65nm Athlons Debut With Lower Power Consumption

TheRaindog writes "AMD has finally rolled out Athlon 64 X2 processors based on 65nm process technology, and The Tech Report has an interesting look at their energy usage and overclocking potential compared to current 90nm models. The new 65nm chips consume less power at idle and under load than their 90nm counterparts, and appear to have plenty of headroom for overclocking. An Athlon 64 X2 5000+ that normally runs at 2.4 GHz was taken all the way up to 2.9 GHz with standard air cooling and only a marginal voltage boost, suggesting that we may see faster chips from AMD soon."

Comments

HTPC

[tedgyz]

The little gem in this story is the Athlon 64 X2 3800+ EE SFF 2.0GHz. At 35W, that sounds like a perfect CPU choice for a super-silent HTPC.

Re:HTPC

[Gothmolly]

If your HTPC needs that much processing power from its CPU, you're doing it wrong.

Re:HTPC

[Mayhem178]

What, your HTPC can't render Final Fantasy: The Spirits Within on the fly? Lame. ;)

Okay, no, seriously. I have an Athlon X2 3800, and it runs deathly quiet for any operation I've thrown at it. Considering that the machine I have it in is my primary gaming PC, I'd say that's noteworthy. And I've never noticed any great amount of heat production, either.

Re:HTPC

[Neon Spiral Injector]

How do you suggest that one decode 1080i H.264 transport streams with AC3 5.1 audio? This processor may be slightly more than required, but not by much.

Re:HTPC

[javilon]

Do you know of any video player that will be capable of taking advantage of two processors?

As far as I know mplayer doesn't, xine doesn't and vlc doesn't.

Re:HTPC

[Noehre]

Why would you decode the AC3? Use passthrough.

Re:HTPC

[Ultra64]

right, because it's totally impossible for a computer to run more than one program at a time.
it's too bad video playing couldn't happen on one cpu while video compression happened on another.
someone should invent that. it could be called "Sametime Many Programs" or "SMP" for short.

Re:HTPC

[Neon Spiral Injector]

OK, I'll give you that. But the HD H.264 requires a huge ammount of CPU to decode. My current dual 1.6 GHz Opteron system can't do it in real time. Doesn't even come close.

So I was thinking the same thing about this new chip. It sounds pretty close to what I was wanting.

Re:HTPC

[Fweeky]

Use GPU-accelerated decoding, ala PureVideo. Not that I wouldn't also want the CPU power to do it too.

Re:HTPC

[eno2001]

What OS and software are you using? I'm able to do it fine on a P4 (with HT) that I bought in 2002 with Gentoo Linux. DV in from an 1080i Sony Handycam and it works great. I also can easily play back 1080i video files. I've got a WMF file in 1080i and it looks great and plays without hassle or even making the CPU break a sweat using Xine.

Re:HTPC

[MBGMorden]

You've got a point there. If you're watching Xvid bittorrent files any 1ghz machine will do, but my last computer upgrade (from an Athlon XP 2100 to a Sempron 3400) was done specifically because hte Athlon XP was choking down to a crawl on Apple's HD movie trailers. Even the 3400 stutters a bit on the really high res stuff, but it does ok on 720p so I don't worry about it too much.

Re:HTPC

[Tack]

GP is talking about h264.

Re:HTPC

[Neon Spiral Injector]

It is just this specific codec, any ffmpeg based player in either Linux or Windows just dies on 1080i H.264. 720p H.264 is fine, as is 1080i MPEG2. I also have some 1080p WMVs that play fine.

Re:HTPC

[tlhIngan]

Do you know of any video player that will be capable of taking advantage of two processors?

As far as I know mplayer doesn't, xine doesn't and vlc doesn't.

If VLC doesn't, then something VERY strange is going on.

In Windows, I use VLC to test out video playback (because it's the only way I can be sure that stuff like FairUse4WM and QTFairUse actually work!). I've decoded 1080p (1440x1080 - strangely, it displays properly on a 1920x1080 panel...) video that consumes about 18-30% CPU (via Windows Task Manager). I have quad proc (2x Core2Duo Xeons), and could swear there were three threads running simultaneously. (VLC actually spawns 18 threads!). I would think one was the OS taking data from VLC via DirectX and shoving it to the display driver to the screen (and managing everything else), while two were decoding video (~50-70% on one processor) and audio (~20% on another processor). The OS stuff seems to consume barely 5-10% of the third processor, and the fourth just remains idle. (At least someone had a clue in the XP scheduler...).

Using Windows Media Player and videocard accellerations, total CPU utilization barely tops 15% (usually less than 10%).

Ditto with QuickTime HD video - the QuickTime Player seems to use videocard accellerations to keep CPU utilization down. (Modern videocards often have H.264 and VC-1 (WMV9) accelleration blocks on them...)

Re:HTPC

[scottnews]

He is not talking about WMF. H.264 encoded files. My AthlonXP 2500+ can handle WMF. H.264 is a different story. The industry is barely getting smooth playback with Core 2 Duo x6800 and high end nVidia cards:

http://www.anandtech.com/video/showdoc.aspx?i=2890 &p=4

Re:HTPC

[NerveGas]

Maybe he wants to play PC video games on his HD big-screen. If I had an HD big-screen, I'd certainly play a few games on it. :-)

steve

Re:HTPC

[eno2001]

Hmmm... H.264 Quicktime files in Cinelerra seem to work OK from the DV I pull in. But I create them myself, so I can't say for non-Cinelerra generated content. And only Cinelerra seems to be able to play them back, not Xine. But that wouldn't seem to be a CPU issue. It seems more like a codec problem.

Re:HTPC

[pinkfloydhomer]

Eh, no. Ever heard of FFDShow? Lanzcos scaling and other scaling algorithms? Noise reduction algorithms? Sharpening algorithms? Proper MPEG decoding (without chroma bug) etc.? Of course, Avivo and PureVideo do exist, but not all like them (or use them, for instance under Linux). A setup as the one described above _will_ use a lot of CPU power. And that's only for DVD. Playing HD on an old, slow computer is not always, well... possible. /David

Re:HTPC

[Neon Spiral Injector]

Quicktime only seems to use a subset of the features of H.264. I can easily create videos that play fine with ffmpeg, but are a corrupted mess with the Quicktime player.

