AMD Packs Six-Core Opteron Inside 40 Watts

adeelarshad82 writes "Advanced Micro Devices has launched a low-power version of its six-core Opteron processor in time for VMworld, a key virtualization show that opens on Monday. The six-core AMD Opteron EE consumes 40 watts, and is designed for 2P servers, among the most popular in the virtualized server space."

Comments

Hardware

[Tubal-Cain]

The six-core AMD Opteron EE...is designed for 2P servers...

All I really want to know is: can you install it in a toaster?

Re:Hardware

[eln]

Netapp already has boxes that run on Opterons, so probably, but you'd need one of their SEs and a support contract to do it.

Re:Hardware

[oldspewey]

If you want a toaster, you want Netburst.

Re:Hardware

[Locke2005]

can you install it in a toaster? Yes, but at only 40 watts, it will take forever to brown your bread!

Re:Hardware

[maino82]

It's like the easy bake oven of processors. It just leaves your cupcakes gooey and soggy.

Re:Hardware

[FudRucker]

whatever you install it in, it becomes the toaster

Re:Hardware

[MiniMike]

No, it is not compatible with Socket 478. That is what you were asking, right?

Re:Hardware

[MonsterTrimble]

Actually, I've been wondering how to heat my food dehydrator. Toaster element! Of course!!!

Re:Hardware

[pla]

With AMD's reputation for producing hot-running processors

What reputation? Since the days of the original Thunderbird core (which still ran cooler than comparable P4s, though admittedly didn't have meltdown prevention circuitry), AMD has consistently given Intel a run for their money in that regard.

Now, the Atom has finally brought Intel back to the realm of "reasonable", but it doesn't seriously compete with AMD, it competes with VIA (and poorly at that - The Nano blows the Atom away, clock for clock and Watt for Watt).

Don't get me wrong, Intel has certainly regained my respect when it comes to performance, but to call AMD the toaster requires ignoring the past 10 years.

Re:Hardware

[commodore64_love]

>>>With AMD's reputation for producing hot-running processors ...

Don't you mean Intel? After all their early 90s Pentiums were the first CPUs to spark the "you could fry an egg" jokes. And the Pentium 4 sitting in my computer is a major power hog (~90 watts), and it's just a single core.

Anyway 40 watts for six processors isn't really that bad. About 7 watts each.

Re:Hardware

[K. S. Kyosuke]

can you install it in a toaster? Yes, but at only 40 watts, it will take forever to brown your bread!

He only needs to replace NetBSD in his toaster by Vista and to overclock it slightly.

Re:Hardware

[TheRaven64]

The K6 series were very hot compared to Intel equivalents. The reason the reputation persisted is probably due to the fact that they didn't add themal throttling on-chip until a lot later than Intel. The P4 was much hotter than a t-bird, but if it overheated it would throttle, while the Athlon would just catch fire. I had quite a few t-birds burn out due to the stock fan not being adequate. We had the opposite problem with our cluster; the P4s were throttling due to uneven cooling, so nodes were all running at different speeds and job scheduling got messed up.

Re:Hardware

[sjames]

Indeed. It's been a very long time since AMD has been the hotter running CPU. It was Intel that introduced us to heatsinks that could hurt you if you dropped them on your toes.

It's hard to believe that at one time CPUs didn't have heat sinks at all.

Re:Hardware

[hairyfeet]

They also couldn't take the abuse that the Intel chips could. At the last shop I worked the boss had what he called "dead buckets" where we would chunk dead parts-one for CPUs, one for RAM, etc, don't ask me why as Doug never did give me a straight answer on that one. Anyway in the dead CPU bucket it was almost entirely AMD chips. It had a total of two Intels, one of which got surged during a lightning storm, the other surged on a blown PSU. We found if the AMDs fan clogged or failed they would often blow.

So while I haven't tried to abuse a newer AMD I still build Intel chips for the SMBs that are doing construction or other jobs where I know the box is gonna get seriously funky. I have switched to AMD for my home builds, as the bang for the buck just can't be beat (just built a REALLY nice quad for a customer that only set him back $700) but on those jobs where I know the box will be subjected to construction grime, low ventilation, etc I stick with Intel. After all it is MY ass that has to replace it if it blows during the warranty period. But I can understand why someone who might have been "burnt" by AMD in the past might be a little gunshy of them now.

Re:Hardware

[morgan_greywolf]

Wow. Guess some AMD fanboys just can't take a joke...

But -- the Thunderbirds (I've owned several), had big problems with the stock AMD fans (which were crap). That basically created the market for companies like Coolermaster, Zalman, and Thermaltake to create aftermarket fans for the Athlons.

Re:Hardware

[PitaBred]

Why should your job scheduling be at all dependent on how fast things finished? That seems like a horrible assumption to make when you're talking about a network of computers. All kinds of things could cause slowdown, and even failure.

Re:Hardware

[evilviper]

What reputation? Since the days of the original Thunderbird core (which still ran cooler than comparable P4s, though admittedly didn't have meltdown prevention circuitry), AMD has consistently given Intel a run for their money in that regard.

AMD was extremely sloppy on power management before the K8/Opteron days.

See my old /. Journal on the subject: http://slashdot.org/~evilviper/journal/70512

In short, while the maximum power of AMD CPUs was about the same as their P-III equivalents, AMD chips (Thunderbird and Athlon XP) would run at their maximum power ALL THE TIME, even when there was NOTHING to do.

This didn't get resolved until the very end of the Socket-A days, when AMD finally REQUIRED all motherboards be tested to work with the S2K bus disconnect feature of AMD CPUs. Before then, AMD CPUs were undeniably hot. However, Intel screwed-up so bad with the Pentium-4 that they leapfrogged AMD's power management problem with a CPU so hot no ammount of powermanagement could save it...

Re:Hardware

[MobileTatsu-NJG]

With AMD's reputation for producing hot-running processors

What reputation?

You're posting on a site that still mods BSOD jokes as funny.

Re:Hardware

[TheRaven64]

You distribute n identical jobs across n identical cluster nodes, then you wait for them to finish. 2 of these cluster nodes have had their heatsinks fail and the CPU has throttled back to around 100MHz, the others are running at 2.4GHz, one node has failed completely. The job scheduled on the failed node is rescheduled on another node when it becomes free. The throttled nodes continue working on the job for a couple of days while all of the others finished after a couple of hours.

Re:Hardware

[Chris Burke]

It's hard to believe that at one time CPUs didn't have heat sinks at all.

Hehe. Yeah, and I remember getting bit by the change and feeling rather foolish. I'd built a computer for my step-sister, and, well, I really never even thought about adding a heat sink. About a year and a half, maybe two years later, she was complaining that it kept crashing so she was just going to the computer lab (defeating the point of her having her own), so I took it and checked it out. Well by this point my own system very much needed a heat sink and fan, and when I opened up hers I immediately noticed the lack of one. I took a spare heat sink/fan (and it was a tiny little thing), put it in, and the machine stop crashing. She had no more problems with it.

Well over a year without a heatsink before overheating had caused enough damage to make it actually start crashing. Kinda amazing compared to, say, the K7 I fried in at most one second by mis-applying the heat sink.

Re:Hardware

[sjames]

I was amazed the first time I saw a CPU with a loose heatsink where the label on the bottom actually charred. Alas, it became a common diagnosis for AMD. Intel overheated as well, but I tended to diagnose that by the machine suddenly slowing down to a crawl.

I am glad AMD added internal thermal monitoring. Before that they weren't the hottest running, but did suffer more death by heat.

Re:Hardware

[Anonymous Coward]

You're posting on a site that still mods BSOD jokes as funny.

It's BSD.

Re:Hardware

[Chris Burke]

I am glad AMD added internal thermal monitoring. Before that they weren't the hottest running, but did suffer more death by heat.

