AMD Shows Upcoming Phenom II CPU At 6.0 GHz+ 159
Vigile writes "Today during a press briefing at AMD's offices in Austin, TX the company showed off some upcoming technology that should be available sometime early in 2009. What was most impressive was the overclocked speeds of the pending Phenom II X4 45nm processors. On air cooling AMD showed the quad-core CPU running at nearly 4.0 GHz while with much more extreme liquid nitrogen cooling help the same CPU reached over 6.0 GHz! It looks like AMD's newest processor might finally once again compete with the best from Intel, including its recent Core i7 CPUs."
Overclocking BS (Score:3, Informative)
For example, here is a video from 2006 where a Pentium 4 processor is overclocked to 5 GHz. [youtube.com]
So no, it doesn't look like "AMD's newest processor might finally once again compete with the best from Intel."
Misleading title (Score:4, Informative)
If you need liquid nitrogen to boost it to 6 GHz, it's not all that interesting. Nehalem 2.66 GHz offering has also been shown to overclock to 4 GHz on air cooling, and some people have got the 3.2 GHz offering up to 4.5 GHz on air. On GHz they're roughly the same, possibly with a slight Intel edge.
I thought both companies were ditching the GHz war and fighting for actual performance supremacy? What's with the silly "my GHz is bigger than yours" competition? Do we have PPW numbers, or just press releases that mean nothing?
Re:Overclocking BS (Score:5, Informative)
You can't get a Core 2 CPU to run at 5Ghz no matter how hard you try.
Given that the Nehalem is reaching the same speeds [theinquirer.net] or higher [fudzilla.com] on air-cooling, I wouldn't be surprised if Intel could match 6 GHz under liquid nitrogen cooled conditions.
Re:first question.. (Score:3, Informative)
what's the power rating for this thing at 4 ghz? 250 watts?
If you actually read the article you would have seen "AMD could theoretically have a 3.4 to 3.6 GHz processor at moderate TDP levels (think 125 watts)"
Re:Is GHz even important? (Score:1, Informative)
GHZ is actually very important. Given that all else remains the same, a 10% increase in clock speed is not greatly different from a 10% improvement in performance in CPU bound applications. Comparing GHZ across different designs is a rather bad idea, but that is not all that is going on here.
Re:first question.. (Score:5, Informative)
That's for the current generation Phenoms. You likely want this article [wikipedia.org], which covers the Phenom 2 procs.
TDP spec at 3.0ghz is 125W, so don't think he's exaggerating that much. I'd guesstimate 150-200W at 4ghz.
Basically (Score:3, Informative)
GHz is brought up when your chosen platform is on top. If you aren't, it's downplayed.
For example the original "MHz myth" was started by Mac fans. When they first went PPC, Apple had a large lag behind Intel in MHz. Well, the Mac fans were all excited about this new architecture and kept talking about how PPC has a positive second derivative of MHz and x86 had a negative one and so on and so forth. They were all excited about how they'd be ahead in MHz in a few years and basically equated MHz to performance.
Well that didn't come to pass. PPC didn't scale up in MHz fast and x86 did. So all of a sudden they started whining about the "MHz myth" and saying that it didn't matter, performance did. When their platform wasn't going to be on top it changed from important to worthless.
Same shit here. When Intel had the high GHz chips, AMD heads were up on the fact that AMDs did more per clock. Now if AMD has the high GHz chips, they'll be touting that as being the measure of awesomeness.
Me, I'll just keep buying what does the job best, forget the clock speed.
Re:Misleading title (Score:5, Informative)
The Phenom seem to perform somewhat better under linux, or at least gcc produces better code for them than it does for intel chips...
(note, this is based on 64bit gentoo, gcc 4.3.0 compiled with -O2 and appropriate cpu type setting on a 2.3ghz phenom 9500 and 2.4ghz Q6600)
Re:Overclocking BS (Score:5, Informative)
Given that the Nehalem is reaching the same speeds or higher on air-cooling, I wouldn't be surprised if Intel could match 6 GHz under liquid nitrogen cooled conditions.
Here [hexus.net] is an example of Core i7 at 5.2 GHz on LN2
Re:Basically (Score:5, Informative)
For games, doesn't much matter. Get a dual core chip that's reasonable and you should be fine. Games do use the CPU but the GPU is by far more important. You can get a quad if you really want but at this time very few can use it at all, and those that can don't tend to be that efficient. A good dual core from the mid range area from either manufacturer should work well.
