow their shit.

Re:Distraction.

[Rockoon]

Thats all well and good... as long as you ignore the fact that Intel only has two chips in the top-20, (#5 and #20) and neither of them are SandyBridge chips.

Re:Distraction.

[Calos]

Actually, from that same article, an $80 card (HD 6670) often provides double the performance of the 6550 IGP. Sorry, but the IGP scores aren't spectacular for the A-3850. You could get away with less on the discrete side to match it, and this is with an i3-2100, which costs less than the A-3850, and is faster (and more power-efficient) at CPU-based tasks.

By all means, the A-3850 probably provides playability for the lowest upfront cost (well, not sure about that, I'd imagine you can get the same or cheaper by buying generation-old CPUs that offer similar performance for cheap, plus discrete graphics, which is exactly what the text in the article suggests). But you just get playability. Right now. In some games... already struggling in newer ones. But you'll be replacing it in a year, or buying that discrete graphics card... just to find yourself quickly CPU-bound (already are with the discrete 6670 on some games).

So, I'm having trouble buying that the A-3850 is a good value for the gamer, or that it's going to save them much money. If you're not gaming, the Intel IGP is just fine, and has some fun new features like assisted encoding (not just decoding), making it faster and more efficient.

The only real benefit I see from the A-3850 is lower power draw at idle. If I was running a system 24/365, that might add up. I'd estimate my system is only up about 3000 hours a year, though. The A-3850 uses, from other Tech Report articles, about 25 W less at idle with the IGP only and compared to the i3-2100 with a discrete 6670. At 95% idle and with the power rate in my area, that's $6.47/year more, using your 100 W at-load differential. Is my use below average? Fine, let's expand the numbers to 24/365 operation (8760 hours), still keeping the same 95% idle figure (which unfairly inflates the usage time, when most of the time added - e.g. when people are sleeping or at work/school - will be at idle). The hypothetical Intel rig will consume an extra $18.89 a year.

Over three years, with a compromise on power use: $30 (power) + $35 (A-3850 price - i3-2100 price + your claimed motherboard differential) + $60 (graphics at least equal if not better than the A-3850) = $125. Subtract from that the amount you spend upgrading the whole "APU" (stupid name), which likely completely wipes this out, or the ~$50 you might spend on a discrete part and the corresponding difference in the power consumption differential... From a cost perspective, it's more or less a wash. You might come out slightly ahead, I'll enjoy a more capable system for three years and not have to putz around upgrading it.

No where close to 10% or 30% of the cost, though. That's mainly my point. I'm not trying to advocate Intel. I think people should look through the numbers and work out what's best for them based on what they use. Shock claims of 10% or 30% TCO are just FUD. Ironic that despite that being demonstrably false, in your eyes, it's me who's flamebaiting.

Re:Distraction.

[theArtificial]

The same awesome sandy bridge which was involved in a $700million recall?

Re:Distraction.

[Calos]

Ha, get over yourself.

As I said, I'm no Intel shill. I'm running an AMD based computer that I built, if that helps.

No, I like the Tech Report because their methods are transparent, scientific, and generally reproducible, from what I've seen on other sites. They also produce great in-depth articles, digging through CPU and GPU architecture. And if it helps your inferiority complex any, they were all for AMD a few years back when Intel was struggling with Pentium D and such. Not even grudgingly accepting, but enthusiastic.

First time I've encountered PassMark. Tell me, why do you place all of your trust in it? I looked at their website. I could find nothing but bland and technically insufficient description of what they test... which is almost entirely synthetic benchmarks anyway. No implementation details, closed source, small suite. Submissions aren't transparent, and for all you know they're playing games with the numbers, either intentionally, or through shitty code. It doesn't even report the results in each section of the test, and applies some "weighted average," the details of which they choose not to declare. How does their software detect hardware? Are fanboys stupid enough to try to spoof it, just to screw with results? Does the testload they provide match what you do with your computer, or any real-world consumer use?

Sorry, but I'll take real applications, real world usage, transparent methods, transparent metrics, and a staff of knowledgeable technical writers over some commercial, closed benchmark any day.

It's ironic. You ask me if the fact that one source contradicts mine would make me reconsider. The answer? Sure, with evidence. It's troubling, at the very least. But what about the reverse. Do you question your beliefs, given that multiple other sources disagree? Or do you stick to PassMark, because it shows what you want it to show? Here's something that should give you pause: Symantec endorses PassMark, as PassMark gave them some of the highest marks for Symantec antivirus performance, both consumer grade and small business. That should set off alarms for anyone who has ever used Symantec software before.

Re:Distraction.

[Belial6]

You were sort of right that the CPUs were way over the speeds the average consumer needed. The thing about processor speed is that the the speed need is not linear. Processors often out strip the current need for most users FOR THE SOFTWARE CURRENTLY AVAILABLE. There are certainly more things the users could use, the software would use requires processors that are faster than what is available. Thus the software isn't written, so their CPU speed IS way over the speed they need.

Once the speeds get fast enough, new classes of software are introduced, and the average users no longer have processors that are faster than they need. I can imagine all sorts of software that users would love to use, but the hardware is not even close to being up to the task of accomplishing it. On the other hand, the fastest i7 is overkill for the software I currently use.

Re:Distraction.

[LWATCDR]

"And please cut the fanboy crap. AMD makes good CPUs in their price points. They don't play with Intel at the high end in the desktop but then I don't know many people that pay $1000+ for a desktop CPU."
I simply said that they make good CPUs at their price points and they do. The new APUs combine a good graphics chip with the CPU. The I3 and I5 are better CPUs with a pretty crummy GPU.
Also that chart is only CPU performance and doesn't take into account the GPU. At this point when looking at systems I feel that one should look at CPU and GPU together. Today we do not often benchmark integer and floating point separately it doesn't make a lot of sense to just look at the CPU alone.