Re:HTPC

[ben there...]

My current dual 1.6 GHz Opteron system can't do it in real time. Doesn't even come close.

CoreAVC's requirements for 1080p24 are:
# 2.8 GHz Pentium 4 or faster processor
# At least 1GB of RAM
# 256MB or greater video card

So if you have a good video card, I don't see why your dual Opteron couldn't do it with CoreAVC. Quicktime is a different story though. But Quicktime has the worst performance of practically any H.264 player/decoder.

Re:HTPC

[ben there...]

Quicktime only seems to use a subset of the features of H.264.

Yeah, it's capable of the Baseline profile and partially supports the Main profile. Quicktime doesn't support any of the following:

• CABAC
• Bidirectional prediction
• Macroblock partitions
• Weighted prediction
• Deblocking

You can turn those off in Nero Recode's Standard-AVC profile to make a Quicktime compatible video, or follow this guide for encoding with x264.

Quicktime also obviously doesn't support High profile. A full list of the features it supports is here.

Re:HTPC

[pclminion]

Do you know of any video player that will be capable of taking advantage of two processors?

Kind of a funny question. The only reason a person would ask is if a single processor in their machine was too slow to play a video on its own. I've never heard of that. Otherwise, what's the point in using both processors to decode video? Only one processor is required, and the other processor of your SMP system will take care of any other processes that need to run. Splitting a task that requires less than 100% CPU between multiple CPUs is silly.

Imagine a task where you have to drag a 3 ton boulder, and you have two pickup trucks. Either truck is capable of hauling the rock on its own, but for some reason you decide to tie BOTH trucks to the boulder. Okay... Maybe there's less "strain" on the engines in that case, but why do it that way when you can drag the boulder with ONE truck and have the other truck completely available for some other purpose?

Re:HTPC

[zten]

Most of the people having problems with realtime H.264 software decoding are using ffmpeg, which simply doesn't do it well (yet). CoreAVC is a faster decoder, but non-free. The other option on Windows-based systems is to use hardware decoding with Radeon X1000 series or whatever the comparable nVidia part is, but I believe software support is limited to Windows Media Player for both the ATI and nVidia solutions.

Re:HTPC

[networkBoy]

3.4Ghz Prescott, 1Gb 533DDR2 no issue with h264 1080p decode, encode at about 20% of realtime.
one up:
PIII 550 512meg PC100 and a PCI Vertex FPGA, no issue with decode and encode 1080p both (just) at real time, wiggle the mouse and there may be jitter.
Next up
Spartan FPGA in a PCIe socket with a core2DuoEE with 4Gb ram, should be capable of 1080p encode at 4x realtime. (just need money :-)
-nB

Re:HTPC

[Spoke]

Do you know of any video player that will be capable of taking advantage of two processors? On the mythtv mailing lists, a number of people have reported better performance (smoother playback, fewer hiccups) when playing HD content when using a dual-core processor. Now, this isn't because the player itself can take advantage of multiple CPUs, but because the player uses a good amount of CPU and so does X. Having two cores lets you dedicate a processor to both processes giving you more headroom.

Having another core is especially important if it's mixed frontend/backend system where the backend may be recording other shows or doing commercial flagging or downloading program schedules or whatever else it may need to do.

So just because the player can't utilize multiple processors doesn't mean that the system won't benefit from having them.

Re:HTPC

[evilviper]

As far as I know mplayer doesn't, xine doesn't and vlc doesn't.

MPlayer does.

For any video that can be played by libavcodec (maybe 95% of them, including practically all the common HDTV formats, and otherwise CPU-intensive ones like WMV9 and H.264) you just need to set the -lavdopts threads= option.

Threads are also supported for encoding, though you inherently get some quality loss by encoding with seperate threads, so it's a trade-off, and I'd prefer to stick with one, faster core.

Re:HTPC

[evilviper]

The only reason a person would ask is if a single processor in their machine was too slow to play a video on its own. I've never heard of that.

Then you don't pay attention, and you've never heard about people working on highdef playback...

Your entire post is therefore moot.

Re:HTPC

[evilviper]

CoreAVC's requirements for 1080p24 are:
# 2.8 GHz Pentium 4 or faster processor

CoreAVC is cheating. Though nobody has figured out quite how, yet. It doesn't decode the h.264 videos nearly bit-exact, like other codecs do.

You can demonstrate this by comparing the checksum of h.264 video frames decoded with CoreAVC to the same video decoded by anything else.

It's safe to say CoreAVC is lower quality, as well as closed source, non-free, etc.

Re:HTPC

[elgatozorbas]

You are making fun of the GP without actually addressing his point, which imho is a good one. In case you just want to perform ONE task, are there players that exploit the presence of the two cores? I don't care of another cpu runs another job, I want them to cooperate. The algorithms need to be parallellised and will not just use two CPUs/cores just because they are there.

Re:HTPC

[pclminion]

Your entire post is therefore moot.

Uh, no. One would expect that people working on decoder software that requires multiple CPUs to run in realtime would probably develop it... for multiple CPUs. Not really the case the OP was talking about I think. The question was, why no multicore codecs for common video formats?

Re:HTPC

[networkBoy]

P4 saw 90%+ loading, pIII saw about 80% loading, but it was really timeslice sensitive, the FPGA, however was topped out, but since it was basically a symmetrical ASIC life was good.
-nB

Re:HTPC

[evilviper]

One would expect that people working on decoder software that requires multiple CPUs to run in realtime would probably develop it... for multiple CPUs.

That makes no sense what-so-ever.

There is no magic codec that requires X CPU time. As you change the resolution, bitrate, and encoding options, CPU requirements change dramatically.

The question was, why no multicore codecs for common video formats?

That isn't even remotely close to the question asked.

Re:HTPC

[aliquis]

Modern graphics card? Or doesn't they?

Re:HTPC

[MadBrassMan]

Under GNU/Linux using the Nvidia drivers, if you select "Sync To Blank" in the nvidia-settings GUI, you'll use a large percentage of one core for playing the video and much of the other for eliminating tearing.