65 vs 45 nm isn't that important except in terms of energy usage. The 45nm chips are going to use less power for equal performance. However this again isn't a huge deal since the GPU is likely to be the big drain in the system.
Cache isn't all that big a deal. Again, just get whatever the midrange is. Games aren't an area where cache seems to make a large performance difference.
More or less, while these things can make a difference, they don't make enough to justify that much worry or money. You will probably find that a $250 processor works pretty much as good as a $1000 processor, whereas a $300 graphics card is going to be 50% faster than a $150 graphics card. Thus it is clear where your money should go.
Personally I have a Core 2 Duo 2.66GHz 65nm chip and it works just fine on all the games I've thrown at it. In general, when games are limited it isn't the CPU it is either the refresh rate of the monitor (it isn't useful to go above that and as with all LCDs mine is set at 60Hz) or the GPU. Now keep in mind the GPU I have is a GeForce GTX 280. So even a GPU that heavy hitting doesn't really seem to need more CPU, for all the games I've messed with.
If I were to build a gaming system today my strategy would be as such:
--Get a midrange CPU. Something probably not more than $300, but not less than $150. Maybe a quad core since I also do audio work, but I'd be looking more at dual cores. I'd make sure it supports DDR2 RAM, since DDR3 is currently too pricey to justify the small gain.
--Get 4GB of RAM. It's cheap, why not.
--Get a nice big drive since bigger drives are faster and games are not getting any smaller.
--Get a video card such that I can afford to get a new one of the same price once every 12-18 months or so.
That last one is key: Your video card is important to games, and it gets outdated real fast. You can't buy one that won't, because new technology comes out all the time. You can drop $2000 on an insane multi-card setup, and it'll still be outdated soon. So, the right answer is to buy less card, more often. I say make it a yearly target. You aren't necessarily going to buy that often, but that's a good target to make sure your price is realistic and you really don't want to buy more often than that. So whatever you can afford per year, get that. Then, when the next worthwhile upgrade in that price range comes along, get it.
That's what I did. Prior to my 280, I had an 8800. They both cost me about $400. I can afford to spend that every year (I spend a lot on my computer, it's important to me). In that case, it was more like 16-18 months, which is fine. You keep your card until there's a new one worthwhile and/or you find a game that doesn't run well, but you are ready to upgrade yearly. Used to be I couldn't afford so much, so I used more midrange cards. I had a GeForce 3 Ti 200 back in the day. Wasn't top of the line, but I could afford to get it, and then to replace it next year if needed.
So get a midrange CPU, plenty of RAM, and a video card that you can upgrade and you should be fine. CPUs have pretty good life these days. It's videocards that are obselete all the time. Good news is that videocards can be gotten for reasonable prices. For example an ATi 4850 will run you about $150, less after rebate. However it is enough to run any game out there at high detail at a reasonable rez. My bet is it lasts more than a year, but at $150, it isn't unreasonable to replace next year if you need.
Re:I've stuck with AMD (Score:5, Informative)
Intel handled (I think they still do) all IPC (Inter-Processor Communication) through the FSB. Which is also the ram bus, and so runs at ram speed. They even did this for inter-core communication, completely screwing any attempts to scale a single intel system to lots of cores and have it still run well.
AMD have 3 independent buses, an inter-core comms for multi-core cpus (runs at cpu clock speed), hypertransport for inter-cpu comms and device comms (runs at multiple GHz independent of CPU speed), and an independent ram bus (runs at ram speed, obviously). This means that an AMD system gains real performance pretty linearly with the number of cores and cpus, and an Intel one didn't.
Intel countered by massively increasing their ram speed, countering the FSB bottleneck for smaller (2 or 4 cores) systems, and by making their cpus capable of more instructions per clock than AMD cpus (a real surprise when it happened), giving them great single-threaded performance. AMD couldn't match the performance of the most powerful Intel Core 2 cpus, so went for energy efficiency in a big way, and generally tried to undercut (instead of outperform) Intel at every turn. AMD's cache architecture was better too, with data not duplicated in all levels of the cache, so AMD cpus effectively had 10% more cache compared to Intel cpus. Intel countered by adding lots more cache to their cpus. AMD also went for forward and backward compatibility in a big way, a BIOS update (and sometimes not even that) is all that is needed to make the oldest socket 939 boards work with the newest AMD cpus. You lose out on a few features (e.g. faster HT and ram), but it makes upgrading an AMD machine much cheaper.
