Sign in with
Sign up | Sign in
Your question

AMD X2 breaks 10000 in PCMark04

Last response: in CPUs
Share
June 17, 2005 6:52:34 PM

If you want to see some radical shiznit, take a look at this rig. AMDs Dual core is smoking!!!!!!! <A HREF="http://www.ocxtreme.org/forumenus/showthread.php?t=460&..." target="_new">linky</A>

<font color=green>NED FLANDERS FOR PRESIDENT</font color=green> Its justa nother gansta PARTY!
Intel P4 Extreme(3.73)@<font color=green>5.6Ghz</font color=green>
Asus P5AD2-E-Premium
Crucial Ballistix DDR2 667@DDR2 855
June 17, 2005 6:57:11 PM

Does that mean that you will give AMD a chance in your next upgrade?

<font color=blue>If everything seems to be going well, you obviously don't know what the hell is going on</font color=blue>
June 17, 2005 7:03:02 PM

The only reason that I havent gotten an AMD yet is because im waiting on their dual core to use DDR2. I have to much invested in DDR2 to revert back to DDR.

BTW I am setting up a new rig as we speak.

<font color=green>NED FLANDERS FOR PRESIDENT</font color=green> Its justa nother gansta PARTY!
Intel P4 Extreme(3.73)@<font color=green>5.6Ghz</font color=green>
Asus P5AD2-E-Premium
Crucial Ballistix DDR2 667@DDR2 855
Related resources
June 17, 2005 7:07:16 PM

I think he got his P4 to work pretty hard, and give any AMD a hard time.

______________
<font color=green>NED AND MOZZARTUSM - REAL (P)RESIDENTIAL CANDIDATES</font color=green>
June 17, 2005 7:09:11 PM

Wise choice.. I'm waiting for a new socket/memory before going to dual core. Current DC might be performing good, but cost too much, and not enough apps really benefit from it to really worth it right now.

<font color=red>Sig space for rent. make your offer.</font color=red>
June 17, 2005 7:23:36 PM

Yeah that's true

<font color=blue>If everything seems to be going well, you obviously don't know what the hell is going on</font color=blue>
June 17, 2005 7:27:45 PM

am i reading this correcrlyt.


-58c for the CPU? how the heck?!
June 17, 2005 7:43:48 PM

Phaze Change / Cascasde - One of those, used for extreme overclocking.

______________
<font color=green>NED AND MOZZARTUSM - REAL (P)RESIDENTIAL CANDIDATES</font color=green>
Anonymous
a b à CPUs
June 17, 2005 8:15:06 PM

Gotta love comments like this one
Quote:
you guys don't have any idea how HOT the x2 is, first time ever I see this cascade from -108c jump up to -89c when running first 2 set of test.

These guys dont live in the same world...well maybe in mozzartusm's world but not mine :wink:

Asus P4P800DX, P4C 2.6ghz@3.25ghz, 2X512 OCZ PC4000 3-4-4-8, Leadtek FX5900 w/ FX5950U bios@500/1000, 2X30gig Raid0
June 17, 2005 8:44:25 PM

lol no doubt!

F@H:
AMD: [64 3000+][2500+][2400+][2000+][1.3][366]
Intel: [X 3.0x3][P4 3.0x2][P4 2.4x5 down][P4 1.4]

"...and i'm not gay" RX8 -Greatest Quote of ALL Time
June 17, 2005 8:47:48 PM

I'm not impressed, and this should be amajor blow to everyone that are planning to buy X2.

IMO, the X2 runs far too fast, and this can cause instability for the user. Maybe AMD sould release a kernel patch to try and fix the problem, but this failure seems to be a real architectural flaw.

The Pentium D's are far more stable for an office environment, since it's not running that fast.

AMD seriously needs to recall all shipped X2 processors, and try and see if they can redesign the core. For now, stick with Pentium D.
June 17, 2005 8:49:30 PM

Lol, just kidding guys, this is great news. Also, I know how porky feels when posting such nonsense.

BTW, this is an overclocked a 4600+. I wonder what a overclocked 4400+ could do (because of the extra cache...)
June 17, 2005 9:07:15 PM

LMAO - I was literaly about to get you (re)banned thinking you was Porky until I saw your second post. :lol: 

______________
<font color=green>NED AND MOZZARTUSM - REAL (P)RESIDENTIAL CANDIDATES</font color=green>
June 17, 2005 9:19:41 PM

What? is porky banned? I came back to this forum after a short absence, hoping to be enlightened by his genious statements...
June 17, 2005 9:27:44 PM

LOL. I was cooking scrambled eggs on my head until I saw his second post. I went into BAN mode also.