No kidding! It was way too long to get that going. If they'd had it from the beginning, I probably would have only been mildly embarrassed when my processor didn't boot and I figured out I hadn't taken the sticker off the bottom of the heatsink, rather than extremely humiliated by my own noobness and a $200 processor-turned-shitty-paperweight. :)

Re:Hardware

[BikeHelmet]

The K6 series were very hot compared to Intel equivalents.

Irrelevent to his argument. See below:

Don't get me wrong, Intel has certainly regained my respect when it comes to performance, but to call AMD the toaster requires ignoring the past 10 years.

Why yes, time flew by that quickly.

Re:Hardware

[palegray.net]

I think you're making horrible assumptions about the nature of the jobs and scheduling system in use. To top it off, you seem to consider other factors a frequent cause for unpredictable computational inefficiency, which usually isn't the case in controlled environments (while other bottlenecks certainly exist, they're typically known values in clustered systems that adhere to predictable congestion patterns). This is probably a consequence of never having actually done any meaningful cluster computing yourself; in the event that my assumption is false, I can't fathom how you were awake for the event, given your apparent level of knowledge on the topic.

Re:Hardware

[Anonymous Coward]

AMD 486, K5 K6 and early K7 were really bad for "dead Buckets"
however later iterations of K7 and all K8 are reliable.

The introduction of large chain suppliers Lenovo, HP/Compaq and Dell releasing more AMD based systems is testament to that.
If the components of a system have x failure rate Lenovo, HP/Compaq and Dell pays X amount honouring support and warranty on that product.

There probably is a failure rate but each supplier expects a failure rate and integrates it with sale price.

Re:Hardware

[TheRaven64]

And the t-bird, which I mentioned in the very next sentence, was introduced in mid 2000 and was shipped for a long time after that; I got mine in summer 2001, which is under 10 years ago. I still have a couple in active use, although one of the machines has had 6 of them die in it and been repaired by cannibalising other machines from the same purchase so they're not exactly ancient.

Re:Hardware

[i.of.the.storm]

Well, with the Core 2 series I think Intel finally got cooler CPUs. All I know is that my Turion laptop has horrible heat dissipation issues (but it's a 12" tablet convertible, so it sort of makes sense). Plus some of the nVidia chipsets that the Turions were paired with are poorly engineered for heat dissipation so it compounds the problem in mine. But I don't think AMD has ever been really better in the mobile CPU department since Pentium M came out. I've decided that I'll probably keep buying AMD desktop CPUs, but for laptops I'll probably have to go with Intel unless AMD really gets their act together.

Re:Hardware

[i.of.the.storm]

Pentium 4 Prescott cannot be beat in that regard.

Re:Hardware

[hairyfeet]

While this may be true, I haven't seen any "real world nastiness" benchmarks, have you? Does anybody have any "real world" stress tests to see how the newer AMD chips hold up to grime, funk, and other nasty conditions? Because in the real world there is nothing more nasty than many construction houses. Plumbers, electricians, carpenters, etc, all have poorly ventilated shops, lots of tracked in dirt, grime, and sawdust, they just seem to take a lot of abuse.

I was sent in to work on this PC one time that was in a trucking depo. You know the kind of place like on "taxi" where all the drivers come to pick up assignments? Well first of all the thing had actually been waxed to the floor by the cleaning crews over the years, so I had to take a truck dolly and roll myself under the secretaries desk (nice legs BTW) to work on the thing. When I get the side off the combination of dirt and cigarette smoke had made a solid block inside the PC. I'm guessing the fans in that thing probably hadn't spun up in a good year. It was either a late model P3 or early P4, I can't remember which, but even after that level of abuse that sucker just kept right on working, slow as Xmas of course. I cleaned her up and she ran as good as new. Hell probably still running to this day.

The oldest Intel I ever replaced was a 10MHz (not a typo-ten MHz) running DOS 3 that ran this funky ISA card for a custom lathe. The HDD slot had gotten flaky but damned if the CPU wasn't still good! The guy ended up paying me nearly $300 for my old P100 and P233Mhz gamer rigs because by 2003 finding ISA slots was getting pretty damned difficult, much less finding ISA slots and a PC that would run DOS 3. I showed them how to boot the 233Mhz once a month to keep the HDD from locking over time and put an image of the DOS 3 with the lathe driver in both. I'm sure that to this day that old workhorse is still cranking out custom columns for that lumber company.

So does anybody have any real world tests, or hell even long term experience with any of the newer AMD chips that can tell me how they survive in extreme conditions? Because the bang for the buck has gotten so good on AMD that I have already switched most of my home builds to AMD, and in the tradtion of "eating one's own dogfood" built myself an AMD 7550 for my current gamer rig. Any real world knowledge would be appreciated, because as I said it is MY ass who is on the hook if they blow during the warranty period.

Re:Hardware

[sumdumass]

I do not think you will find those types of evaluations because it isn't really a function of the processor but the cooling of it.

The so called dead buckets with exposure to grime is where Intel had the CPU throttling and AMD didn't. AMD also used some heat sinks that had the fins very close together which trapped lint and grim pretty well but that was alleviated by using a different heat sink or by cleaning the heat sink periodically. I do not know how many Intel's I taken on trade for new computers just to find out that the entire reason they were dogs is because the heat sink was clogged and the CPU was throttled back to the equivalent of a 1998 processor.

When I place a system into a harsh environment, I often use a positive ventilation box to enclose the system. This filters most of the dust and stuff before it even touches the computers internal components. If you know the system will be exposed like this, it's something you should think about too. You can get really elaborate systems costing crap loads of money or you can slap one together using Furnace filters, some furring strips and maybe some cheap, 1/4 inch plywood and just use a box fan on low sitting beside it and blowing in. Some computer cases also have enough filters and screens to pretty much qualify as the same. If making one, you don't need to enclose the entire computer, just enough so that most the air around it has passed through the filter which mean you want a small breeze at the open ends.

The funky ISA card you mentioned was a CNC controller. I do support for several machine shops and we have them all over the place. We also use wireless serial ports to save on cable runs and allow one machine to control 6 or 12 CNC machines. In all the machine shops, we do the positive ventilation around the computer which has increased it's speed and reliability enormously.

Re:Hardware

[hairyfeet]

Well I know the stock AMD CPU fans did seem to be a lesser quality than the stock Intel, but then again they learned with netburst not to cheap out on the cooling so that may be the reason why. Unfortunately my construction customers like the box to "look purty" so I can't just throw the thing in a box like I'd like to, but I do put 120mm fans, use oversized cases, and put filters whenever possible. Lucky for me my customers will usually bone something and call me out long before the filters get too nasty, so I can just kill two birds with one stone.

As for the CNC controller, I hope the ones you deal with aren't a royal PITA like this one was. The machine was bought in the early 80s, from a company that apparently went out of business like mid 86, so naturally there weren't any other OSes that this thing would run on but DOS 3, and of course by 03 ISA card slots in PCs were pretty damned rare, so too was anybody that knew dick about old DOS commands and setting it up. By the time the poor guy walked into our shop he was practically in tears because every shop he had been to had said it would be at least a week and a half before they could get something that would work, and he had a $50k contract due that needed work done by the lathe in 3 days. Needless to say he was ready to panic and I thought he was gonna have a breakdown when the boss said "I don't think we got anything that will run that".

I swear I though he was gonna hug me when I said "I got a couple of old gamer rigs at home that'll probably run that." He ended up buying these two boxes that weren't worth $100 together for $300 off me, PLUS paid me a full day with overtime PLUS Doug to let me loose for the day just so I could spend the day fixing his problem. When he saw that CNC lathe fire up I thought he was gonna dance, he was so damned tickled. So I truly hope yours ain't a royal PITA like that one was, because good luck getting that old thing to run on anything modern. Of course he said it would cost the company some really serious $$$$ to get that thing replaced, IIRC he said that thing cost close to 100k new. Really was a nice lathe though, made beautiful custom column scroll-work. I bet that original P100MHz is still sitting there, pumping out columns to this day. Between it and the 233MHz I wouldn't be surprised if he gets another 2 decades out of it. They really don't make workhorses like those anymore.