Re:Distraction.

[Rockoon]

Passmark is measuring real systems, mainly as assembled and delivered by companies like Dell, to Joe Public who then runs PassMark.

You can make excuses all you want for why you would rather believe the obviously less realistic benchmarks of the tech sites, but they are still just excuses.

While you think I need to get over myself.. you need to get over the obvious flaws in the "information" delivery framework that you have been listening to.

You claim that I am listening to one source, when in fact...

...for benchmarks of the i7-2600K I am listening to 4257 sources (869 benchmarks of this CPU in the past 30 days alone.)
...for benchmarks of the i5-2500K I am listening to 2882 sources.
...for benchmarks of the i5-2400 I am listening to 692 sources.

This is crowd benchmarking. I am not listening to one source. I am listening to a crowd of sources.

With their software I can even narrow things down to specific motherboards, or specific memory chips. Compare memory bandwidth of various setups (actual results of many people) .. or single threaded integer performance.. or multi-threaded simd.. etc...

You are in the stone age.

Re:Distraction.

[Catnaps]

You're assuming that the benchmarking code is reliable, and are utterly blinded by the idea that it's crowd-sourced benchmarking that you can't even stop to question the fundamentals. And you're saying someone's in the stone age. Right.

Re:Distraction.

[Catnaps]

Oh, one more thing. People don't use computers for benchmarks, they run these things called games and applications on them. Perhaps you've heard of them?

Psst... real world benchmarks > synthetic benchmarks.

Re:Distraction.

[Intropy]

You can get an Athlon II X3 455 for $80. Intel's cheapest competitor is probably the Pentium G850 for $100. But if you're spending $100 you could upgrade to the Phenom II X4 840. $120-$125? Phenom X4 955 Black vs. Core i3-2100 is close; the former has better parallelism and is $5 cheaper, while the latter is more energy efficient. If you want to get into overclocking then the AMD offerings soundly beat the Intel ones. If you want much more CPU you really have to make a big jump in price unless you want really poor marginal returns on your dollar. Once you get up into the $190 range or so Core i5-2400, Intel has superior performance/price. It certainly appears to me that it is reasonable to conclude that AMD offers superior value at the low to mid range while Intel does at the high end without being a "fanboy spreading FUD."

Re:Distraction.

[Lanteran]

Get some RAM, mount a small partition to RAM for usage during compiling. Should offer much more of a performance boost than a more powerful processor, hard drives being the main bottleneck on performance in this case.

Re:Distraction.

[jones_supa]

And sometimes it's nice to participate in the discussion rather than just increment some number.

Re:Distraction.

[Anonymous Coward]

The same awesome sandy bridge which was involved in a $700million recall?

The chipset released at the same time as Sandy Bridge was recalled, for a single-transistor design flaw which can cause the SATA ports to fail. (Take home lesson: one mistake out of millions or billions of transistors in a chip design can cost $700M if it's not caught before volume production starts.) The processor itself was fine.

Re:Distraction.

[Anonymous Coward]

Thats all well and good... as long as you ignore the fact that Intel only has two chips in the top-20, (#5 and #20) and neither of them are SandyBridge chips.

ROFL. Your link is a chart which has tons of discontinued, ancient, and slow discount-bin CPUs (like an Opteron 252, of all things) placed high just because some random garage vendor selling on Amazon's website is clearing them out for a song. It appears that price dominates in whatever calculation they're using to determine "value". A super cheap price boosts the CPU way up the chart no matter how slow (and discontinued!) it is.

Which makes it utterly unsurprising that Sandy Bridge chips fail to appear. They're the current standard-setters for value, by which I mean that AMD is (and has been for a few years now) stuck with being reactive in pricing. Whatever price Intel sets for a given performance level, AMD has to undercut a bit, while not being too much lower since AMD's costs per CPU are higher. (This is for two reasons: one, Intel's simply better at low cost volume production simply because they have in-house fabs and a lot more volume, and two, AMD's performance deficit means they must use more cores to compete in any given segment, which increases die size compared to Intel.) So, it's not too surprising to see AMD "first" in a chart which heavily favors the lowest price.

What's not clear is whether that has anything to do with "value". To me, "value" means: if I spend $X, how much performance do I get? Generally speaking, you'll find that $X buys you a given Intel CPU, and $X - ~5-10% buys you a somewhat slower AMD CPU (due to the reactive pricing I mention above), right up until AMD runs out of steam and can't compete at all any more (at somewhere around the $200 price point these days).

Re:Distraction.

[MikeBabcock]

I've never seen a modern PC that off-loads video chat encoding to the GPU. Mind citing one? Skype? MSN? ooVoo? etc.

Re:Distraction.

[MikeBabcock]

I think you'll find that's not even remotely true.

Here's a better chart too -- Tom's Hardware Gaming CPU for the money.

Under $100: AMD
Best at $115: AMD Phenom II X4 955
After $150, Intel starts being on the chart instead.

PS ...

CPUs priced over $220 offer rapidly diminishing returns when it comes to game performance. As such, we have a hard time recommending anything more expensive than the Core i5-2500K, especially since this multiplier-unlocked processor can be overclocked to great effect if more performance is desired. Even at stock clocks, it meets or beats the $1000 Core i7-990X Extreme Edition when it comes to gaming.

Overclocking Demonstration

[I.M.O.G.]