Re:HTPC

[Neon Spiral Injector]

A checksum wouldn't be quite fair. Lossy compression format standards are often described as a bitstream, not as an encoder or decoder. So it is left open to interpretation as to how exactly the samples should be reconstructed. Any slightly different decoding by different program will yield vastly different checksums. So what CoreAVC is doing isn't exactly against the spec.

It would be interesting to see a visual difference of identical frames one decoded by ffmpeg and the other by CoreAVC. That would give a bit more of a clue as to where CoreAVC is skimping.

Re:HTPC

[evilviper]

So it is left open to interpretation as to how exactly the samples should be reconstructed.

That's not true with most modern codecs, and it's certainly not true with h.264.

So what CoreAVC is doing isn't exactly against the spec.

Quite the opposite. There's ample evidence CoreAVC is significantly going against spec:

http://forum.doom9.org/showpost.php?p=914265&postc ount=1

http://www.uploadtemple.com/view.php/1165615987.pn g

Re:HTPC

[pclminion]

There is no magic codec that requires X CPU time. As you change the resolution, bitrate, and encoding options, CPU requirements change dramatically.

Yes, but you can typically say with certainly whether a particular codec will be able to run with reasonable parameters on a single core or not. Can you do realtime decoding of 1080i video on a Pentium 300? Probably not, who cares.

That isn't even remotely close to the question asked.

Riiiight. Let's go back to the original post:

Do you know of any video player that will be capable of taking advantage of two processors? As far as I know mplayer doesn't, xine doesn't and vlc doesn't.

Let's see... He asks about multicore codecs... Examples of which include three common codecs, all of which are known to work fine on a reasonably modern single-core system. How is this not equivalent to what I said?

Re:HTPC

[evilviper]

Yes, but you can typically say with certainly whether a particular codec will be able to run with reasonable parameters on a single core or not.

"Reasonable" is entirely in the eye of the beholder.

The fact of the matter remains, just because most systems can't play back high resolution video on a single core, does not mean those writing it are going to make it threaded.

Examples of which include three common codecs, all of which are known to work fine on a reasonably modern single-core system.

WHAT THE HELL ARE YOU TALKING ABOUT?

He lists three programs which play videos. He does not mention specific codecs.

I have no idea where you get "known to work fine on a reasonably modern single-core system" from. How did you possibly determine that? How did you reach that conclusion?

It's a bit like saying "bicycles are know to be able to exceed the speed limit"...

Re:HTPC

[pclminion]

I have no idea where you get "known to work fine on a reasonably modern single-core system" from. How did you possibly determine that? How did you reach that conclusion?

Why, you're absolutely right. There is no proof that video players work. I've never seen a computer play video, nor have I ever met a person who has. Point conceded.

Re:HTPC

[evilviper]

Wow. I didn't realize I was talking to an insane person. My mistake.

It's a question of cores

[BadAnalogyGuy]

If you have only one core, you need to rely on the OS to not get in the way of running processes during task switches. With more than one core, processes can be split amongst the cores so that they do not need to be interrupted all the time by the OS timer interrupt handler. The more cores you have, the better you can scale up, even if the cores themselves are slower than a competing single core chip.

It's like driving down the highway in your train vs riding the rails in your Audi. Sure, you can try to drive the car on the train tracks for a while, but eventually the springs will break and your tires will pop and you end up walking to your final destination. But if you took the train, you'd probably tear up the road and it would take a while since you couldn't get much traction with the large metal wheels, but since you're carrying a whole lot of stuff in the train cars being pulled behind you, your bandwidth / time ratio is very favorable.

Re:It's a question of cores

[KingArthur10]

Concerning your analogy: I was thinking more along the lines that a train runs on a single track and sometimes has to be held up for another train to use the same track. They have some track switching, but most operations are serial. A car on the highway might not be allowed to go as fast as a train, but it's got four lanes to maneuver through. A bunch of cars will reach their destinations faster than a bunch of trains because the trains have to share single tracks often.

Re:It's a question of cores

[dodobh]

Which is why most places have two lines. And compared to a train carrying a thousand people, even the four lane highway is very inefficient.

Re:It's a question of cores

[Overly Critical Guy]

Take a look at his name before you respond to his analogy.

Re:It's a question of cores

[OneSmartFellow]

Except that the memory access is still the stumbling block. When each process gets dedicated memory, then we'll really get good performance, but, it seems that is not a high priority for most MoBo manufacturers these days

Re:It's a question of cores

[quizzicus]

Your ideas are intriguing to me and I wish to subscribe to your newsletter.

Re:It's a question of cores

[DaveWick79]

So if you can't buy gas, neither the train nor the cars move very quickly. It's starting to come together... :)

Re:It's a question of cores

[ultranova]

So if you can't buy gas, neither the train nor the cars move very quickly. It's starting to come together... :)

Except that trains use wood, electricity or diesel oil as their power source. I've never heard anyone suggest they'd run on gas. Or did you mean water vapour, AKA steam ? But steam cars weren't very fast...

Re:It's a question of cores

[Rod Beauvex]

I've seen one do 88... :D

Re:It's a question of cores

[drinkypoo]

There are gasoline-powered train engines, although they are usually restricted to doing pulling tasks in the yard.

Re:It's a question of cores

[pclminion]

Concerning the more serious first part of your post, it seems that ideally what you want to do is dedicate one CPU/core to interactive tasks, and another core for batch tasks. That way, the interactive tasks can easily interrupt each other as often as necessary on one CPU, while the other CPU cranks along on the batch tasks with a much longer time quantum without any unnecessary interruptions.

Re:It's a question of cores

[networkBoy]

Mostly because to get the same torque as a diesel they are gorssly inefficent. The flip side is that they don't need a half hour (or more in cold areas) to warm up to operating temperature.
-nB

Interesting..

[joshetc]

considering my 3800+ X2 runs at 2.8ghz with 1.5V. 2.9ghz really doesnt seem like much for a higher end model.. I'm thinking they will need at least 3.1ghz or so overclocks on air to have much of a chance in most highend enthusiast rigs.

Re:Interesting..

[gone9teen]

You do realize, well obviously you don't, the clock speed of a processor means nothing between different models when it comes to performance. A newer 2.0 Core 2 Due processor SMOKES my 3.0 Pentium 4.

Re:Interesting..