This leaves us with the situation where AMD cpus are great for highly-communicating parallel operations, and are great in clusters and datacenters due to having higher performance per watt (so they cost less to run and need less cooling). They also make for cheaper desktop systems both to build and to run, important if you're on a budget. Intel cpus are great for ram performance, and high-speed single-threaded ops, important if you are building a super-powerful gaming rig. Intel's pushing of their on-board graphics chipsets has also caused Intel cpus to end up in a lot of pre-built machines.
Though to be honest, You don't need a cpu costing more than £60 to play anything released recently at full speed, and AMD is incredibly competitive at those prices (e.g. my AMD X2 5600). The real expense is in graphics these days, though my 2-generation-old nVidia 8800 (rev 1) GTS 320MB hasn't struggled on anything I've bought recently, even on high settings...
Looks like the performance race might be slowing, unless someone comes up with a cheap, working holographic projector :)
Re:Overclocking BS (Score:3, Informative)
GHz does not equal performance. Which is why a 2GHz Core 2 CPU can beat out a 3.6GHz Pentium 4.
Please don't bring back the myth.
Re:Basically (Score:3, Informative)
Oh, no. USB3 will be backwards and forwards compatible. You can run USB3 on USB2 devices and vice versa, but the speed difference is 20x faster so it's a big deal (way faster than firewire). AKA 600MB/s. Meaning that you could tune HD tv and use ridiculously fast flash drives/it could be straight competition to Esata as well.
Gaming processor = look at techreport's system builder guide on the front page. They show justification for why they choose something and newegg links to the whole system and is updated monthly I believe, so it's pretty easy to compare. You'd be amazed how cheap the stuff is sometimes :) In my personal opinion, phenoms are dirt cheap for a quadcore and getting a ati/amd 4850 can get you some monster performance real cheap ($170 for a vcard that very competitive right now that you could drop in another for sli at a later time to run as fast as the baseline fastest card on the market).
Re:Basically (Score:2, Informative)
I wrote this in response to another, and decided against posting it (since the GP specifically asked about gaming, but I missed that), but then decided to use it in response to your post. It expands on your points.
Addressing the GP's question: It depends on your application/use-case. A database server can do fine with a relatively slow multi-core CPU, as long as it has fast enough IO and a LOT of RAM. A scientific computing/numerical analysis machine will likely need the fastest FPUs it can find -- hence the popularity of the PowerPC architecture in that field. Depending on the domain of scientific computing (discrete data mining or other discrete mathematics problems), something like a Sun UltraSPARC T2 could potentially do very well. These sorts of machines often have a LOT of RAM too, since hitting swap can turn a 10 hour long run into a year long run. Gamers need fast FPUs, usually provided by the graphics card.
The average user will probably benefit most from a quick IO subsystem combined with 4GB of RAM. Fast IO will allow your applications to be loaded quickly, and lots of RAM will help keep your applications and data cached, so you won't have to hit the disk for them again. (Which means the second time you start an application will be several orders of magnitude faster) Fast processors are nice, but only if you actually do processing. There's no point in keeping a 4.0 GHz Core Quad Extreme idle, or waiting on disk or user input, except for bragging rights.
I wouldn't worry about things like the L2 cache much as long as you get a modern processor or are doing serious number crunching. A cache is only useful if it gets "hit" -- if the value the processor is looking for is already stored in cache. A cache miss means going to RAM or disk (or reprocessing) for the value. A MB is enough to store the most used parts of your OS's kernel. Scientific computing applications are often designed to fit in the L1 or L2 cache -- if the main loop is going to run a hundred trillion times, it's better to keep its instructions in the cache so you don't have to go to RAM for them, wasting 2-4 cycles each time. Also, if the main loop is bigger than a cache, you will ALWAYS have to go to the next layer of memory, either from L1 to L2, or from L2 to L3 or RAM.
Other processor subsystems are similar -- you would already know what you needed if you "really" needed it.
If you want to compute faster, figure out which component/subsystem is slowing you down. And look for specific recommendations about that component/subsystem. The rest of your system will fall into place around it, at least as fast as your current one, since everything has gotten incrementally faster. The adage regarding premature optimization is true of both software and hardware.