BTW, im just about to install my new RAM. 1000Mhz

<font color=green>NED FLANDERS FOR PRESIDENT</font color=green> Its justa nother gansta PARTY!
Intel P4 Extreme(3.73)@<font color=green>5.6Ghz</font color=green>
Asus P5AD2-E-Premium
Crucial Ballistix DDR2 667@DDR2 855
June 17, 2005 9:38:48 PM

hmmm...scrambled.

Congrats on new memory. :cool:

______________
<font color=green>NED AND MOZZARTUSM - REAL (P)RESIDENTIAL CANDIDATES</font color=green>
June 17, 2005 9:41:25 PM

Yeah, congrats on new memory. Which brand? Also, keep us updated of the results :) 
June 17, 2005 9:47:05 PM

Corsair. I had maxed my other RAM out or at least I think it was. I should have this new rig going in a few hours so I will report back then. Later on guys!

<font color=green>NED FLANDERS FOR PRESIDENT</font color=green> Its justa nother gansta PARTY!
Intel P4 Extreme(3.73)@<font color=green>5.6Ghz</font color=green>
Asus P5AD2-E-Premium
Crucial Ballistix DDR2 667@DDR2 855
June 17, 2005 9:49:07 PM

Arent they picking up DDR2 in 06?

<font color=green>NED FLANDERS FOR PRESIDENT</font color=green> Its justa nother gansta PARTY!
Intel P4 Extreme(3.73)@<font color=green>5.6Ghz</font color=green>
Asus P5AD2-E-Premium
Crucial Ballistix DDR2 667@DDR2 855
June 17, 2005 9:53:14 PM

Cya, and good luck with the upgrade :) 
June 17, 2005 10:01:56 PM

Quote:
Let's see you daaance, sucka! You've got nothin' on me!
Let's see you...! Let's see you...!
Daaance, sucka!
You got served by Wuzy!

You aint from da hood!!!! Me and Ned are HARD CORE Gangstas! QUIT TRIPPIN before NED puts a CAP in YO AS5!

<font color=green>NED FLANDERS FOR PRESIDENT</font color=green> Its justa nother gansta PARTY!
Intel P4 Extreme(3.73)@<font color=green>5.6Ghz</font color=green>
Asus P5AD2-E-Premium
Crucial Ballistix DDR2 667@DDR2 855
June 17, 2005 10:18:57 PM

Please stay on the topic, we're here to discuss the far to high perofomence on the X2. I would really sack any IT professional that bought such systems if I was the boss of a company, just think about the reliability problems among the employees.

Really, I understand now how funny beeing a troll is. Maybe I should re-register as Porkster2 just for fun...

BTW, south park is best, I loved that episode...
June 17, 2005 10:26:20 PM

Quote:
Maybe I should re-register as Porkster2 just for fun...


OR...not.

______________
<font color=green>NED AND MOZZARTUSM - REAL (P)RESIDENTIAL CANDIDATES</font color=green>
June 17, 2005 11:45:28 PM

No. I've only ever got one person banned, and the entire forum, bar 1 or 2 asses, agreed.

______________
<font color=green>NED AND MOZZARTUSM - REAL (P)RESIDENTIAL CANDIDATES</font color=green>
June 18, 2005 8:28:25 PM

Im back with the new rig up and going. Old age must be getting to me, I have just now got the friggin thing going. Im as dumb as a stump when it comes to RAID configs. So what have I missed? Wusy, shagging sheep is not the answer im looking for. :tongue:

<font color=green>NED FLANDERS FOR PRESIDENT</font color=green> Its justa nother gansta PARTY!
Intel P4 Extreme(3.73)@<font color=green>5.6Ghz</font color=green>
Asus P5AD2-E-Premium
Crucial Ballistix DDR2 667@DDR2 855
June 20, 2005 3:39:03 AM

SHEEP FUK3R!

<font color=red>"Battling Gimps and Dimbulbs HERE at THGC"</font color=red>

"<font color=blue> Wusy</font color=blue> <-Professional sheep banger"
June 21, 2005 3:25:05 PM

AMD has a lot of good things for 06 and 07. AMD is going DDR2 in 06 with 2 differant sockets. The only problem is one is a socket 940 (not the same pin layout as the Opterons) and is for 1 way CPUs. I think they are going to strip out the extra HT busses to make room for the extra pins needed for DDR2 and better voltage control. Their other socket is something like 1207 and this will be for their 2-32 way cores. They are adding a 4th HT bus so the extra HT bus, DDR2, and better voltage control combine for a lot of pins.