[joshetc]

Duh, all athlon 64 dual cores to date are clock for clock nearly identical though. This means clock speed does matter. I can't believe you got modded up for making such a shitty assumption on a "geek" website.

Re:Interesting..

[Anonymous Coward]

meanwhile core 2 duo systems continue to best AMD's offerings.....

Re:Interesting..

[Zaatxe]

I can't believe you got modded up for making such a shitty assumption on a "geek" website.

Maybe Slashdot's been attracting Digg's readers...

Re:Interesting..

[teg]

Duh, all athlon 64 dual cores to date are clock for clock nearly identical though. This means clock speed does matter.

They're almost identical - cache sizes vary, and, more importantly, the new ones (65 nm) have higher cache latency

.

As someone once said...

[Anonymous Coward]

A bad analogy is like a leaky screwdriver.

Re:As someone once said...

[rbanffy]

wish I had modpoints...

Overclocking potential should be increased.

[ZahnRosen]

I simply can't wait to get my hands on one of these and start overclocking. Getting them out the door at 2.9 GHZ should mean there's some overclocking head room left in the chips. Probably not too much, but I'll take what I can get.

Nice but a little slower. Surprise!

[IPFreely]

Anand has a nice review of these new processors, including performance comparisons.

The surprise is that it was a little slower than it's 90nm counterpart. They chased it down to the cache latency going up from 90nm to the 65nm part.

Other than that, it looks good.

Re:Nice but a little slower. Surprise!

[MrFlibbs]

According to the AnandTech article you referenced, saying that "it looks good" is a bit of an overstatement. Here are a few quotes from the article:

          "It's clear that these first 65nm chips, while lower power than their 90nm
          counterparts, aren't very good even by AMD's standards."

          "Performance and efficiency are still both Intel's fortes thanks to its Core 2
          lineup, and honestly the only reason to consider Brisbane is if you currently
          have a Socket-AM2 motherboard."

In every single AnandTech benchmark, Intel wins in both raw performance and performance per watt. And if raw power consumption is important to you, the winner was a 90nm AMD SFF part. In no case was a 65nm AMD better at anything.

The article does point out that a mature 90nm process is being compared to an immature 65nm process and thus future steppings are bound to be better. However, this doesn't change the fact that the current crop of AMD 65nm parts are a major disappointment.

Availability? In notebooks?

[Qwavel]

I'm thinking of buying a new notebook. When will these be available?

Lower heat (and performance, ....)

[IYagami]

...But most of the time irrelevant.

Anandtech has two good reviews here (lower power) and here (lower performance)

The main reason is the increase of L2 Cache Latency from 12 cycles to 20. But in most of the benchmarks the difference is very low.

All I know

[foniksonik]

Is that 90nm should be enough for anyone....

Re:All I know

[serialdogma]

AMD disagrees, they think that 90nm is a bit too much for everyone.

Re:All I know

[Courageous]

Mod +1, Attempting to Be Funny

C//

Good news, guys!

[Black Parrot]

Next time your class stud mentions his 9", you can counter by mentioning that your 6.5" consumes less power and gets the job done faster!

Re:I wouldn't overclock an AMD

[vslashg]

I've had bad experiences overclocking AMD processors (considering extended usage >6 months and at 100% load) with proper refrigeration. Unfortunately those processors tend do break (with no prior warning) eventually. What's a pity, since they overclock well.

They tend to break when overclocked, and yet they overclock "well"?

I don't think that word means what you think it means.

Re:I wouldn't overclock an AMD

[Com2Kid]

o_O

Nice assumptions there!

For the record, I used to up the vcore on my AMD at home to warm up my back computer room in the winter. A 300mhz speed boost helps as well. (It was a 700mhz Duron, OC'd to 1ghz in the winter, 900 in the summer)

I have always had good luck overclocking AMD chips, but I am running so many laptops now that I am living in an Intel world!

Obligitory

[Rik Sweeney]

Where's the 64 bit ver...

Oh wait

Re:SMP

[Anonymous Coward]

Whooosh....!

Take my advice....please

[OneSmartFellow]

If you are about to buy a AMD chip, ensure you buy a AM2 version, this is becuase non-AM2 versions do no support low level Hardware Virtualization (which means that XEN - and competitiors - can only operate in a paravirtualization mode)

Re:Take my advice....please

[Slashcrap]

If you are about to buy a AMD chip, ensure you buy a AM2 version, this is becuase non-AM2 versions do no support low level Hardware Virtualization (which means that XEN - and competitiors - can only operate in a paravirtualization mode)

Correct me if I'm wrong (and provide copious evidence since you're telling people how to spend their money), but isn't the current hardware virtualisation support in Intel and AMD chips generally considerably slower than paravirtualisation?

Short explanation, it doesn't virtualise everything. Long explanation, I can't be bothered with right now.

Re:Take my advice....please

[Courageous]

This is if you want to run Windows guests. Linux guests are best run paravirtualized for performance reasons. But point well-taken.

C//

Re:Take my advice....please

[TheRaven64]

True, but only as far as it goes. On x86, you have four protection rings. The hypervisor lives in ring 0, the kernel gets moved to ring 1, and the apps go in ring 3 as usual. When they designed x86-64, AMD 'helpfully' removed rings 1 and 2, so now the kernel and the apps have to share the same ring. They also removed the segmented memory model, so you have to use (more expensive) paged protection mechanisms to protect guests from each other. This makes paravirtualisation more expensive on 64-bit x86 systems than 32-bit ones. If you have a system that supports HVM, you can put the Hypervisor in the special hypervisor mode, use hardware-assistance for shadow page tables, and generally implement paravirtualisation more efficiently.

I can't remember if current releases of Xen do this, but if they don't then they definitely will in the next six months.

Re:Take my advice....please

[Courageous]

You know, I have to say: thank you for taken the time to write up this meaningful reply.

As it so happens, we at my company are in the midst of a giant virtualization study (and prototype), starting first with a large VI3 deployment, to follow shortly by Xen deployments and likewise some Virtuozzo. Your comment on paravirtualization perhaps not having the performance on x86-64 systems as might be expected was interesting enough that I'll now have something to pay close attention to when we get there early next year. Would the issue extend to I/O or it confined to compute?