Re:Hardware

[etwills]

early 90s Pentiums were the first CPUs to spark the "you could fry an egg" jokes

At first people ignored that, then they laughed at it, and then they fou^H^H^Hfried it

Not a good idea...

[fuzzyfuzzyfungus]

But with a 40 watt chip you could get that into a laptop, if you felt like it. Not the thinnest, lightest, or quietest laptop around; but plenty of 14-15 inch units under two inches thick(though often not far under) were running P4s at least that power hungry back before P-Ms became cheap enough for common use.

If you were willing to deal with the size and weight of those high-end gamer laptops, the ones with quad core i7s and SLI, you could probably build a 17-inch dual socket system....

Re:Not a good idea...

[WaroDaBeast]

Fudzilla claims these 40 watts we're talking about translate into a 60W TDP though.

Re:Not a good idea...

[drizek]

Games don't scale to 6 cores. At all. In fact, most desktop applications have a hard time scaling to 6 cores, even ones that can do 4 or 8 cores.

Re:Not a good idea...

[angelbunny]

because games tomorrow will not use 6 cores and an i7 is a better idea because it uses threads which operating systems can not even properly seem to take advantage of let alone video games, right?

The post is about making a laptop. laptops are not exclusive to video games.

Re:Not a good idea...

[Anonymous Coward]

Bullshit. I've personally written games (in python, multiprocess, though not multithreaded, because hey, it's python) that scale to more cores than I can afford right now (although if I had enough money, I could conceivably get enough ; and yes, in a few years such hardware will be affordable). And while I don't know how many cores each of my desktop applications can scale to, I know my desktop overall scales to more cores than are currently for sale at any price (that fit into a single ATX formfactor box; I'm not counting networked supercomputers).

Your "software doesn't take advantage of current hardware" argument is so 2004. Time to upgrade your software.

Re:Not a good idea...

[WuphonsReach]

40W chips are very interesting for business desktops where you don't need the latest / fastest CPU chip. Especially if they have 4+ cores and are inexpensive. The lowest power Phenom is 65W, the 905e.

(All of our new builds have been the energy efficient Athlon64 X2s for the past few years. It cuts down on heat inside the PC by a lot. Especially when compared to the 95W/125W AMD CPUs. Noise is also greatly reduced because the CPU fan, and other fans, can spin at a lower speed.)

Re:Not a good idea...

[emilper]

one application might not scale, but Eclipse plus Firefox plus a couple of virtualboxed windows-es running various versions of IE plus a web server plus a db, GIMP, file manager, YM client, Skype and a couple of other applications running at the same time scale nicely ...

TFS is a bit light on details

[clone53421]

Here are a few quick bits from the article:

• Full name: Opteron 2419 EE
• Cost: $989
• Begins shipping: Today
• Power consumption: 40 watts
• Clock speed: 1.8 GHz
• Compatable with DDR-2 memory (cheaper than DDR-3; AMD claims this could save about $1000 per server)
• Compare to the 2377 EE, 40-watt quad-core @ 2.3 GHz: approximately 1/3 more performance from the new six-core chip.

Re:TFS is a bit light on details

[morgan_greywolf]

Compare to the 2377 EE, 40-watt quad-core @ 2.3 GHz: approximately 1/3 more performance from the new six-core chip.

Depends on what kind of server. If you're talking about a Web server, IIS 5.1 and later or Apache 2.x and better with multithreading on, yes. If you're talking about Apache 1.x or 2.x without multithreading, or some older versions of IIS, no.

Re:TFS is a bit light on details

[clone53421]

Interesting; straight math didn't seem to support the article's claim of a 1/3 performance gain, but I assumed the increased parallel capability must be responsible for the extra performance. I'm glad someone with more knowledge was able to clear that up.

Re:TFS is a bit light on details

[idiotnot]

The important information FTFA is here:

"AMD also estimated that the power consumption for a fully populated 42U rack would be 9.2 KW using the six-core Opteron 2425 HE, a 55-W part. Replacing those chips with the 2419 EE would require 7.5 KW, about an 18 percent power savings."

That's just in the rack consumption. I would imagine these probably run cooler, too, which will help with HVAC costs.

AMD seems to be doing a better job shrinking down dated designs at this point. While Intel is selling the Atom, which is undoubtedly cooler and less power-hungry, it's still based on a very old CPU design, which isn't up to heavy computing tasks. AMD, OTOH, has now established a pretty good record of taking mainline processors, and developing lower-power versions. They scaled down what used to be a pretty hot Athlon core (Thunderbird) to the Geode (as used in the OLPC). They followed that with a 45W Athlon 64 X2. Now the Opteron. Intel does have a 35W Conroe, but it's in Celeron cripple-mode badging, a shadow performance-wise, of the original C2Ds that initially came out on that core.

I hope that AMD does release a desktop version of this, but I don't know if they could keep it profitable ($900+ eek.)

Re:TFS is a bit light on details

[dirtyhippie]

I love how AMD is touting the lack of DDR3 support on a new chip as a "feature".

Re:TFS is a bit light on details

[dirtyhippie]

Actually the geode used in olpc is not a scaled down thunderbird. The Geode LX series is based on old cyrix chips (although the Geode NX is a scaled down athlon).

Re:TFS is a bit light on details

[warrior]

It's actually more complicated than that. The six-core chips have the ability to configure up to a quarter of the 6MB L3 cache as a probe filter. This keeps most snoop traffic from reaching down into the L2 and L1 caches of the other cores on the same die and all cores on other die in a multi-socket system. The result is better memory latency and improved memory bandwidth. Here's a link: http://techreport.com/articles.x/16448

Re:TFS is a bit light on details

[Anarke_Incarnate]

The FAIL is yours. You assume that the 6 core offerings from other manufacturers are free? The issue is that the savings in populating the RAM is STILL a $1000 net savings.

Re:TFS is a bit light on details

[Anonymous Coward]

it supports both

Re:TFS is a bit light on details

[angelbunny]

No, I think Hurricane78 is right on this. Why pay $2k for a 2 processor (12core) setup on the CPUs alone and think they are saving $1k by using ddr2? The difference in price is very little today. Not only that but when it comes to vm imho ram is far more important than cpu power. If you want a good setup you'll make sure you get the best ram for the job first then worry about the cpus second.

Who would buy a 12core system for VM and not have it use ddr3 today? It nearly defeats the point.

Re:TFS is a bit light on details

[Anarke_Incarnate]

in many cases for these applications, MORE RAM is better than a faster RAM. Also, the $2000 for 2 CPUs is pittance compared to the overall cost of running those machines. You fail at math as well. You save money with a similar setup going for these CPUs. The savings on RAM are on TOP of that. These are not gaming machines. They are usually more concerned with concurrency than with sheer clock value. This is why many servers do NOT have a single fast core.

Re:TFS is a bit light on details

[John_Booty]

I love how AMD is touting the lack of DDR3 support on a new chip as a "feature".

That was my initial reaction too. However, remember that most (not all, obviously) typical server tasks aren't particularly memory bandwidth-hungry. Email, web serving... even databases aren't usually coming anywhere close to saturating the bandwidth DDR2 can provide, even with several virtualized OSes sharing that bandwidth.

Re:TFS is a bit light on details

[cbhacking]

Leaving aside the question of security, why the hell would somebody run such an outdated web server on brand new hardware? Web servers, database serers, mail servers... all software of that nature has been becoming increasingly multi-threaded over the past few years. IIS 5 is pushing a decade old; the only place you'd run something like that would be on a virtualized system that probably can't access more than 1 core anyhow (making it perfectly reasonable to use software that isn't incredibly parallel).