I also attended the event, and wrote up a more detailed account of the demonstration and word record result for the overclocking audience: http://www.overclockers.com/amd-fx-bulldozer-breaks-cpu-frequency-world-record/

Re:Overclocking Demonstration

[raymorphic]

Pretty "Cool" Bulldozer(processor)

And memory?

[Anonymous Coward]

Bottlenecks only become more noticeable.

No New Egg Warranty...but...

[David_Hart]

... the kit includes a free pair of thermo insulated mitts, just pay for shipping and processing... (grin)

Overclocking a what?

[girlintraining]

This is like supercharging a Pinto. Sure, it's the fastest a Pinto has ever gone, and maybe the highest RPM ever seen, but it's still a Pinto. AMD is two generations behind now on its fabrication facilities and chipset designs. Extreme overclocking is interesting, but hardly useful except as a PR stunt, unless they plan on selling liquid helium with every CPU purchase.

Re:Overclocking a what?

[Anonymous Coward]

Give away the processor, and sell the liquid helium. The gillette model all over again.

Re:Overclocking a what?

[Anonymous Coward]

AMD has been far ahead of Intel in chip design. They have been behind in core design. AMD had gotten rid of the FSB with a NUMA earlier, real quad cores with L3 cache earlier, and properly integrated graphics earlier (Intel's CPU performace goes down quick when the integrated GPU is used in SB).

Re:Overclocking a what?

[0123456]

AMD has been far ahead of Intel in chip design. They have been behind in core design. AMD had gotten rid of the FSB with a NUMA earlier, real quad cores with L3 cache earlier, and properly integrated graphics earlier (Intel's CPU performace goes down quick when the integrated GPU is used in SB).

And in the real world Intel has still been ahead of them for most uses since at least the Core-2.

As for this demo, good luck running your CPU at 1.9V for more than a few weeks.

Re:Overclocking a what?

[I.M.O.G.]

1.9V lifespan would be measured in minutes or hours, not days or weeks. At room temperature, 1.9V would damage the processor almost immediately, if not kill it.

In fancy demonstrations and testing like this, the main magic comes not from voltage, but from super conductivity within the processor at the cold temperatures. The cold also increases the lifespan of the chip when using exorbitant amounts of voltage. This doesn't matter to most people or maybe even the poster I'm replying to, but I'm mentioning it here for the general knowledge of anyone who may be interested in speaking accurately about this.

Ultimately though, its the cold that raises the frequency ceiling - sure the voltage helps, but with super-cooled temperatures alone the frequency ceiling is raised even at default voltage./p.

E-peen

[Dunbal]

Just like drag racers - exactly how practical are they to get to and from work? I want a reasonably fast reliable chip that will last me the 5 years or so till my next upgrade, not something I can run at 8GHz on liquid nitrogen for maybe an hour before the chip dies from thermal stress.

Re:E-peen

[timeOday]

Just like drag racers - exactly how practical are they to get to and from work?

It's not even that, since the record is not for processing speed - just clock speed! One hopes they at least ran some benchmarks, but the article doesn't say anything about it. So, this is more like "highest RPM for an internal combustion engine" or something like that. (Which, not coincidentally, is most easily done on a small-displacement engine with little torque).

Re:E-peen

[LWATCDR]

Umm... Just about any CPU will run for 5 plus years. Depending on what you are using your PC for just about any CPU you get from Intel or AMD will do that for.
And if this CPU will handle this level of abuse odds are that it will last for a very long time on your desktop system.

Re:E-peen

[sjames]

That's pretty much why the rated clock speeds are somewhat lower. There is a non-orthogonal relationship between overclockability and the ability to run reliably at the rated speed for a long time.

Mostly though, it's just a bit of a publicity stunt.

Re:E-peen

[Anonymous Coward]

And if this CPU will handle this level of abuse odds are that it will last for a very long time on your desktop system.

Who says the CPUs did "handle" this level of abuse? Working for an hour or two in an over-the-top demo doesn't mean you would want to take the chips used in the demo, clock them normally, and use them for another 5 years. At the voltages in question I guarantee you damage was done.

AMD and Intel both like to do these extreme overclocking stunts now and then. They don't mean anything, they're just stunts.

Protection

[suso]

Does anyone else think that for handling liquid nitrogen and liquid fucking helium that these "experts" seem to be throwing caution to the wind? Seems more like a rave than a science experiment.

Re:Protection

[Tim C]

One of the lab experiments in the first year of my physics degree involved using liquid nitrogen. If you're quick you can scoop a little out of the container with your fingers and splash it over the bench; it really isn't that dangerous unless you're a complete idiot (and no, sticking your bare fingers into it to splash it over the bench in this case does not quite constitute complete idiocy).

Re:Protection

[nschubach]

Even if it was dangerous... what a cool story to share. "Yeah, I was overclocking a new line of processors and I spilled this all over my arm."

Re:Protection

[suso]

That's possible via the Leidenfrost Effect, but not for long.

Re:Protection

[nschubach]

This makes me wonder why you'd use such cooling on hot processors then. You'd have to maintain a close relative temperature across the entire processor in order to prevent that vapor barrier.

Re:Protection

[wisty]

Liquid nitrogen vaporizes, forming a gaseous layer which protects exposed skin. You *can* put it in your mouth, and look like you are breathing steam, but you can't swallow it (as the gaseous layer gets forced aside causing you frostbite when the liquid presses against your internal linings, and it rapidly expands in your stomach, which makes you expand too, possibly fatally). See http://darwinawards.com/personal/personal2000-25.html

You can wash your face with it, but your hair / body hair *can* perforate the gaseous layer, resulting in localized frostbite, and hair loss. If you jumped into a pool of it, it would kill you, but *just don't do that*. Geez.