(FYI, we are currently using Woodcrest systems with VT, and do not have any AMD-V systems planned, although I am aware of the nested page tables subject).

Current test facility is twenty 4 core Dell 1955 blades with 16 gigabytes each, 4 ports of GE on the back of each blade, fully interwoven (bonded/etherchannel) through CISCO 4948 switches.

C//

Re:Take my advice....please

[OneSmartFellow]

I don't make any claims about speed. However there are distinct 'advantages' to hardware (full) virtualization including the ability to run an unmodified Guest OS, and complete application isolation.

Re:I wouldn't overclock an AMD

[GotenXiao]

I've been running my S754 Athlon64 2800+ at 400MHz over spec (2200MHz, +/- 50MHz) for over two years. It still whips 3.2GHz P4s into shape.

Love the testing

[Dibblah]

Okay. Let's test a low power CPU. We need to stick it in a low-power board to get good measurements, of course. Let's ignore that we've got a 6150 with integrated graphics.Then let's stick on a big-ass 7950 which consumes over 70w on it's own at idle.

Is this a mistake in the article, or is this just... Insane?

Nice.

Re:Love the testing

[GodsMadClown]

They used the same video card on the Intel test rig too. They're just trying to keep as many components as possible in common between the platforms so that the power draw comparisons are more useful.

Not too complicated really. As to why they chose that particular video card, I don't know, but I'd wager that the reviewer just had it on hand.

Re:Love the testing

[lagfest]

Why is this insane? They are comparing it to other processors, and using the same platform so they can compare the results.

Re:Love the testing

[Dibblah]

It's insane when more than 50% of the power budget is going to the video card, not the CPU they're 'testing'.

Re:Love the testing

[pclminion]

Yeah, it seems like a silly way to do it. Why not just put a power sensor on the main ground pin for the CPU (or do they have multiple ground pins?) and measure the power draw of the CPU directly? Come on, these are overclockers and hardware hacker geeks, surely they can do something simple like that.

Re:Love the testing

[lagfest]

The power consumption of the system in idle state appears to be consistent to +-1%. We don't exactly need 6 decimals of accuracy here.

Re:Love the testing

[joshetc]

Because nobody uses JUST a CPU. One of the power saving features of AMD CPUs comes from the integrated memory controller which makes the motherboard not require one reducing overall power consumption.

Re:Love the testing

[jelle]

"(or do they have multiple ground pins?)"

Multiple? More than half the pins on ICs like such CPUs are usually for power, not just to keep the input power stable, but also to prevent 'ground bounce' where the ground goes up from 0v when the chip draws a peak current.

And, hum, 'just' add a 'power sensor'? You can measure strong currents with a coil around the wire, or smaller currents with a small resistor in-line (and then measuring the voltage over the resistor), but both influence the power flowing from/to the chip in an unwanted way, so that's not likely what you want.

Unless you have one of them 'James Brown' power sensors ('can you feel da powa').

something simple like that, yeah...

Re:Love the testing

[Dibblah]

For me, the interesting point from this testing is to figure out what a low power CPU consumes. I really don't want to have to guess what their big-bad-ass card is chewing up as a percentage of this, I'm only interested in processor and onboard graphics (for example, for server usage, where onboard is absolutely fine).

Adding in an unknown load for a graphics card doesn't help in knowing what the *processor* consumes, which is what the article pertains to be about.

65nm version marginally slower than 90nm version

[MajorJuggler]

Apparently Brisbane (65nm) has a 20-cycle L2 cache latency, vs. the 12-cycle latency from the 90nm versions. http://www.anandtech.com/cpuchipsets/showdoc.aspx? i=2893&p=3

OT

[Spaceman40]

We really need a "Whoosh" mod.

Re:I wouldn't overclock an AMD

[Anonymous Coward]

They tend to break when overclocked, and yet they overclock "well"?

I don't think that word means what you think it means.

Perhaps he means 'well' as opposed to 'rare'.

I prefer mine medium.

As I was once told...

[Valacosa]

Faster isn't always better. "Efficient" should not be the word one uses to describe your sexual performance.

Re:flamebait

[Zaatxe]

Nice to see some bigots with moderation points.
"Flamebait" != "I disagree"

But it seems by the modding you got that "Troll" == "I disagree and STFU!"

Re:I wouldn't overclock an AMD

[dtjohnson]

I overclocked the very first stepping of the first 500 Mhz Athlon to 700 Mhz and it has been running 24/7 for 6 years. Now, it is finally being replaced with a newer (AMD of course) system and I opened up the case for salvage parts and there...there was the overclocking board still attached to the 'Slot A' CPU and still working perfectly. I'd forgotten it was even there. So there's a 6-year data point on the overclocking/longevity scale.

Re:I wouldn't overclock an AMD

[TheRaven64]

It depends on the cooling a lot. At home, I used a 1GHz Athlon overclocked to 1.33GHz for about two years, before lending it to a friend, who used it for another three. To my knowledge, it still works, although he's recently upgraded so it's sitting in my hallway now.

Conversely, at work I have a 1GHz Athlon machine running 24/7, but rarely being taxed (it runs my Subversion server and a few other things. Sometimes I to timing tests on it, but not very often). It has got through 3 motherboards and 4 CPUs in three years.

The difference? My home machine had a CoolMaster sounds-like-a-jet-engine fan, which I later replaced with a near-silent Zalman cooler. The ones at work had supplied-by-the-lowest-bidder heatsinks and fans, on both the CPU and the chipset (biggest cause of motherboard death was the fan on the chipset dying and preventing air escaping).

Re:Sempron

[MBGMorden]

I research my purchases very well. No Athlon XP made can hold a candle to the Sempron 3400. It's in no way a crippled processor - it just has less L2 cache than the Athlon 64's (and if that's crippling then Athlon 64's a cripple compared to Athlon X2's because they have fewer cores. Both are crippled compared to Opteron's due to their lower L2 cache. Any particular processor is crippled compared to a processor with higher mhz, etc).

Bottom line, it was a significant upgrade for me for a low investment cost, and it's a good processor. The fact that it's not fast enough to play 1080p content doesn't mean it's broken - there are a whole lot of good processors out there (some Athlon 64's included) that can't play that video without skipping.