Now when I talk about better voltage control, this will include a new level of Cool-n-Quiet for dual cores. This will let AMD control the speed and voltages of the cores seperately. Well that lets you run your main application on the main core at full speed, but still run the background apps and such on the second core and let the speed be dynamically adjusted on both cores depending on individual load. This lets the first core run at say 2.4Ghz and 1.3v and the second core run at 1Ghz at 1v. This is nice for a desktop but great for a laptop. Intel's Yonah when on battery power turns off the second core, but AMD can have complete control over both cores in all situations. Running an Athlon64 3500+ like mine at 1Ghz and 1.1v draws less than 10w likely and the new 65nm chips would likely draw much less at lower speeds. If they adjust C&Q to go down lower to say 600Mhz and say .8v, then the power savings would be even better. I would not be supprised if they could get it into the 2-3w range. You might think that you should just turn off the core, but the power draw is minimal at this point and this is actually enough to do something like surf the net or even play a DVD with something cranking away on the other core. They are currently estimating te power draw at 35w, but that is with whatever speed cores running at full speed. Likely this is for around 2Ghz CPUs. If they do add say a 3x multiplier instead of the 5x C&Q usues as the minimum, then the thing will generally idle at less than 10w and possibly as low as 5w.... with 2 cores.

Now what I am hoping is they push to make DDR3 use the same socket on the motherboard and have it so you can replace the CPU and memory to goto DDR3 without replacing the motherboard. I doubt it will happen this way. DDR3 uses less power than DDR2 and ramps up faster.

Now one thing we need to watch for is AMD is going to drop a Quad core in 06. This will be a mighty CPU. With 4 HT busses AMD could do a 32 way system without the latency going through the roof or using a propriator repeater that costs a fortune. Think of having 8 1U 4-way servers with 8 HT plugs on the back so each box can connect to 2 others to make a very impressive grid of 32 CPUs. Now think of each one using 8GB of dual channel DDR2 667-800Mhz. Now think of 4 cores with 2MB L2 cache for each core. Now run the cores at 2.4Ghz and you will have a HPC (high power computer) that would likely blow away anything on the market. That would be 128 cores with 128MB of cache, 256GB of system memory. All in a box about the size of 2 mid-tower case slapped together. Seeing as the server would be modular the costs would not be that bad. The company would only need to build one board and just run the traces from the HT busses to external plugs on the back of the board.Slap in 4 hot swappable drive using serial SCSI or SATA2 and redundant 700w power supplies and you are set. I bet this could be as cheap as $200,000... which seems like a lot, but would be a bargain compared to what say a 64 way Itanium2 (likely millions for one) would cost and would likely run 10 times faster.

One of the things on the rumor mill right now (that Anand agrees with) is AMD listed a slide show as Aero Glass Graphics here on slide 60 (and 61 on the next page):

http://epscontest.com/presentations/05q2_analyst-day.ht...

It might be a very basic 2D graphics engine, it might be an SSE4 for LongHorn's graphics engine, it might be a booster for HD-DVD playback... I don't know, but a lot of talk was going on at Computex. Anand Tech wrote up a little bit here:

http://www.anandtech.com/tradeshows/showdoc.aspx?i=2444...

The only thing is this is for 2008, but it talks about on core IGP and I/O functionality. I think it could be a lot earlier than that if they wanted. I'm guessing we will likely see something like this at least for the Turion sometime late next year (for the graphics at least, but likely not the I/O or sound). Having the entire chipset on die in 2008 would be a nice CPU. Think of what 4 cores at 45nm with everything on one chip (north and south bridge) add in a coproccessor, 2Ghz by-directional HT bus, Pacifica and Presido, PCI-e 2.0, and DDR3 then wrap it all up with 64 bit and multi-threading and you will have a mighty chip.

Now you might say integrated graphics just cant cut the mustard, but think of say an Radeon x300 with 1 pipeline. Normally this card is clocked at something like 400Mhz with 4 pipelines, but think of 1 pipeline a 2.4Ghz... that is 50% faster. Now on the memory side. Well they use something like 400Mhz 64 bit memeory... well this would have 1Ghz or faster DDR3 in dual channel for 128 bit memory That is 5x faster memory, so even using a small portion of it should still run VERY fast. Now think that it is built into the CPU so you have basically a super low latancy acces to the CPU and system memory. I would not be supprized if it would perform about in the range of a Radeon 9600Pro. Not a speed demon, but definately no slouch for something that is basically free.