It's like the difference between a hot coal, and a pot of hot coffee. Drop the hot coal on your lap, and it will vaporize the top layer of your clothes, causing no real damage (and creating lots of comic relief for observers), unless you leave it there for a significant period of time. Drop a pot of boiling coffee on your lap, and you may no longer be able to reproduce. But people just assume that the hot coal is more dangerous, because it is hotter.

Re:Protection

[Kvasio]

Next on Fox News:
"My CPU overheated while I watched porn. Wanted to add some nitrogen and that's how I lost my John Thomas"

Re:Protection

[blueg3]

No. Liquid nitrogen is easy to use safely. It's common in undergrad-level labs, can be reasonably easily purchased (like dry ice) by just about anyone, and only requires minor safety procedures. Liquid helium is more dangerous, but is common in graduate-level labs and is not particularly hard to work with if you know what you're doing. It is really quite expensive, though, and a complete waste to use for cooling something that's producing ~100W of heat, since it doesn't have a very high heat capacity.

It's probably easier to injure yourself with home power tools than with liquid helium. It's just that fewer people have experience properly handling the latter.

Re:Protection

[MrHanky]

Yeah, if you get too excited during the 'hands-on time', you'll end up losing your willy.

Re:Protection

[mikael]

It's not that dangerous - the system operates just like a regular refrigerator. The cooling system is filled with helium gas at room-temperature. A compressor is used to compress the gas to high pressure - this causes heat to be emitted. As the pressure decreases, the helium liquefies before reaching the processor where it then heats up again and becomes gas again and the cycle is repeated.

It's not like some dude is standing on a wheeled office chair above the PC, with a flask of liquid helium in one hand, a funnel and some rubber tubing in the other trying to keep the end of the tube aligned with top of the CPU.

Re:Protection

[NatasRevol]

Is that part of the Republican debate?

Re:Protection

[dpilot]

I picked up this trick in an undergrad lab from a professor, though he didn't quote it by name. Pour LN2 out of the container onto the palm of your hand - it steams and falls out onto the floor. The key is to position your palm so that nothing collects there - everything rolls off - and not to do it for too long, because even though there is a vapor barrier that, it is chilling your hand.

It looks pretty neat though, and has a lot of wow factor to someone who doesn't understand that it can easily be done safely.

Re:Protection

[VolciMaster]

Does anyone else think that for handling liquid nitrogen and liquid fucking helium that these "experts" seem to be throwing caution to the wind? Seems more like a rave than a science experiment.

I was handling liquid nitrogen straight out of high school (16-17), and used liquid helium in a limited fashion at the same employer

Re:Protection

[Low Ranked Craig]

sounds more like something Tony Weiner would be up to these days...

Re:Protection

[tixxit]

I think they probably have a piece of metal between the processor and the liquid nitrogen. The metal on the cold side would be close to the liquid nitrogen's temperature and on the hot side it'd be closer to the procs temperature. You wouldn't get the effect, because of the gradient in the metal.

Re:Protection

[tenco]

As long as contact with the skin isn't long (unlike: get it on your clothes or in your shoes), you should be fine. I frequently handle liquid nitrogen in the lab and don't use special protection gear for it.

Re:Protection

[wgoodman]

If you start cold, the processor never gets hot enough. They don't let it heat up then start dumping it on.

Just like the FPU

[Viol8]

For the same reason that the FPU was first integrated into the 386DX. People forget that it used to be on a seperate chip (287). It would seem perverse to have a seperate chip for the FPU these days and I suspect in 5 or 10 years the same will be said for standalone GPUs.

Re:Just like the FPU

[paedobear]

486 (and it was only enabled for the DX) - the 386 series still needed a separate FPU. The difference between the 386 SX and DX was the size of the data bus.

Re:Just like the FPU

[nschubach]

I happened to grow up on a 386SX... I know that pain all too well.

Re:Just like the FPU

[tepples]

It would seem perverse to have a seperate chip for the FPU these days and I suspect in 5 or 10 years the same will be said for standalone GPUs.

Was there ever a competitive market for FPUs, or did only one FPU make and model work with each CPU? GPUs are meant to be replaced.

Re:Just like the FPU

[LWATCDR]

Ahh no.
The FPU was integrated in the 486DX you could get a cheap 486SX that didn't have an FPU because heck who needed one except for CAD users.
http://en.wikipedia.org/wiki/X87
For the history of the x87 family.
10 years? Naw I give it five max. Once APUs can play games at 1080p with all the eye candy almost no one will buy a separate GPU. They day that they can driver two 1080p displays at that level it will be all over.

Re:Just like the FPU

[Anonymous Coward]

For the same reason that the FPU was first integrated into the 386DX. People forget that it used to be on a seperate chip (287). It would seem perverse to have a seperate chip for the FPU these days and I suspect in 5 or 10 years the same will be said for standalone GPUs.

The 486 was the first to have an intergrated FPU. The 386 and the 486SX used a copressor.

Re:Just like the FPU

[Runaway1956]

Thank you - I was sorta scratching my head on that one. I had a 386 SX, then a 486 DX, and I was pretty sure that the FPU was separate on the 386.

Re:Just like the FPU

[_Shad0w_]

There was only one manufacturer at the time, really, Intel. Although Cyrix did start off as a match co-processor developer, come to think of it.

Re:Just like the FPU

[MadKeithV]

Was there ever a competitive market for FPUs, or did only one FPU make and model work with each CPU? GPUs are meant to be replaced.