Re:TFS is a bit light on details

[tietokone-olmi]

Except that with multiprocess concurrency (i.e. non-multithreaded Apache on Unix), you actually gain in a NUMA setup like the Opterons have been from day 1. See, in the optimum case in a NUMA environment, the server process that handles a request gets an entire memory bus for itself. That's far more scalability than with multithreading in the absence of memory duplication, which AFAIR Linux doesn't implement on a per-thread basis in the same address space.

This is why Opterons practically own the 4-socket x86 space: unlike with Intel's older "hub-style" busses, on a NUMA system aggregate memory bandwidth goes up as sockets are added because the number of memory busses increases also.

Re:TFS is a bit light on details

[mikehoskins]

> That's just in the rack consumption. I would imagine these probably run cooler, too, which will help with HVAC costs.

I understand that for every "power watt," it takes 1-2 additional "cooling watts" additional power, in a server room.

So, if a rack takes 10KW, expect an additional 10-20KW of electricity to cool the server room.

I'd, then, estimate 30KW total for a 10KW rack, just to be safe.

So, an 18% savings on 10KW (1.8KW saved), is really saving you on the order of 3.6KW to 5.4KW, when you include cooling!

At $0.10/KWH, you save a bunch of money in electricity.... Just an 18% savings on 10KW (20KW to 30KW total with cooling), means a total savings of $259.20 to $388.80 a month!

Re:TFS is a bit light on details

[hardwarefreak]

Depends on what kind of server. If you're talking about a Web server, IIS 5.1 and later or Apache 2.x and better with multithreading on, yes. If you're talking about Apache 1.x or 2.x without multithreading, or some older versions of IIS, no.

Apache 1.x forks child processes. AFAIK, 2.x does the same. On *nix, any process can be scheduled on any CPU in the system. So your argument doesn't hold water. *nix has been inherently "multi-threaded" or "multi-tasking" for a very long time, on the order of 30 years.

If you're arguing about Apache Win32 and how Windows handles processes and threads, then it doesn't matter, as Windows sucks as a dedicated server platform, web or otherwise, and isn't worth discussion.

Re:TFS is a bit light on details

[yuhong]

To be more precise, the Cyrix MediaGX series, which was sold to NSC and renamed as the Geode, which was again sold to AMD, who makes today's Geode GX/LX chips.

Re:TFS is a bit light on details

[hansamurai]

It's not lack of support, it's that it supports both. Have a bunch of servers you want to get rid of that have DDR2? Just grab that RAM and pop it in this machine, or buy new DDR2 ram at a cheaper price than DDR3 for little performance difference.

Their desktop socket AM2+/AM3 is the same way.

2P

[Enry]

Do they mean Dual Processor? I've never heard the term 2P server before.

Re:2P

[lukas84]

I've seen this many times in German.

Re:2P

[smoker2]

Neither have I. Either they mean 2U or it's a case of " language changes - get used to it !" which roughly translated means "we made up a new term for an old idea to make ourselves look relevant (and we didn't know how to spell 'dual')".

Re:2P

[oldspewey]

Maybe they mean 2U

Re:2P

[Christian Henry]

Do they mean Dual Processor? I've never heard the term 2P server before.

My guess would be 2-partition (as in, two virtual partitions on a single physical server).

Re:2P

[Hadlock]

I've never heard it before either. Seems like a SEO-only term. You see it in the head tag, but the marketing dribble doesn't mention it. Server specs for anything coming up "2p" includes 2 processors, but correlation != causation, especially with marketing folk.

Re:2P

[morgan_greywolf]

My guess would be 2-partition (as in, two virtual partitions on a single physical server).

Now I wonder why he would think that ...

My thoughts about IBM pSeries hardware, AIX, and Linux.

Oh, I get it now....

Re:2P

[Anonymous Coward]

Apparently, "2P" does indeed mean dual processor in adspeak.

Citation: http://searchoracle.techtarget.com/generic/0,295582,sid41_gci1362417,00.html

Re:2P

[Kjella]

I've seen it before, usually used in a context where you have 2P, 4P, 8P = dual-processor, quad-processor, octo-processor machines because noone wants to go around remembering what that should be abbreviated like. Of course, with cores per chip varying widely just saying you have a DP/2P machine says little these days.

Re:2P

[squallbsr]

2P = 2 Processor (as in physical CPU slots)...

Re:2P

[Chris Burke]

Of course, with cores per chip varying widely just saying you have a DP/2P machine says little these days.

Naw, it's just been transmuted to mean processor sockets. Which from a system architecture standpoint is the more meaningful way to do it. You can put anything from a single-core to a six-core processor in a given socket (assuming they all exist in the necessary package), but you can't change the number of sockets in your motherboard. So "2P server" tells you that there are two sockets which you could potentially populate with 2, 4, or 6 core processors.

Or just think of "processor" as all the stuff in the package, and that a "processor" could have multiple "cores", and it works just fine.

Re:2P

[CarpetShark]

I've seen it many times in UK confectionery stores.

Re:2P

[kramulous]

Works fine until we start getting specialised cores with specialised instructions. I've adopted sockets and processors. A pizza box can 8 processors (two sockets). A single processor, in this case, is still a generalised processing element.

Re:2P

[Chris Burke]

Works fine until we start getting specialised cores with specialised instructions.

I don't see why heterogeneity should change anything.

The processor on my desk contains two x86 cores, and a 'north bridge' memory controller with DMA engine and APIC. Soon there will be ones that also have a graphics core on them as well. I'd still call that a processor.

The Cell Processor contains a single general-purpose Power core, 8 relatively specialized cores, and a memory controller, but I feel comfortable calling the sum of them a processor.

You could call the FP unit in the processor on my desk a 'specialized core with specialized instructions', but it's still part of the overall processor. If the FP unit was its own separate DSP core, I'd still call it part of the processor.

I've adopted sockets and processors. A pizza box can 8 processors (two sockets).

So... this nomenclature doesn't take into account future developments in Multicore Mania? This story is about six cores in one socket, so you're already out of date. ;)

Also, while we can easily refer to a motherboard as having two sockets, a socket is still just a thing that you plug chips into. What do you call the thing you put in the socket? A package, but that's literally a container not a thing contained. I suggest "processor" when that is the type of thing that is in the package that goes into the socket. :)

A single processor, in this case, is still a generalised processing element.

Meaning, a single core that runs a particular instruction set considered 'general' for these purposes.

I just think your definition of 'general' is too specific. :)

The first "GP" in "GP-GPU" stands for "general purpose". Sure graphics cores aren't all that well suited for truly general purpose computing, but what difference does that make if the processor is using it for data processing? Same with the Cell SPEs. Same with math "coprocessors" once they were moved onto the chip.

Personally I would suggest that we continue to call everything that fits in one package, and contains processing elements of whatever nature, a "processor", and the thing it goes into a "processor socket". Leave "cores" as a separate thing, because it makes it easy to talk about a processor with heterogenous cores -- "two x86 cores, one graphics core". Also it prevents confusion when architectures get more complicated, like lets say two x86 cores sharing a single FP unit -- is the FP unit, specialized as it is, not a processor, or not part of the processors surrounding it? What about SMT? Cores are fairly well defined physically, but perhaps not so well defined logically. See what I'm getting at?

Re:2P

[kramulous]

I see the point you're making, but what happens when I write a parallel program? I make a request to the array services daemon that I want 4 processors and that gives me a single socket (in my current, massively outdated system :) ), or in your nomenclature a single processor. I see it as having 4 processors (or cores if you will) and hence the need to get a naming scheme that is consistant across hardware and software.

And please don't say threads :s

Re:2P

[Chris Burke]

I see the point you're making, but what happens when I write a parallel program? [snip] I see it as having 4 processors (or cores if you will) and hence the need to get a naming scheme that is consistant across hardware and software.

And please don't say threads :s

Well sorry, I hate to break it to you, but regardless of whether you use my nomenclature or yours, that's the only term that is going to be consistent across hardware and software.

Take for example a case I already mentioned: The Nehalem (Core i7) processor. It has 4 cores, each of which supports 2-way SMT (simultaneous multi-threading).