Now what good is this... well a chipset adds a lot of power draw and takes up a lot of space on the motherboard. This would be nice for reducing cost and heat on a desktop, but VERY important for blade servers and laptops. At 45nm I would not be suppised to see a quad core at 2Ghz drawing only 35w and that is the CPU, GPU, and chipset rolled into one... for 35w.
June 21, 2005 4:19:56 PM

Wow, long post. Can I lend your chrystall ball though :)  ?
You seem to state as fact a lot of things that are certainly not confirmed, and that frankly, I haven't even heard rumours about. Most of them sound reasonable, but is your source anything else than ahem.. your ass ?

Lets go over them:

>The only problem is one is a socket 940 (not the same pin
>layout as the Opterons) and is for 1 way CPUs. I think they
>are going to strip out the extra HT busses to make room for
>the extra pins needed for DDR2 and better voltage control.

Agreed on the removal of the HT links, but I am unsure, does DDR2 actually require more pins ? I thought not, but could be wrong on that. If DDR2 doesnt need more pins, the additional pins are likely just for better power delivery.

>Now when I talk about better voltage control, this will
>include a new level of Cool-n-Quiet for dual cores. This
>will let AMD control the speed and voltages of the cores
>seperately

Source ? Sounds great, but is this true ? Thing is, I have *never* seen any symetrical multi processor system run its cores/cpu's at different clockspeeds. I assume there is a good reason for that.. Not that I see a fundamental reason why it couldn't work, but if it where easy, it would have been done. There has to be a catch...

>If they do add say a 3x multiplier instead of the 5x C&Q
>usues as the minimum, then the thing will generally idle at
>less than 10w and possibly as low as 5w.... with 2 cores.

At these low speeds, static leakage becomes a far greater factor than dynamic power, so you will get greatly dimishing returns. IOW, even if you'd clock the cpu at 50 MHz, it would not consume a whole less than at say 800 MHz. A 3x multiplier won't do nearly as much as improving clock and cache gating a la Banias/Dothan.

>Now what I am hoping is they push to make DDR3 use the same
>socket on the motherboard and have it so you can replace the
> CPU and memory to goto DDR3 without replacing the
>motherboard

Not bloody likely. That would require DDR3 to be electrically AND pin compatible with DDR2. I can only think of one time where you could use a newer kind of RAM into an older board, and that was FP RAM to EDO RAM (which brought next to nothing btw).

>Now one thing we need to watch for is AMD is going to drop a
> Quad core in 06

Another rumour I have heard as well, but won't take for fact until I see it. AMDs roadmaps peg this at 2007; do you have any other source expect for a vague reference to this on the INQ ?

>With 4 HT busses

Yet another scoop :)  Do you have a (credible) source ? It wouldn't be a bad idea I guess, but a chipset like Horus makes a lot more sense to keep motherboard design manageable for >4 sockets. With 4 HT links per cpu, routing that to 16 or 32 sockets is going to be a nightmare me thinks.

>Now run the cores at 2.4Ghz

A bit optimistic, no ? Current dual core Opterons max out at 2.2 GHz; doubling the cores AND increasing the clock seems a bit much to hope for, even if 65nm is on time, and provides as much an improvement as 130->90nm did. If indeed a quad core opteron arrives in 2006 (big IF), I wouldn't bet on anything above 1.8-2 GHz without letting go of AMDs 95W maximum.

>I bet this could be as cheap as $200,000... which seems like
> a lot, but would be a bargain compared to what say a 64 way
> Itanium2 (likely millions for one) would cost and would
>likely run 10 times faster.

10x faster ? :)  I think not..

>Aero Glass Graphics <..> It might be a very basic 2D
>graphics engine, it might be an
>SSE4 for LongHorn's graphics engine, it might be a booster
>for HD-DVD playback... I don't know, but a lot of talk was
>going on at Computex

Gee, I even missed that on those slides.. interesting.


= The views stated herein are my personal views, and not necessarily the views of my wife. =
June 21, 2005 5:19:08 PM

I state the rumor stuff or ones I am unsure of as such, but yes a lot is not confurmed...

Yes DDR2 uses more pins... I think it is about 40 pins more. HT uses about 30 pins per full HT bus if I recall correctly. I actaully e-mailed a few reveiw sites about Cool and Quiet on the X2 and 2 of them said that controlling the cores individially will have to wait for the next socket. It is also listed on slide 68 of AMD's pressentation as "Partitioned PowNow!

http://epscontest.com/presentations/05q2_analyst-day.ht...