Re:Sempron

[Endo13]

Obvious troll, but I'll bite. The 32-bit Sempron is nothing more or less than the continuation of the old 32-bit Athlon XP CPU line, and Semprons carrying the same numerical designation as their old Athlon XP counterparts have exactly the same specs. Why they changed the name I'm not exactly sure, but it's still the exact same CPU. Celerons on the other hand are just Pentiums without most of the L2 cache, which makes them heavily crippled since the P4 with its long pipeline depends very much on their on-die cache.

A major disappointment for whom?

[John Jamieson]

Whenever AMD or Intel moves to a new process, they do not expect much from the first cores(they are happy if they get as many cores from a wafer as they did before-which if my sources are correct, Intel didn't do, and AMD has).
A lot of people forget that when Intel moved to 65nm, the new chips were slower in many ways, and the clock speeds were lower than the top end 90nm P4's.
By industry standards these AMD 65nm chips are a SUCCESS.

My only beef with the 65nm Athlons is that I cannot buy one at newegg, or order one from DELL. In my world, if I cannot order a PC with one, or buy it at newegg, IT IS A PAPER LAUNCH!

Where is the article on UNDERVOLTING?

[John Jamieson]

ALL the test sites seem to be preoccupied with overclocking and overvolting.

I cannot help but think there are enough of us interested in undervolting potential that someone should write a blurb about that.

35W Sempron

[zerOnIne]

These are nice, but I'm just trying to track down one of the new 35W Semprons that AMD makes (model # SDD*). Unfortunately, nobody carries them at retail. And unless I want to order them in packs of 12, or pay for shipping from Sweden, I seem to be completely out of luck.

Re:35W Sempron

[tindur]

Where do you get them in Sweden?

Re:35W Sempron

[zerOnIne]

A place called Dustin Home. Search for "sempron" and it'll come up, though I have no way of telling if it's in stock. The ones in question are marked "EE" for "Energy Efficient", with model numbers starting with "SDD" as opposed to "SDA". AMD has a nice chart on what the model numbers mean. I just wish that AMD would be a little more explicit in the names on the packaging.

A follow-up on L2 cache performance

[Dr. Damage]

I have posted an update to my initial look at AMD's 65nm processors here:

http://techreport.com/onearticle.x/11486

The update addresses some anomalies in L2 cache performance and raises some possibly related questions about die sizes for the 65nm Athlon 64 X2. It appears this chip is not just a die shrink with the same performance characteristics, after all.

Past mistakes

[Duncan3]

My current rig is an Athlon 64 3200+, 10k RPM drive, ATI 9800 Pro, 2GB of RAM.

There are 2 HUGE mistakes there. The CPU and the drive. Both are HOT, and hungry hungry for $power.

My next machine I'm looking exclusively at the dual core 35W CPU's, leaning a little to Intel over AMD. For the drive, I'll probably go for a SATA laptop drive, since by 10 min after booting, absolutely everything is in RAM anyway - turns out drive performance is 100% irrelevant.

The 9800 runs all the games I've playyed since (Lineage2, GuildWars, AutoAssault, WoW) at 30+ FPS, so it's good enough for anyone who doesn't live in their parents basement :) That said, go buy a X1900 and run Folding@home!

Finding data on low-power video cards has been a little trickier, noone seems to make anything under 1.21 jigawatts. Even the 9800 is hot stuff.

Sounds alot like a Mac Mini, so clearly Apple has figured out the same stuff, and that may just be the cheapest way to go once they move them to core 2 duo.

Going to take more than low power to draw me back-

[purduephotog]

I'm one of the fortunate 'many' that have had nothing but problems with AMD's x2 chips. Everywhere I turn I see the same problem listed and lamented about in forums- random total hard freezes with the X2 chips running a variety of mobos and configurations.

Some say to disable CnQ, others all USB (like THATS a fix nowadays), while still others recommend a full re-installation of everything and hope it works (doesn't).

I've swapped PSUs, memory, motherboard, drives, RAID cards, firewire cards, keyboard wedges (scanners), USB devices, disabled everything, and I can still get a lockup.

Moral of the story? I'm going back to intel. Yes, they're power hungry, more expensive, and rather not the company I'd like to support, but at least I never had a problem with their chipsets.

(And why are the crashes so annoying? 7 hours+ to rebuild a RAID-5 array everytime the computer freezes. And it still freezes if I remove all components and put in an old video card... but there is NO discernable pattern to it.)

Freeze examples:

Pressing the 'back' button on Firefox.
Getting up to go take a piss while downloading Ubuntu
Compressing a DVD
Ripping a DVD
writing a DVD
(yes I removed all DVD writers from the system and still got a lockup)
3 PSU's (Antec, 2x Ultras) no affect

Mind you I was totally stable when I had an AMD 3000+ chip- but I gave it to my parents when I decided to 'upgrade' and before I knew this hell was going to befall me. Had I known that, I would have shanked the whole system to ebay in parts and hoped they goto the four corners to never again see the light of day...

Conroe Duo Extreme, here I come when you drop to 118$/CPU.

Re:Going to take more than low power to draw me ba

[jsmethers]

I hear the same problems from people who assemble either an AMD or Intel based system. Either way, I suggest that since you do not know how to diagnose and repair such problems, you save yourself the trouble and purchase a premade system from someone who does.

Regardless, your problem is either:

Hardware:
* Crap power supplies - biggest problem I've personally seen lately.
* Crap memory - have not seem much of this lately.
* Under voltage memory - biggest problem I've seen complaints about DDR2. Lots of DDR2 memory is sold to auto configure voltage for an Intel system, but require 0.1 to 0.2 more volts for an AMD system. If your memory is less than 2.0 or 2.1 volts, go into the BIOS and set it to 2.0 volts. If that does not work, set it to 2.1 volts.
* Overheating - which is actually from improper assembly by "dumbass user" of heatsink and thermal grease on CPU.
* First revisions of hardware with known bugs -- requires proper drivers to be installed, which also comes down to "dumbass user" for purchasing the hardware in the first place and not being able to find the latest drivers.

Dumbass/Incompetent User:
* Install ALL necessary and latest drivers - motherboard, video, etc.
* Improper assembly
* Static shock to components
* Overclocking

Good luck. I'll buy your stuff for $100 if you cannot get it to work =).