I mean, on the software side, the thing that you're trying to schedule are definitely threads, you ask the OS to schedule your thread, not a processor, and the thing you're interested in is how many threads can you usefully run in parallel. And for Nehalem, the answer is not 4 (number of cores/processors), but 8 (number of hardware threads).

There's a lot more wiggle room in the definition of "processor" than there is for "thread", and the "thread" terminology is the only one that properly bridges hardware and software. You're going to be hearing it more in the future, not less, so I suggest you get used to it. Sorry! :)

Who feels like explaining a bit

[Anonymous Coward]

Anyone here know enough about CPU design who can guesstimate what the lower bound on CPU energy consumption is? I think I understand that you can lower the operating voltage of the chip, but this leads to more computation errors due to thermal noise. Or lower the clock speed of course... but flops/W would stay the same. Or use a lithography process that produces smaller size features. Then if you get too small though things don't quite work the same due to quantum effects etc. Does using more cores help? How is AMD going about this problem?

Gaming/compiler performance?

[bl8n8r]

It's a nice marketing strategy -- "My cores outnumber your cores" -- but where is the performance gain when the CPU speed is almost half that of a dual-core 3.2GHZ?

Re:Gaming/compiler performance?

[clone53421]

6 x 1.8 = 10.8
2 x 3.2 = 6.4

If you can take full advantage of the six cores, there's a lot more computational power despite the slower clock speed.

Re:Gaming/compiler performance?

[jittles]

It's called multi-threading. Most servers don't need a ton of horse-power, but they want to handle as many simultaneous tasks as possible. More cores means that you can theoretically handle more requests at once.

Re:Gaming/compiler performance?

[lytlebill]

From TFA: "According to IDC data quoted by Brent Kerby, a product manager for the chip, about 82 percent of cloud and Web servers only use about half of their available processor power at any given time." Not intended for gaming or compiling. Low power, multiple cores, it's a server chip.

Re:Gaming/compiler performance?

[maxwell demon]

2 * 1 = 2 < 3 = 6 * 1/2

Re:Gaming/compiler performance?

[WiglyWorm]

This is a server processor. If you are either gaming or compiling on your server, you are doing something wrong. My servers here at work tend to do a high volume of low processor intensity transactions... therefore, more cores (and more simultanious transactions) is far more important than high speed.

Also, by shoehorning this into a 40w envelope, they're obviously going for power efficiency over horsepower. Interesting fact: power usage is one of the largest costs of a data center, and it's growing.

Re:Gaming/compiler performance?

[JWSmythe]

    I had this argument with someone once. They didn't quite get it. The machine they were using was a 4 CPU 700Mhz server. In their logic, 700Mhz * 4 = 2.8Ghz. I wanted to move them to a 2 CPU 1.4Ghz machine, which I promised would be blazing fast. In their mind 1.4Ghz * 2 = 2.8Ghz, so there was no difference.

    There were a bunch of reasons for the move. The hardware was old. The form was huge (like 5u tall) and power hungry. The OS needed to be updated badly, and we couldn't take it offline for a day to do that. One day there was a fault of some kind (it's been a while, I don't remember specifically), so we moved it over to the new machine that I had wanted to move them to. They were amazed. Their $40,000 server had been replaced by a $2,000 server (original costs for both), and it was running faster and better than before. After the move, I repaired their old server, upgraded the OS, and made it ready. I offered to move them back, and they refused. :)

    About a year later, we had a 2CPU 2.4Ghz machine ready for them, and I offered again, "May I move you?" This time there wasn't a complaint. We just scheduled a window and did it. I set a 3 hour window, and we had it completed in about 15 minutes.

    I agree, I'd rather have CPU speed AND cores. I'd sacrifice extra cores for more speed. CPU speed has stagnated, while they're growing cores. I remember this happening in the past too, around the time CPU's were 200Mhz. You could get motherboards that supported one CPU, then 2 CPU, then 4 CPU, but the speeds weren't going up. You could give me 100 CPU's at 200Mhz, but I'd rather have one at 10Ghz.

    I'm sure people will throw a bunch of excuses of why. I remember back when the 50Mhz CPU was the fastest available, there were all kinds of reasons thrown around of why CPU's would "never be faster". People were very insistent that they were right. There were RF interference issues. If CPU's got to RF speeds, radio and TV would cease to work. If we got up near 2.4Ghz, people would be cooked because it's the same frequency as microwave ovens. There was no way to deal with the thermal issues, and computers would be ovens requiring liquid cooling (like liquid nitrogen or helium, not water cooling). Blah, blah, blah, blah. As we've seen, we did get well beyond 50Mhz. It's just a matter of time. I'm just disappointed that we end up stagnating. It's probably financial issues. The market will support a slower multicore CPU, but people won't spend the money on faster CPU's right now.

    I always love the "latest greatest" craze. It's entertaining. People will spend mad money on latest greatest, and I'll wait 6 months or a year to buy the same thing at a fraction of the cost. Maybe I'm part of the problem there. I won't drop $500 on a CPU, but I'll drop $100 on last years model that's only slightly slower.

    At least right now it's nice, since I can buy older and older hardware, and really not be far behind the curve. :)

Re:Gaming/compiler performance?

[JohnFluxx]

> I always love the "latest greatest" craze. It's entertaining. People will spend mad money on latest greatest, and I'll wait 6 months or a year to buy the same thing at a fraction of the cost. Maybe I'm part of the problem there. I won't drop $500 on a CPU, but I'll drop $100 on last years model that's only slightly slower.

Say you wait 6 months before upgrading. $400 divided by 6 months = $2 per day.
Assuming you earn $20 an hour, if the new chip saved you just 6 minutes per day, then it's a worth while investment to upgrade sooner rather than later.

Re:Gaming/compiler performance?

[Anonymous Coward]

"If you are either gaming or compiling on your server, you are doing something wrong. "

Now I finally understand why MUD was banned on our group server.

Re:Gaming/compiler performance?

[CaptainPatent]

This is a server processor - one of the biggest concerns is performance per watt. I'm assuming you're referring to one of Intel's dual cores (but even if you look at dual core AMD offerings the wattage is a significant increase) First you're ignoring pipeline utilization. Generally speaking, slower and newer processors make better use of their pipelines. Of course older procs don't have all of the design improvements that increase throughput, but the factor that's ignored is that a lot of "high-end" ultra-fast cpus had to have their pipeline dumbed-down a bit to achieve those marks. Guess what - that's the actual marketing ploy here. Even if you ignore pipeline utilization, the 3.2GHz Processors I found weigh in anywhere from 35 Watts (kind of - it's AMD's rating of 3600+ and not a clock speed on the FX 64) to 130 Watts on the Intel Xeon. When you're paying for the electricity on a server room with hundreds to even tens-of-thousands of these processors - this may become the standard in where to go.

Re:Gaming/compiler performance?

[Belial6]

I like your math, but you do have to change it just a tad. The $2 savings is only if he is paying for the work. If he is getting paid, then there is no savings in getting the job done 6 minutes sooner. If he is hourly, he will get paid $2 less each day, while spending an extra $2. This means a $4 a day loss. If he is salary, then there is no change in his income, but he still pays out the $2 a day in equipment costs, and thus still loses money.

For the one paying the wages, there certainly can be a savings. So, for a company that is paying an employee, your math can be correct in some instances.

That all being said, from a non-economic standing, it may still make sense to upgrade. I know, I would rather have the extra 6 minutes of time, even if it is just spent getting a cup of coffee, or just being productive on something else. Ok, Ok, even if it is spent posting on Slashdot about how I would rather have the extra 6 minutes.

Re:Gaming/compiler performance?

[mindbrane]

I wait until the latest and greatest has gained the larger part of retail shelf space then buy the most established and robustly tested stuff at the price nearest it's exit price just before retailers stop carrying it. The problem with the latest and greatest stuff is that often there's bells and whistles that won't make it because the market goes in another direction.