Well, there was Weitek.

Re:Just like the FPU

[Anonymous Coward]

FPUs work in synchronous with the CPU. The program has to wait the result of the FPU instruction before going on with the program. Having an FPU outside of the chip increases the communication overhead and makes the system slower.

GPUs in contrast work asynchronous with the CPU. The CPU sends a batch of commands (today it is more of uploading a program to execute) to the GPU and goes on with its other tasks. The separate memory bus and modules of the GPU makes it possible that the GPU program executes without any performance impact on the CPU. This makes the system much faster. This advantage will be important for long. At least in the high-end gaming and parallel computation market.

Certainly it has disadvantages. The worst is that all textures (arrays of data) must be copied from CPU-RAM to GPU-RAM. But this movement is a linear transfer of huge chunks of data that can easily be optimized by hardware. Better than blocking the CPU with random access of bytes in the RAM.

Re:Just like the FPU

[Ferzerp]

I see this argument over and over, but it isn't analogous. The very first x86 integrated FPU was *faster* than the i387. Right now, the integrated GPU is an order of magnitude less capable than the top end discrete GPU.

Your comparison does not hold true.

Re:Just like the FPU

[UnknowingFool]

For average consumers though, discrete graphic cards are almost not necessary anymore. Sure gamers will always want more FPS with moar shaders but average consumers have far less stringent requirements. The next step I see is going to 4k but that won't be for a while.

Re:Just like the FPU

[asdf7890]

Early 486SX processors were DXes from batches where some had failed the FPU tests. The power and ground connections to the FPU part were burned so that part would not function, and the rest of the chip just took this as meaning the FPU was not present (they were still separate units linked like a 386 and 387 were, they just lived on the same bit of silicon instead of in separate packages - in the Pentium class chips onwards everything was more tightly integrated so just disabling the FPU units this way would not work).

Much to the annoyance of some people 487 chips that added the FPU to a system with a 486SX were actually full-blown 486DX chips internally - when installed the 486SX unit was completely turned off (if it wasn't surface mounted, you could remove it completely). The irritation was that the 487 was often sold cheaper than a 486DX even though they were internally exactly the same part. IIRC some people found that you could cut off the extra pin a 487 had so it would function as a 486DX, if you didn't mind voiding a warranty or two.

Re:Just like the FPU

[0123456]

It would seem perverse to have a seperate chip for the FPU these days and I suspect in 5 or 10 years the same will be said for standalone GPUs.

The FPU didn't use 300W.

Putting a high-end GPU onto a CPU would double the cost, require adding a huge memory bus to support the required bandwidth, make it insanely difficult to cool and force you to upgrade the CPU whenever the GPU got too slow for new games.

Re:Just like the FPU

[MikeBabcock]

And when Intel began integrating the math chips on all future processors, I was sad because the IIT math chips I'd had in my previous systems was much more capable than those integrated by Intel. See all about x87s here.

Re:Just like the FPU

[Kjella]

10 years? Naw I give it five max. Once APUs can play games at 1080p with all the eye candy almost no one will buy a separate GPU. They day that they can driver two 1080p displays at that level it will be all over.

1. Heat. If you have a hot CPU and a hot GPU, having both together is not such a good idea.
2. Upgrades. No more swapping graphics cards without swapping CPU (and if you're unlucky then mobo, memory etc. too)
3. If you look at the benchmarks for your favorite game, a discrete nVidia may be a better choice.

Don't get me wrong, for laptops, nettops and SoC chips this makes perfect sense but for a powerful desktop a discrete graphics card makes just as much sense. I don't think the discrete GPU is going away any time soon.

Re:Just like the FPU

[rwa2]

Poor you, sounds like you didn't have a turbo button.

Re:Just like the FPU

[0123456]

Once APUs can play games at 1080p with all the eye candy almost no one will buy a separate GPU.

Yeah, because no-one will find a use for all the extra power that a discrete GPU would give you.

The only reason you can play modern games on a PC with a low-end GPU with anything like the graphics you could get from a high-end GPU is that most games have been crippled to be playable on consoles with five-year-old GPUs.

Re:Just like the FPU

[Rockoon]

AMD made 80387's as well as their popular 40mhz 80386

Re:Just like the FPU

[julesh]

FPUs work in synchronous with the CPU. The program has to wait the result of the FPU instruction before going on with the program.

Err -- no it doesn't. The CPU can continue executing integer instructions, and doesn't wait for the FPU until it issues another floating point instruction (or FWAIT, which doesn't compute any floating point results but synchronizes the two processors for the purpose of handling exceptions, etc.).

I think more relevant is that the FPU is reliant on the CPU providing it with a constant stream of instructions in order to not sit idle, whereas GPUs have their own instruction memory that they can read independently of the CPU, so can carry on without its attention for long periods of time.

Re:Just like the FPU

[Belial6]

Some people will still want faster graphics, but the OP's point was that the vast majority of people won't want to spend the extra money on it. There really is a "good enough for 90% of the people". Once 90% of the people don't see a benifit in paying the added costs of a seperate GPU, the cost of seperate GPU systems will increase, and fewer stores will carry systems that have it. This will accelerate the process. At the same time, the integrated graphics will continue to improve.

This isn't a "Nobody needs more than 640k" situation.

Re:Just like the FPU

[Fjandr]

Intel was contracting out to other companies fabs at least from the days of the 8088 (that's the oldest Intel CPU I am actually in possession of).

Intel's main rift with AMD came about as AMD started manufacturing compatible CPUs using the x386 architecture after Intel declined to renew their manufacturing contract. IIRC, AMD at one time manufactured the bulk of Intel's CPUs.