Re:Going to take more than low power to draw me ba

[The_Dougster]

* Under voltage memory - biggest problem I've seen complaints about DDR2. Lots of DDR2 memory is sold to auto configure voltage for an Intel system, but require 0.1 to 0.2 more volts for an AMD system. If your memory is less than 2.0 or 2.1 volts, go into the BIOS and set it to 2.0 volts. If that does not work, set it to 2.1 volts.

Thats exactly what I had to do to make mine stable. I bumped up the DDR2 voltage a bit and haven't had a problem since.

Re:Sempron

[carl0ski]

It's been a long time since sempron was a byproduct of Athlon XP

Just at the time Athlon XP was being axed
the entire XP line was renamed and rebadged to Sempron.
they were identical


However new 32bit Semprons are based on the Athlon 64
they are slower and have less cache however the slowest k8 Sempron is faster than most Athlon XP predecessors

Hell now all Semprons are AMD64 (x64 x86_64) compatible

Re:Sempron

[Xabraxas]

Obvious troll, but I'll bite. The 32-bit Sempron is nothing more or less than the continuation of the old 32-bit Athlon XP CPU line, and Semprons carrying the same numerical designation as their old Athlon XP counterparts have exactly the same specs. Why they changed the name I'm not exactly sure, but it's still the exact same CPU. Celerons on the other hand are just Pentiums without most of the L2 cache, which makes them heavily crippled since the P4 with its long pipeline depends very much on their on-die cache.

Not exactly true. Some 32-bit Semprons were cut down Athlon64s, with 64-bit extensions disabled. Then there are 64-bit Semprons that are cut down Athlon64s (less cache) but without 64-bit castration.

When I was young...

[Jesus_666]

...we had 90cm and we liked it that way!

Re:Going to take more than low power to draw me ba

[purduephotog]

Hey, thanks for the offer- really. But as I tried to say, I have been as thorough as you can without owning a tech shop.

I have tested 2x different video cards. Still locks up randomly.
I have tested 3x different PSUs. Unless you have suddenly lumped Antec TRU supplies in with 'crap' I'd have to say that's rather unfair- An Ultra and a NEO as well.
All the drivers? Of course! And as I said everything *WAS* stable *UNTIL* I installed the X2 chip. From then on it's been down hill- with NO discernable pattern.

Memory is CORSAIR XMS 2GB, run at just a bit higher voltage- Oh yeah, it passes MemTest86 for 9 hours.
CPU? Not overclocked. Stock everything. Why? Because I need to get to a stable system.

Anyways, I do appreciate your comments. This is the 10th AMD system single chip I've built and the first 'dual core' system, but I'm not a stranger to dual-CPU configs. I know all about temperatures and cooling requirements- used to design and sell waterblocks back when you couldn't just place an order for a CNC machined one.

What it comes down to is that this particular chip, and the manufacturers that make the boards with them, have problems that can't be dropped into the 'dumbass' user slot. With that sort of hardware support from a mobo maker you'd wonder why the chipmaker is getting a shit response.

And with that.... I'm off to backup some more data before the system locks up again.

Re:Sempron

[tedgyz]

I couldn't have said it better.

Re:Sempron

[Endo13]

Fair enough. I neglected to mention the socket 754 Semprons that were only 32-bit. Other than that, my previous post is correct.

Re:Going to take more than low power to draw me ba

[WuphonsReach]

MemTest86 is no good for tracking down timing or voltage issues with "almost good" systems. It doesn't put anywhere near enough load on the CPU and other components while it does the memory testing. In general, MemTest86 will only catch completely bad memory modules.

If you *really* want to be sure of your memory (and system), run Prime95 in torture test mode. Exercising your disks and video at the same time is also recommended (while you check operating temperatures). You'll want to run for at least 4 hours, but preferably as long as 48 hours.

Re:I wouldn't overclock an AMD

[WuphonsReach]

The ones at work had supplied-by-the-lowest-bidder heatsinks and fans, on both the CPU and the chipset (biggest cause of motherboard death was the fan on the chipset dying and preventing air escaping).

Check out the Asus motherboards (M2N-E, M2N32-SLI) designs. They use a passive copper heatpipe / heatsink setup for the chipset cooling.

(I'm a big fan of fanless solutions on servers when I can get away with it. Such as the GeForce 6200 LE card.)

Re:Going to take more than low power to draw me ba

[jsmethers]

I have two Antec Smart Power 400 power supplies, dead after less than three months of use. Both protected by APC UPS's while in use. So, yes, I currently place Antec in the 'crap' power supply category.

You did not mention specifically what power supplies you used. Again, THIS IS THE BIGGEST PROGLEM I HAVE SEEN. You need to make sure that the power supply you use is rated at minium for the ATX v2.2 standard. You need to make sure that it provides enough wattage on each rail. This means at least 35w+ on the 3.3v, 35w+ on the 5v, and 18w+ on each 12v. You may need even more depending, e.g. lots of hard drives or SLI video cards.

Crap power supplies typically result in
* machine not powering on -- even if lights indicate power available
* machine sometimes powering on, only powering on speradically
* machine turning off almost immediatly after being turned on
* machine turning off spontaneously -- could also be an indication of other faulty system components.
* machine locking up -- also a good indication whether someone is using a UPS and/or has their refrigerator on the same circuit.

These issues typically indicate that the power supply cannot or can nolonger supply correct and necessary power to the system. The fail safe circuitry in many power supplies are typically responsible for the first four of these behaviours. This could be because of a faulty power supply not producing enough power, Really crappy power supplies are typically responsible for the last. i.e. they may end up frying your system.

You did not mention whether you are using a UPS. Some issues apply to it. Either a faulty UPS, or a faulty eletric utility grid.

You did not mention what motherboards you used, or whether they're DDR or DDR2 or 939 or AM2. The earlier 939 motherboard did not have BIOS which supported multicore chips. You did try flashing the motherboard to the latest BIOS, correct?

Also, early revions of BIOS in the AM2 motherboards had problems correctly detecting and configuring for the memory's SPD. You did try flashing the motherboard to the latest BIOS, correct?

You did not say anything about checking the voltage on the memory. If you are using DDR2, you did insure that the memory is set to 2.0v minimum, correct?