I just built an intel Quad core for under $400, but used some components from cannibalized systems. Right now USB, SATA, and the PCI bus all have new standards on the way and I can't see building a new box on the latest and greatest until those three items have been well tested. I'm not sure too many informed buyers will look at new stuff until, figuratively speaking, the equivalent of SP1 comes out.

Re:Gaming/compiler performance?

[afidel]

Power usage is actually going down per unit work, by a LOT due to virtualization. Compared to my standard server from just 3 years ago I can do 17:1 virtualization today without any major over-subscription of resources.

Re:Gaming/compiler performance?

[scorp1us]

Except that those requests are I/O bound, and therefore are in a blocking-wait and not scheduled by the OS. How many processes are actually in a run state? Not many most of the time. I am lucky if I have two. But having greater than 1 instruction pipeline does wonders for responsiveness. Look even at hyper-threading. Did a great job for responsiveness, even with only one CPU.

Re:Gaming/compiler performance?

[Spatial]

CPU speed has stagnated

It hasn't stagnated at all. You're equating cycle rate with performance, that's incorrect.

Each processor architecture does a different amount of work each cycle. Counting only the number of cycles is like comparing the running speed of two men by the number of steps they take each minute - but one guy may be a midget and the other eight feet tall. Clock speeds remain similar but performance doesn't correlate.

For example, a 3Ghz P4 isn't even half as fast as one core from a 3Ghz Core i7. The number of instructions per clock have been continuously improving with each new architecture.
Phenom is faster than Athlon X2. Phenom II is faster than Phenom.

Core 2 is faster than Pentium 4. Core i7 is faster than Core 2.

So you can have what you want - improvement continues in both per-core performance and the number of cores.

Re:Gaming/compiler performance?

[clone53421]

Analogy fail. It's more like saying 6 assembly lines producing 1.8 cars per hour can make more cars per hour than 2 assembly lines producing 3.2 cars per hour.

Assuming you have stuff pipelined correctly so as to take full advantage of all six cores, of course... which I explicitly required in my previous post. If you're only able to keep 2 cores reasonably busy, the 6-core processor will be slower because 4 of its cores are mostly idle.

Re:Gaming/compiler performance?

[CAIMLAS]

One possible way to solve this is to figure out how to scale single processes over multiple cores. As far as I know, that hasn't been done yet, and I don't even know if it's possible. But if it is possible, that adaptability seems like it'd kick the whole "programming for multiple cores" thing in the ass pretty readily.

Multithreading? Check, we've got that. Multiprocess? Sure, we've got that too, though there are still a lot of applications which don't do it, or do it wrong. But if it could be done transparently (at the OS level), it'd be a win-win for everyone, I think - even if there was a 10-20% overall performance penalty as a result. It'd at least buy us some time in trying to figure out how to either increase clock speeds again or how to effectively program for multiple cores.

Re:Gaming/compiler performance?

[JWSmythe]

    While I do agree, I've been considering replacing my desktop (and adding a second one), when money allows. CompUSA is still alive in my city, so I wander the isles when I'm there. I can get a $300 machine that's faster than any of my old desktops. Ok, so it's an eMachines, big deal. I bought one about 2 years ago to be a mail filter server. It was an AMD64 with 1Gb RAM. I upgraded the RAM immediately (we needed more for the mail load). The motherboard died, so I picked up an Asus board for about $50. So, it's a crappy $300 eMachines computer, but now there's nothing crappy about it except the name on the case. :)

Re:Gaming/compiler performance?

[angelbunny]

I disagree. Often times the cache ends up defining the performance in the end. Sure it doesn't effect raw number power but if that is all we where doing today we would go for gpus alone with the 8000+ threads in a lot of the cards today.

If you have a 6mb cache (shared cache not private cache) between 2 cores or 6 cores are those 6 cores really going to run faster in a real world gaming situation based on fps? I'm not saying the 6 core cpu will be slower just that the performance will not be 3x like you imply it will. It would be closer to a 1.2 to a 1.8x performance increase for most games that are designed to be fully threaded properly and if both have the same clock speeds.

What I'm saying is in most gaming situations the 2 core 3.2 will blow away the 6 1.8 even in the best of threaded situations.

This is why the 6 core cpu is not aimed for more than servers and even in the server market it is aimed entirely for vm where the multiple cores will help the most.

Re:Gaming/compiler performance?

[clone53421]

While I don't disagree, I'd like to point out that you described situations where the 6-core processor can't be taxed to its full capability.

In other words, your main point was that the only situation where it can be fully loaded is "in the server market [and] aimed entirely for vm where the multiple cores will help the most". That's a valid point, but it's not presented foremost in your post. It's sort of hidden down at the bottom.

Re:Gaming/compiler performance?

[mindbrane]

Totally agree. I still run 2 single core Athlon 64s S939 as servers. I'm typing this on my old intel D865 2.8 gaming box running Ubuntu. My dual core Athlon 3800 is my production box and runs Handbrake all day every day while I hack around with code 'n stuff. The Quad runs Vista (ya I know) as a gaming and Home Theater box in the living room. From the old P4 on up, my biggest problem is not being able to justify junking boxes that are still very usable, but they're multiplying like pet rabbits and I've got to find good homes for them all. :)

Re:Gaming/compiler performance?

[kramulous]

CPU speed has stagnated,

Not even close. Floating point performance on a single core has still been increasing according to Moores law. 256 bit wide registers, vectorised/packed instructions, xsse4.1, etc

Re:Gaming/compiler performance?

[jamesswift]

"compiling on your server, you are doing something wrong."

I do 'make -j16' all the time on a server.

Re:Gaming/compiler performance?

[JWSmythe]

    Well, if you saw my resume, or knew the work I've done, you'd know better. Hell, if you've kept up with my journal on here, you'd have a hint. Not that I have anything to prove, but I've been in and out of ;arge DC's for over 10 years. I've been in enough Equinix facilities where I'd forget which one I was in at any given point, and would get lost inside. Well, it's easy enough to get lost in them, it's not like they keep a logical floor plan on any of them. I've also been hosted in Switch&Data, Level3, Qwest, AT&T, and worked in small ones where I didn't even care to remember who they were.

    The "may I move you" was a partner company's database server. The guy in charge was bent on staying on that machine. It wasn't the best business choice to stay on that machine, so the "may I move you" was the polite way to begin. It's a good way to start a holy war with management if you say "I am moving you", even though it's their purchased equipment. It's also a good way to go from a senior position to unemployed by morning. In the end, they did purchase an almost identical machine to the DB server that I moved them to, so the new machine was their purchased hardware. Since the old machine was so big, and archaic, they didn't want it shipped back to their office, so it was stuck at mine.

Re:first post

[JWSmythe]

5th core

Re:first post

[Anonymous Coward]

c-c-c-core breaker!

Re:first post

[maxwell demon]

Core dumped.

WTF is 2P?

[h4rr4r]

How many U to a P?
Or is that supposed to be dual CPU/dual socket?

Re:WTF is 2P?

[mister_playboy]

U to P ratio? Is that the nerd's replacement for H to W ratio? I'm all about a 1:1 T to A ratio myself.

Re:WTF is 2P?

[mcmonkey]

Shouldn't T to A be 2:1?

Re:first post

[oldspewey]

Are we getting close to TDP yet?

Re:2P ...

[lagfest]

or not 2P, that is the question.

It might get unpleasant if you hold it in too long.

Well

[Anonymous Coward]

That's what i call a 6 pack

Miss the part about virtual servers?

[Mathinker]

> If you're talking about Apache 1.x or 2.x without multithreading, or some older versions of IIS, no.

Summary:

> ... among the most popular in the virtualized server space.

So your comment ignores the fact that this CPU will probably be running 6 (or more) VMs, which could just as well run single-threaded code....

Re:Miss the part about virtual servers?