Re:Just like the FPU

[LWATCDR]

Wow it is amazing some times what people read.
"Yeah, because no-one will find a use for all the extra power that a discrete GPU would give you."
I never said that at all. I said almost no one will buy a separate GPU. Notice that the word "almost" which means that some people still will.
This issue here are economies of scale. Even today the majority of systems probably use integrated graphics. Most systems sold today are notebooks and most notebooks use IG. Throw in all the corporate desktops, school machines, and the average home users and I am willing to bet that most users already are using IG today.
Now as you said most modern games are console ports. Do you see that changing? I sure don't anytime soon because that is where the money is right now. The game makers will all jump for joy once the average desktop and notebook can do good HD graphics. The can then have a good sized market on the PC to target.
The end result will be as IG gets better and better the market for desecrate GPUs will get smaller and smaller. As the market gets smaller and smaller the costs will go up faster and faster for less and less gain. You will have the same situation that you have with audio today where a few people spend the money for a high end sound card but most people are very happy with the sound buit into the motherboard.

Re:Just like the FPU

[tyrione]

Ahh no. The FPU was integrated in the 486DX you could get a cheap 486SX that didn't have an FPU because heck who needed one except for CAD users. http://en.wikipedia.org/wiki/X87 For the history of the x87 family. 10 years? Naw I give it five max. Once APUs can play games at 1080p with all the eye candy almost no one will buy a separate GPU. They day that they can driver two 1080p displays at that level it will be all over.

It won't be over. It will be a new form factor for GPGPUs on Systems. You'll see a cluster of scaled down GPGPUs in a daughter card for PCI-E 3.x and the system will be ripping through all sorts of mundane tasks now offloaded to that card.

Re:Another use for liquid nitrogen right now...

[UnknowingFool]

Well considering that the highs have been in the 110F range with heat index, 100F is what Texans should call a "cold front". Remember in Texas there are only 2 seasons: Hot and Not Hot. ;)

Hothardware ?

[Lennie]

Shouldn't this be really, really cool hardware instead if they are using liquid-helium and all that ?

Providing additional utility ...

[Rambo Tribble]

... these setups can replace several small, kitchen appliances.

Plastic deformation of the motherboard

[Anonymous Coward]

From an article: "The PCIe section of the motherboard curling from extreme temperatures" (complete with a picture)

Aha, so this is why motherboards are generally mounted to the case with lots of screws?

Re:Two links to one article

[Kvasio]

check here: http://en.wikipedia.org/wiki/Point_(geometry)

Manufacturer should be disqualed from OC records

[Anonymous Coward]

Overall clocking a CPU the fastest (that we know of) (for right now): cool. Congratulations, dudes. You did something awesome and we all recognize how nifty that is.

Over-clocking a CPU the most: AMD employees ought to be disqualified, at least when they're using AMD chips. For an overclocking record, you shouldn't be connected to the organization that rated the chip in the first place. If I were to make my own 3-GHz-worthy Pentium 4 then I could rate it as a 100 MHz processor and I would win an overclocking record too. How do we really know these aren't 6 GHz Bulldozer chips to begin with, but under-rated, other than AMD's word that they're not? Not that I seriously believe shenanigans are happening, but the potential for it makes it not cool. AMD employees going for an OC record should have to use an Intel chip (or ARM or whatever, but not AMD) and vice-versa.

2.4Ghz in the 90's

[bmwEnthusiast]

Big deal, I had a cordless phone that was 2.4GHZ in the 90's. :)

Intel GMA (Graphics My ...)

[tepples]

Once APUs can play games at 1080p with all the eye candy

Given Intel's history of lagging behind NVIDIA and AMD in GPU capability, I don't see Intel GMA becoming able to "play games at 1080p with all the eye candy" any time soon. Does Sandy Bridge change this? Or were you talking about AMD CPUs with an AMD GPU on-die?

Re:Intel GMA (Graphics My ...)

[gman003]

Sandy Bridge was (from what I've heard and read) a major improvement, but not really a game-changing one. Not enough to game at full settings at 1080p, but enough to watch videos at 1080p.

Obviously

[DinDaddy]

You can break almost anything with a bulldozer. Records are especially easy.

Re:Obviously

[Anonymous Coward]

There's a barn over there; please go stand behind it. I'll be right back, I just left something in the house.

Letdown

[Anonymous Coward]

I was hoping they actually overclocked a bulldozer.

It's been a long time..

[Peter Cooper]

But.. imagine a Beowulf cluster of these!

Other benchmarks

[Aggrajag]

I am still waiting for actual benchmarks.

Re:Goatsehertz record.

[GameboyRMH]

I at least lol'd at your creativity.

Also Goatse is on a Russian domain now? In Soviet Russia, Goatse disgusted by YOU!

Re:Manufacturer should be disqualed from OC record

[EmagGeek]

If you were Intel or ARM, would you really put any effort into winning an OC contest with an AMD chip?

Hell no. You'd say "Well, we got 100MHz out of it, but couldn't overclock it any further because it was getting really unstable."

After all, you're Intel, and have zero motivation to win an OC contest with your chief competitor's chips, and have every motivation to make them look bad.

The real question is...

[Anonymous Coward]

Why are they designing chips with wires twice as short as what they really need for the desired ~4 GHz clock frequency? How much better would their chip be without that unnecessary constraint on circuit layout?

Cheating.

[screwdriver]

Isn't it cheating if the manufacture of a CPU (you know, the people who stamp the magic clock speed figure on the chip) overclock their own CPU? I mean, they could easily under-rate the chip and then "overclock" the hell out of it.