You did not say anything about the temperature of the processor. An AthlonX2 (both 89w and 65w versions) should give around 30*C CPU @ 15*C ambient temperature or 40*C @ 20*C ambient temperature when idle. You did install the monitoring software from the motherboard vendor, correct? Temperature probes are not always accurate, but they give a reasonable indicator of proper heat sink installation and system air flow.

Good luck.

Re:Sempron

[aliquis]

Not even close.

Sure the Sempron is a budget CPU from AMD, but it's not close to a Celeron from Intel.
Also he probably got the 64 bit Sempron, not the 32 bit version which is the budget version of Athlon-XP, I would guess a Sempron64 beats and Athlon-XP.

Fanboy/joke/whatever: The Sempron is more like a Pentium 4 Extreme Edition, but with lower power consumtion of course ;/

Marginal Voltage Increase = High power

[akuma(x86)]

You all realize that power scales with voltage CUBED right?

So 1.42 volts / 1.35 volts ~ 9%
But 9% cubed is about 1.09^3 = 30%

So 30% more power isn't exactly marginal.
Otherwise, CPU vendors would sell the chip at 1.42 volts.

Re:Going to take more than low power to draw me ba

[purduephotog]

Actually I'm one of those too that had the Antec "3 for 3" special- 3 replaced power supplies in 3 years. Each one of the failed ones were rather obvious as the 5V and 12V rails were very low- 4.3V and 11.2V, respectively. This one is cranking out 5.3V and 12.5V as measured with the Fluke. Again, all new PSUs which are pushing the requisite voltages (and I haven't seen any drops while monitoring.

CPU temps are all low- 40C, 43C under load (I did see a 45C once...) as measured with 'Core Temp' and MBM5 / SpeedFan.

UPS is an APC 1000 / Smart-UPS. Doubt its that.

This is socket 939- so DDR400/PC3200 ram. Bios was flashed when the crashes began to the latest version- and I didn't notice at the time but the ram timings were reduced to 'safe' values- 166 instead of 200- upping them back didn't help either. Memory is at 2.5V (would have to reboot to double check and the raid-5 is rebuilding).

The machine works just fine- it doesn't power off, it just 'pauses'. It's literally sitting next to a constant buzz and then all the associated clicks, whirls, beeps, and bumps suddenly stop, and you're left with just the sound of air flowing over components. Of course video still works so I can see my bloody mouse cursor sitting theredoing nothing.

But keep throwing out ideas. No help on the forums so far maybe you'll hit one I haven't looked thru. Or maybe I'll get another hard lock and I'll go back to the hard liquor....

Re:Going to take more than low power to draw me ba

[jsmethers]

If we can totally rule out power supplies and cpu temperature issues, I'd say it is a driver/hardware bug. Get the latest motherboard drivers direct from the chipset manufacturer, get the latest video drivers from the chipset manufacturer, and install the latest processor driver from AMD.

If you've got a Nvidia based motherboard, go directly to nvidia. If you've got a Via based motherboard, throw it away =). Same with Nvidia or ATI based video cards. Few people seem to know about the AMD windows processor driver.

You can run cpuz to see what revision chipset the motherboard has to check for known bugs in it.

You can run FreeBSD or like to get real output for recuring hardware issues that windows ignores/hides. If it boots without errors, put it in a repeat cycle of buildworlds.

At this point I'd need to get my hands on the hardware itself.

Re:Going to take more than low power to draw me ba

[purduephotog]

As I've said, latest drivers. Confirmed today, again.

So I got to spend the evening with my inlaws. 3 stiff drinks later and I was ready to have 5 more. But when I came home I can say, with certainty, that 6 hours of prime 95 on both cores showed no errors and no crashes. 49C/54C on the cores.

You did give me one idea I missed tho- I changed the SATA driver in the bios from 'ide' to 'sata'. I switched it back. No crashes yet. It's still early tho- I went 4 days without a lockup previously.

Stable drivers.... sorry... I love AMD. I lost 10K on Intel last year alone- I want the little guy to keep pounding them. But this isn't any way to win market share- this shouldn't be necessary.

Asus A8N board just arrived... decisions decisions....

Re:Going to take more than low power to draw me ba

[purduephotog]

*grin*

15 hours of prime 95 just fine. On both cores.

2 hours later I decided to check gmail- poof- down goes the computer.

Such a fun bug...

Re:Going to take more than low power to draw me ba

[jsmethers]

So, we can pretty much rule out a problem with the cpu and memory.

The variables in the problem are poorly specified. We do not know exactly what hardware you are using and have used to test with and in what combinations.

What power supply? What amprage on each rail (3.3v, 5v, 12v)? What ATX version?
What motherboard? What PCB revision? What chipset revision? What BIOS version?
What memory?
What hard drive? What chipset is on the PCB?
What addin cards?
etc.

What OS? What version? What patch level/service pack?
What drivers? Their versions?

Other possible issues:
Maxtor SATA hard drives had problems with nforce3 and nforce4 chipsets.
You did not attach the molex power connector on motherboard for your video card, and your video card requires it.
If you are using an nforce chipset, NVidia storage drivers may still be buggy. Do not install them.
Your motherboard may have bad capacitors.

BTW, use Orthos instead of prime95 for dual core stress testing:
http://sp2004.fre3.com/
http://sp2004.fre3.com/beta/beta2.htm

Like I said, I'd pretty much need to get my hands on the hardware itself to diagnose the problem at this point. Anyway, good luck and happy holidays.

Re:Sempron

[Xabraxas]

Fair enough. I neglected to mention the socket 754 Semprons that were only 32-bit. Other than that, my previous post is correct.

No it still isn't correct. There are at least three different Semprons.

1. Socket A Sempron. This Sempron is really just an AthlonXP.
2. Socket 754 Sempron. This is an Athlon64 not an AthlonXP, but it has 64 bit extensions disabled and has less cache.
3. Socket 939 Sempron. Cut down Athlon64 of the same socket.

I beleive there is now an AM2 Sempron also but I haven't read about the specs yet.

Wow

[d_54321]

Golly, a square processor chip measuring 65 nanometers by 65 nanometers. That's tiny.

How can you even see something that small, much less install it in your computer?