[morgan_greywolf]

So your comment ignores the fact that this CPU will probably be running 6 (or more) VMs, which could just as well run single-threaded code....

Clearly that's the market that this chip is targetting. I'm simply pointing out that 1) even if you're not in that space, this chip compares favorably to a 2.3 Ghz Quad Core or even a 3.2 Ghz Dual Core, so long as you're running multithreaded apps on any OS that uses a sane threading model. Just understand that single-threaded apps won't compare unless you're running them virtualized.

OTOH, separate blades are going to give you to better performance, no matter how you look at it. On the gripping hand, if you have 10 blades, each running this 6 core CPU (conceivable in 40 watts!) and a VM hypervisor... sounds delicious.

Re:Miss the part about virtual servers?

[afidel]

Try 32 half width/half height blades in 10U each running 2x6 core CPU's running VM's (actually I just checked and HP doesn't have a 2x blade based on these CPU's for some reason despite the fact that they are lower power than the 50W or 80W Intel's they currently use for the 2x). Of course the problem is that most VM implementations are memory not CPU bound so Intel wins for VM density with 192GB in a half height blade (HP BL460 G6). 512 cores and 12TB of ram in one rack is an amazing amount of computing power, but cooling that kind of power density is difficult to say the least.

Re:Miss the part about virtual servers?

[morgan_greywolf]

I'll take two orders with hot sauce on the side!

Wouldn't be a laptop I'd want to use

[ciroknight]

Most laptops today have much more power efficient chips (AMD's line tops out at 35W, Intel's 25W, most do quite a bit less, especially with all of the fancy power-saving junk thrown in like QuickStart and SpeedStep w/ deeper-sleep DC4). And both of those numbers are just embarrassing with chips like the newer dual-core Atom chips which run at 4W or less at full-tilt and do most everything anyone demands of a laptop anyways.

Now if only someone would wise up and build a 15" laptop with an Atom chip, and LED display and a 9-cell battery... mmm, 8+ hours of battery life.

Re:Wouldn't be a laptop I'd want to use

[K. S. Kyosuke]

The thing is that server laptops have already appeared on the market. You can easily buy a laptop with a Xeon, why not with a 40 W Opteron? Even though, like the Xeon-powered ones, I would rather call it a portable rather than a laptop.

Re:Wouldn't be a laptop I'd want to use

[jawtheshark]

I always fancied one of these. Probably just me.

Re:Wouldn't be a laptop I'd want to use

[fuzzyfuzzyfungus]

If you really want to go overboard on "dubiously suitable for laptop use", you can get one of these.

Dual UltraSPARCs, up to 16 gigs of RAM, full sized 64 bit PCI slot, 3GbE ports. Of course, it's 22 pounds, and they don't even say what it costs.

Re:Wouldn't be a laptop I'd want to use

[gbarules2999]

What do you need that extra speed for?

Re:Wouldn't be a laptop I'd want to use

[Ant P.]

Now if only someone would wise up and build a 15" laptop with an Atom chip, and LED display and a 9-cell battery... mmm, 8+ hours of battery life.

They can't. Intel doesn't allow the Atom in full-size laptops at all, or something stupid along those lines.

Re:Wouldn't be a laptop I'd want to use

[Christophotron]

What do you need that extra speed for?

less time wasted staring at an hourglass cursor, of course.. atom is dog-slow and it might be fine for these netbook toy-laptops but I would never buy something resembling a full-sized laptop with a dinky atom processor in it. it's barely tolerable with firefox and xp.

if you aren't seriously bothered by the speed of atom processors then i will assume you run linux/fluxbox (or the equivalent) and a very lightweight web browser. and if that's the case (you're already running an OS that is not restricted to x86) then why wouldn't you prefer the more-power-efficient ARM processor over an atom?

i already have an eeepc and i considered purchasing another one but i just don't see the point until they can make a better processor for it that still gets decent battery life. although i dislike apple, the poster above has a point about the new macbook pro with regards to the battery life and speed. still too expensive, though.

Re:Wouldn't be a laptop I'd want to use

[Gr8Apes]

What do you need that extra speed for?

less time wasted staring at an hourglass cursor, of course..

hourglass? Perhaps that's the problem?

Re:Wouldn't be a laptop I'd want to use

[Tsuraan]

I know we're on slashdot and all, but the sony vaio Z-Series laptops (the ones with the BIOS disabled VT extensions that you need to hack around) have a 13.3" model with a core2, LED display, and 9-cell (90WH) battery as an option. It draws 10W with wifi disabled, or 12W with wifi enabled. It's not quite 8 hours with the wifi, but in airplane mode it can get over 8. And then, when you plug it in, you have 2 2.4GHz processors to work with.

Re:Wouldn't be a laptop I'd want to use

[sowth]

I would rather call it "super crotch-burning sterilizing nightmare." My Asus EEE gets hot enough just running Firefox. I can't imagine what a 40 watt processor would do. Ouch!

I for one would rather research go into how to reduce processor usage for laptops so they don't get so damn hot. Yeah, new tech reduces how hot the CPU gets while doing the same work, but bloat is outpacing it too far.

Re:Wouldn't be a laptop I'd want to use

[sowth]

I don't know. This laptop looks really fast. Look at it fly!

All about threads

[James McP]

This is for situations where you need lots of processes running but that those processes are either easily completed, are low-impact, or limited by bandwidth or the user. Web servers love lots of cores

On gaming you could separate the game into a user environment thread, a physics thread, an object management thread, a pair of AI threads, and still have a core left over for general OS activity.

I know that in theory compilers could also pull loops and modules out to separate threads but I haven't the foggiest clue if that really happens.

Re:2P ...

[oldhack]

You'd be pissed off either way.

Ug! I hate puns.

Re:2P servers?

[earnest murderer]

2 Processor

2P = Dual Socket

[HighWizard]

Single socket (1P), Dual socket (2P).

Re:2P servers?

[Alyred]

Obviously, it's a server that's installed in a hairpiece.

Good for sore muscles

[Mark_in_Brazil]

This makes me feel all warm and toasty inside.

My electric heating pad, which helps me with little muscular issues, is 50 watts, but that dissipation is spread out over a 30 cm x 65 cm surface.

Re:2P ...

[mcgrew]

Schrodinger's beehive -- two bees, or not two bees? THAT is the question!

2P or not 2P is a bad question. As General Patton said (at least in the movie portrayal of him), "never turn down a chance to piss!"

Less Bang For The Buck

[u64]

Doesnt it fall into the same pit as Intel CPUs, low bang for the buck ?

I dont know what server-people need/want. But for me i want the price-tag

to remain while i get faster and better-idle-consumption CPUs.

The CPU that deliver the most bang for my buck is the winner, for me.

All other CPUs are just a big waste of time.

Re:Less Bang For The Buck

[MBGMorden]

$1000 for the processor is peanuts for certain applications. Generally for something like a database server most of the cost goes into software anyways (assuming you're not using an OSS database - if you are that's fine and some of this doesn't apply, but my employer simply doesn't allow it). With MS SQL Server for example, if you're not licensing by CAL's (which is a budgeting headache) is licensed per processor - at about $5000 or so per processor. Neat thing though is multiple cores don't count - only an actual physical processor counts, so for licensing reasons the more cores they can pack onto a single chip, the cheaper my licensing becomes. I've been setting up dual quad-cores on most systems lately just to keep the official processor count at 2, but six-cores would help even more. Remember too that most decent servers will be running SCSI/SAS drives in a RAID config, and many will also have fairly large tape drives (an LTO3 or 4 will up the cost another few thousand by itself).

With those prices, most of the servers I've setup recently have run $15k-$25k with software. None of them are going to get those $50 Celeron's anyways, so these chips generally won't up the cost that much.

What Does Intel Have...

[Nom du Keyboard]

What does Intel have to compete with this on price/power/performance?

Re:What Does Intel Have...

[EmagGeek]

The Atom...