Re:Cheating.

[cmburns69]

If I understand the summary correctly, it's not the record for the most overclocking-- rather, it's the record for the fastest clock speed of a computer.

8+ GHZ is no simple feat.

Re:Cheating.

[Targon]

The only thing that would be cheating is if the manufacturer claims it will run properly at a given speed and then when a customer does it, it does not. Every company will ALWAYS provide some extra room for overclocking, but what level of it is safe? AMD does not CLAIM that a retail part will run at those speeds, but the fact that it can be done with extreme cooling shows the design is stable.

Record broken by Bulldozer?

[Anonymous Coward]

Maybe they shouldn't store such important information on stone tables?

It's all about memory speed!

[BlueCoder]

Why haven't we learned our lessons yet? Processors are bottlenecked by memory speed. We were only able to get processors to go faster in the past by large memory caches and superscaller design. It's a dead end.

We need new memory technologies. We need new memory and memory buses that are able to run faster than processors. Preferably an optical technology not subject to radio and inductive interference. Solve this and systems will seem like 10 thz superscaller designs.

Then we need to reverse past trends and scale down superscaller technology and divert the transistors to more cores and power efficiency. It's like there is no innovation going on anymore, no design; it's all brute force through shrinking transistors.

Re:It's all about memory speed!

[Mr Z]

Don't you think those of us working on these chips haven't thought of that? This problem was recognized back in the 1940s , for goodness sake. Quote:

Ideally one would desire an indefinitely large memory capacity such that any particular . . . word would be immediately available. . . . We are . . . forced to recognize the possibility of constructing a hierarchy of memories, each of which has greater capacity than the preceding but which is less quickly accessible.

--A. W. Burks, H. H. Goldstine, and J. von Neumann

Preliminary Discussion of the Logical Design of an Electronic Computing Instrument (1946)

There's a limit to how fast even the processor's own registers can go, such that some designs include "local register files" to individual ALUs, because the main processor registers are too slow.

Re:It's all about memory speed!

[Calos]

This is baiting, but I'll reply...

Shrinking transistors isn't brute force, per se. It makes things faster, and if done correctly (read: not like AMD's 32nm process, which consumes just as much power at load as older generations), it reduces power. And it's not just shrinking. Intel's move to FINFETs is a great example of that. I fear people outside of the industry don't grasp what all goes into this fundamental change.

On the memory front, things are changing... things have to. Memory (DRAM) can't be shrunk just by transistor shrinks. Research is being done on things like spintronics and phase change. These things will get there in time. And with shrinking transistors you can get shrinking SRAM, which is directly integrated on-chip, meaning high-speed and no interconnects. SRAM is one of the faster memory types around, and you can't beat not having any real interconnects. Building bigger caches, which you seem to deride, through transistor shrinks, which you deride, allows using more and more of the fastest memory cheaply available.

More than that, there's a ton of work always done in memory prediction on the CPU. Predicting what is coming next, pre-fetching it before its needed. Tons of work is still being done in this area, and it lessens the impact of slower memories.

Interconnects? Optical is probably a long way out, from what I've seen in literature. Could be some company is well ahead of published research, though. Nonetheless, 3D integration will probably happen first... and effectively eliminate a lot of these slow buses. Companies have been talking about this for ages, IBM is throwing R&D at it, rumors suggest TSMC wants to manufacture it by end of next year, last I saw.

So, sorry... I disagree, and can't help but wonder if you did much looking at all before complaining.

Yup , my mistake

[Viol8]

I shouldn't rely on memory. Still 386 or 486 the point still stands.

You miss the point.

[Chirs]

I bought a new laptop this past January. I *could* have bought a blazing fast i7 with discrete graphics but I settled for an i3 with integrated. It was good enough for my needs, cost less, and the battery lasts longer.

Sure, I could use blazing fast speeds, but I don't *need* them. And I don't want them enough to pay the extra costs in dollars and energy consumption.

Re:Manufacturer should be disqualed from OC record

[ArcherB]

Overall clocking a CPU the fastest (that we know of) (for right now): cool. Congratulations, dudes. You did something awesome and we all recognize how nifty that is.

Over-clocking a CPU the most: AMD employees ought to be disqualified, at least when they're using AMD chips. For an overclocking record, you shouldn't be connected to the organization that rated the chip in the first place. If I were to make my own 3-GHz-worthy Pentium 4 then I could rate it as a 100 MHz processor and I would win an overclocking record too. How do we really know these aren't 6 GHz Bulldozer chips to begin with, but under-rated, other than AMD's word that they're not? Not that I seriously believe shenanigans are happening, but the potential for it makes it not cool. AMD employees going for an OC record should have to use an Intel chip (or ARM or whatever, but not AMD) and vice-versa.

This isn't necessarily a record for overclocking. This is a record for CLOCKING. It doesn't matter if the chip was rated at 9GHz or 9MHz. The fact is that no chip, regardless of its initial rating has ever reached a clock speed that high.

Re:Manufacturer should be disqualed from OC record

[RalphSleigh]

Pretty sure its an absolute Ghz record, not one relative to the chips nominal speed.

more impressed with efficiency over speed

[schlachter]

If we're going to shoot for extremes, I'd be more impressed with record efficiency over record speed. It could be far more practical.

Liquid He in Texas?

[damn_registrars]

I figured it would have been impossible to keep Helium from going up in flames in Texas (along with the rest of the state).

Re:Liquid He in Texas?

[burning-toast]

Liquid helium is impossible to get to go up in flames as it is inert and non-flammable. You are thinking of liquid hydrogen. OTOH, your point about Texas going up in flames still stands.

Re:Liquid He in Texas?

[damn_registrars]

Liquid helium is impossible to get to go up in flames as it is inert and non-flammable. You are thinking of liquid hydrogen.

Actually I rather assumed that people knew Helium to indeed be non-combustible, and would recognize my post as a joke. Or, they would realize I was making fun of Texas and tag my post flamebait.

Whatever, I have karma - and a large southern state - to burn.

Re:Manufacturer should be disqualed from OC record

[Anonymous Coward]

(Original GP AC here)

If you were Intel or ARM, would you really put any effort into winning an OC contest with an AMD chip?

So you bow out of OC contests, since there's no good way to enter them.

I'm not putting overclocking down, but there's something "amateur" about it (and I mean that in the nicest and hackeriest way). You're making the chip do more than The Man in the suit says it's rated to do. If you are The Man in the suit, then it doesn't make sense.

It's not unusual for pros to stay out of amateur contests or record-breaking. Not because it's unfair for pros to go up against amateurs (that would be to imply they're somehow "better" and that's totally not what I mean), but because the pros have conflicts of interest.

BTW, some people have corrected me that it's a clocking record, not an overclocking record. Fair enough, Slashdot just got the summary/title wrong.

Jung disagrees

[CSMoran]

n/t

Pretty cool...

[Anonymous Coward]

Windows detected this higher speed anomaly and compensated to enhance the user experience.
Boot times and MS word performed as though running on a PIII at 600 Mhz. Not too shabby.

GMA: Graphics My ...

[tepples]

For average consumers though, discrete graphic cards are almost not necessary anymore.

Unless their PC has an Intel CPU. Or has Sandy Bridge yet caught up with the six-year-old Radeon X1900 in the Xbox 360 or the GeForce 7800 in the PS3?

Re:GMA: Graphics My ...

[UnknowingFool]

Did you read anything I wrote above or are you one of those people that believes a GPU needs to be so powerful that you can cook with it.

Sure gamers will always want more FPS with moar shaders but average consumers have far less stringent requirements.

I don't believe average consumers need a 7800 or the x1900 if they are watching kitten videos on Youtube. For casual games, the Intel HD 3000 is usually enough. And I hate to break it to you but they beat AMD's 890GX.

Final proof

[Anonymous Coward]

That AMD is better than Intel.

Still slow?

[Anonymous Coward]

Even at 8.5GHz, can it touch a stock-clocked i7-2600K in real-world application performance?

Re:Goatsehertz record.

[Vegemeister]

In Soviet Russia, goatse hosted at goatse.su.

All hype and no availability?

[rrohbeck]

Just give me the damn chips instead of benchmarks. I'm planning a new rig and I'm ready to place a big order with Newegg. Right now it's i7 based. I've been delaying, waiting for Bulldozer, but I won't wait forever.

Re:All hype and no availability?

[slater.jay]

I already gave up and bought an i5.

This begs the question...

[Anonymous Coward]

...But does it run Crysis?

Rule of the shorter term

[tepples]

Copyright protection in US under Berne Convention is only valid as long as the work is under copyright protection in its country of origin

Citation needed that the United States recognizes the rule of the shorter term for European works. Wikimedia Commons, for example, doesn't seem to think so.

What does it mean Clock for clock?

[xiando]

1 MHz on one CPU is not equal to 1 MHz on another CPU. You can not say "AMD chip X is 3.0 GHz and Intel chip Y is 2.9 GHz so chip X is faster". Intel chips generally give you more clock for clock. Real-world benchmarks of a CPU running 8.429 GHz would be interesting.

Celeron MK2?

[triffid_98]

I'm fairly certain they had a Celeron up to 8.1Ghz several years ago. If this is actually competitive clock for clock with current offerings than I'll be suitably impressed, but clock speed by itself isn't good for anything except frying CPU traces via electron migration and bragging rights.

Fool.

[Anonymous Coward]

Just buy a 990fx mobo (gigabyte has good cheap ones which allow you 1800 mhz or 2 ghz ram speed and also 16x * 2 crossfire) and plug a low end amd cpu (any will do, even semprons for $35) in it and you can wait as much as you want. that's what i did. but i plugged a phenom ii 965 black edition in it, and now i am not sure whether i need to get a bulldozer anytime soon. phenom 965 works great by itself.

Re:Fool.

[rrohbeck]

1. AM3+ will only be good for the first Bulldozer generation (Zambazi.) The next generation will use the FM2 socket.
2. Right now an AM3+ mobo plus a Phenom II isn't all that much cheaper than an i5 with a LGA1155 mobo.
So - unless Bulldozer has a significantly better price/performance ratio and its raw performance is in the vicinity of i7, an LGA1155 is the better choice.
I'm waiting for Zambezi benchmarks.

THAT :

[Anonymous Coward]

With Cedar Mill chips half a decade old breaking records as recently as last month, and Bulldozer breaking the World Record before it hits market – without any doubt this is only the beginning for Bulldozer. AMD will be the first to hit 8.5GHz, and it will happen before long, the only question remaining is which overclocker will hit it first!?

you can bet that a lot of overclocker groups have already started planning the rigs they will break 8.5GHz even as of this moment. it probably wont take even a month after these chips start getting shipped.

wrong

[Anonymous Coward]

it was another chip. (cedar mill based) and it is a decade old. it only broke record recently. this chip has broken it before it was released. so, go figure.

No, no, no, no...

[Kaenneth]

Does it run !!Dwarf Fortress!!

Overclocking

[Anonymous Coward]

It's just a way for chip makers to make sure the warranty is voided.