Re:Gimme MHz

[K. S. Kyosuke]

Ask Koreans (the northern ones), they can supply you with several megadeaths by now.

Re:first post

[maxwell demon]

Are we getting close to TDP yet?

TDP = Thread Dead Point?

Was it the processors or the memory?

[DamnStupidElf]

A switch from PC100 to DDR would yield quite a bit more performance than just going from 4 processors to 2. While it's true that a single processor is better because it has a unified cache and no contention with other processors for the memory/IO bus, those days are over for now. Multiprocessing is making a comeback, and unless there's an amazing revolution in chip design it will be easier to get bigger overall MHz numbers by multiprocessing than my speeding up individual processors. If you really want the MHz, look at IBM's Power line. They actually have multicore chips with high clock rates.

Re:Was it the processors or the memory?

[JWSmythe]

    I won't argue the memory probably helped with the overall speed, just as better drives did, but all in all I'd say it was the CPU speed. The faster memory and better drives are an added bonus. :)

    It was really hard to get him to step away from that old server, since there was such an investment made in it. In the end, after we decided not to put that server back into production for that purpose, I couldn't convince anyone to use it. I don't know it's final fate, but I did suggest that it would make very nice boat anchor. :) It took two of us to even consider racking it, even though I did manage to move it once by myself (and suffered the pain after). I figured if they ever used it as an anchor, the best option would be to cut the line when they wanted to move again.

    The damned thing was huge, noisy, hot, and slower than even the desktops we were buying at the time. I suggested using it as a file server, since it had a decent sized raid on it, but it made the relatively quiet server room rather loud. It was fun annoying people by turning it on in my office. You could hear it running two offices over. :)

Re:Gimme MHz

[Bakkster]

That worked great for the Pentium 4, didn't it? Faster clock != more instructions per second. The only way to get close to 4GHz on the Pentium was with a 31-stage pipeline. http://en.wikipedia.org/wiki/Instruction_pipeline

This means, on an instruction like if(a+b>c){}, the actual branch gets delayed by about 20 cycles if the processor guesses incorrectly whether the if statement should execute or not. Add the overhead due to such a fast clock (the P4 could only have 4 logic gates per pipeline stage due to the speed).

I'll keep my more efficient, better laid out processors over raw GHz, thank you very much.

Re:Gimme MHz

[h4rm0ny]

Except for the "wah!" there's truth to what you say. Well, excluding the "give me death" part as well. In fact, let's just concentrate on the middle bit.

It would be nice to see some greater increase in speed. There are reasons to have that. But in the meantime, I will accept code properly written to utilise multiple cores as a substitute.

We're getting robbed of latin related fun

[spazimodo]

The previous generations of multi-core CPUs weren't 2-core and 4-core, they were dual and quad-core. These new chips should pretty obviously be called sex-cores. Not since the 667MHZ PII have I been so disappointed.

Cyrix

[mollog]

Cyrix! I had one of those. A DRX 50, IIRC. I upgraded and overclocked my 16MHz 386 HP Vectra to 50MHz with one of those. Those were the days.

I running a funky AMD four core with a TLB bug. Works fine on Windows XP. I own Intel stock but use AMD chips. I'm looking forward to using one of these low power chips on a HTPC.

Re:Gimme MHz

[PitaBred]

My Phenom II 955 will even do 3.6GHz on the stock cooler now, probably even a lower thermal envelope than the P4's of the time. And the Core i7's (I'm too cheap for a full system replacement) are faster and cooler still.

Re:first post

[MobileTatsu-NJG]

6th Core, but I'm the bad one so the 7th core will have to be the 6th one.

Re:Gimme MHz

[Bakkster]

The P4 only managed 4 gates at 4GHz because it was 90nm manufacturing. Phenom II and the i7 are 45nm, and the faster gates enable the higher clock speeds naturally and without huge tradeoffs, unlike the P4 where GHz drove development instead of performance.

Re:Gimme MHz

[Chris Burke]

My Phenom II 955 will even do 3.6GHz on the stock cooler now, probably even a lower thermal envelope than the P4's of the time. And the Core i7's (I'm too cheap for a full system replacement) are faster and cooler still.

To add to my sibling post, the P4 was supposed to be at ~10GHz by now. Intel's projections based on further technology shrinks, and further extended pipelines, was pretty reasonable from a transistor speed and clock speed/IPC tradeoff perspective. They failed to account for an explosion in leakage current in smaller technology nodes, and for a market shift towards performance/watt or performance/(watt*$) as the metrics of choice. That's what doomed the architecture.

The article is confusing ACP for TDP

[George_Ou]

The article is confusing ACP for TDP. The 6-core is NOT "inside 40 watts", it's inside 60 watts TDP. Moreover, AMD is getting killed in energy efficiency if you look at the standard benchmarks such as SPECpower. The âoelow powerâ AMD chips consumer more power on average than Intelâ(TM)s highest wattage highest performance CPU parts.

There's wiggle room for both, really.

[Gazzonyx]

Doesn't scheduling (process vs. thread) depend on the OS? IIRC Linux schedules threads the same as processes. IE. quantum slice size/interactivity credits and dynamic priorities are calculated per thread and penalties are assigned to their parent process. Windows, OTOH, does all of the accounting at the process level. We'll pretend for the moment that Windows fork is POSIX compliant.

Also, I'm not sure if it only applied to Green Threads (which have mostly fallen out of favor - at least in the JIT world. I think some interpreted language runtimes may still use them.), but I have heard of threads being broken down in to fibers.

That being said; you're right. The concept of a 'processor' has been abstracted so much that many EULAs have to define the term verbosely when the license limits the number of processors that the app can run on. I think the default definition, currently, is that a processor is a single socket and all of its cores. It will be interesting to see what happens as socket packages become essentially multiple NUMA style processors (with their own L[123] caches/dedicated RAM and memory controllers) that share only a high speed internal and external bus.

Sorry for rambling, it's quite late here.

Re:There's wiggle room for both, really.

[Chris Burke]

Doesn't scheduling (process vs. thread) depend on the OS? IIRC Linux schedules threads the same as processes. IE. quantum slice size/interactivity credits and dynamic priorities are calculated per thread and penalties are assigned to their parent process. Windows, OTOH, does all of the accounting at the process level. We'll pretend for the moment that Windows fork is POSIX compliant.

Okay, it's true that the definitions of 'thread' and 'process' and what have you vary as much as implementations, but that's all that is, implementation details, the majority of it in the OS. How the OS decides to schedule various things isn't really the issue.

In the most general definition of "thread" as a "thread of execution", it merely means a "context" (relevant register values including page table pointer and instruction pointer) executed on the machine. This definition is a super-set of all other definitions. After all, contexts is all the hardware actually understands. All the additional qualifiers are added by the OS: e.g. in linux a "process" has its own page table, while "threads" within a process do not, but the hardware doesn't care, it just sees a context switch and whether or not the CR3 values are the same or different is just a coincidence (that the hardware may or may not exploit).

So, at the end of the day, the thing you as OS-scheduler-writer or you as parallel programmer are interested in is how many simultaneous threads of execution the hardware supports. The names of the data structures used to manipulate those threads may vary in your OS/library of choice, but that doesn't change what the basic unit of interest in the hardware is, and it certainly isn't the "processor".

It will be interesting to see what happens as socket packages become essentially multiple NUMA style processors (with their own L[123] caches/dedicated RAM and memory controllers) that share only a high speed internal and external bus.

Well I think that's either already happened or on the cusp of happening. With large L3s, you can have different latencies to different parts of the L3 from different cores on the chip. Intel's MCMs communicate on-package using the external data bus. I don't see much future for dedicated per-core DRAM controllers. That doesn't make much sense. For one, it's highly useful to have a shared cache at some level (L3 currently), and the DRAM naturally goes on the other side of the L3. DRAM pins coming off the package are a scarce resource, and tying those pins to a core would be wasteful when that core doesn't need the bandwidth but another does.