That is pretty strait forward if you ask me. :D 

As for lower CPU speeds I did read an article that showed the benifits of downclocking the CPU and they did go quite low (800 at .8v):

http://www.pcstats.com/articleview.cfm?articleID=1798

The benifits on the Venice core deminished quite a bit after going below 2Ghz, but they were measuring temps not power draw. I have also heard the Venice core did add a 4x multiplier, but as far as I know C&Q only uses the 5x still.

As for DDR3 fitting in a DDR2 socket... I know it will likely not happen. As far as I know it could be though as it uses the same number of traces. The voltage is lower though, but that is something that the BIOS can control and possibly auto detect (based on the CPU or memory). The big problem is the problem of people plugging a DDR2 stick in a DDR3 slot. That alone is the main reason it will likely not happen.

Hum... running a Quad core at 2.4Ghz does not look to be much of an issue. The X2 is at 2.4Ghz alread and overclock in the same range as the current Venice core (2.6-2.8Ghz) so adding extra cores does not seem to limit the clock speed capabilities. The quad core would be a 65nm part and unless they have some major issues I think they can do 2.4Ghz pretty easy.

OK so I exagerated, but the Itanium system would likely cost millions and would likely still run slower and if you are talking standard 32 bit applications, then 10x is a possibility as it still does not run 32 bit native. Also seeing as AMD has 27% of the 4 way segment already, making a modular board based on a 4-way box is the best way. HT bus links can be made externally (as some boards already do this) and the developement costs are almost non-exsistant using them. That is what AMD made them for after all. Also the Opteron scales better than any other CPU out there and this only gives it more of an advantage for 4, 8, 16, and 32 way systems and dual (and more) cores are bringing a big performance boost. It will likely be the fastest non-cluster based computer out there.

They memtion a coproccesor, but do not say what it does. They mention the Aero Glass Graphics but do not say what it does. On slide 70 they list something called Thread-level Parallelism (TLP). This has several interesting things under it like:

Multiple programs running at the same time
Multiple OS's on virtualized hardware images
Multiple computers running the same program
Single application partitioned to run as multiple threads
The holy grail of computer science

I like that last one, but the one above it looks interesting too. It seams they will be able to slit a single threaded application up with hardware to run as multi-threaded. That could be very interesting if true.

Rumors.... well maybe. :D 
June 21, 2005 5:53:19 PM

> It is also listed on slide 68 of AMD's pressentation as
>"Partitioned PowNow!

Could also be that only the voltage is changed dynamically per core. If a core is iddling, it may not need as much power, but I agree, it seems likely dual core K8s will eventually clock each core differently.

>Hum... running a Quad core at 2.4Ghz does not look to be
>much of an issue

Except for power consumption and binning. On a mature 65nm process, I also don't expect this to be a problem, but on a brand new one, it would be quite an achievement. Current 2.4 GHz dual core chips are specced at 110W TDP, and draw ~80W real world power. Double that, than subtract the 65nm factor.. 65nm better be exceptionally good for this to happen, especially in 2006.

>Also seeing as AMD has 27% of the 4 way segment already,
>making a modular board based on a 4-way box is the best way.
> HT bus links can be made externally (as some boards already
> do this) and the developement costs are almost
>non-exsistant using them

Wrong. What you forget, is that current 4x2 setups (like Iwill) only need to route 2 HT links from one board to the other. When you move to 4 HT links per chip, the number of permutations grows insanely; Assume 4 sockets per board like you said; per daughterboard that is 16 HT links, 6 of which are used 'internally'. That leaves 12 links that need to be bridged with connectors. Worse, If you design an optimal mesh for 32S using 4 HT links per cpu, you will see that you'd need to connect those HT links to all different boards.. thats a nightmare from a design perspective, and the phyical distance might also be too large to bridge with HT.

Either way, as you can imagine, such a board that needs to route 16 HT links, 12 of which with connectors (versus 8 links, 2 of which with connector for current boards) will be considerably more expensive than a regular 4 way board, so there is no way it will be used for anything less than 8 way, so you won't even have the benefit of volume.

Either that, or # HT links between boards will be severely limited, and thereby, scalability will be far less than optimal. On top of that, even with all spare HT links 'connectored', I'm pretty sure only 2 boards will be linked at a time, meaning worst case memory access to a remote cpu is still an awefull large ammount of hops (like 1 onboard, 6 boards further, and then one on the remote board.. thats 8 hops each way.. not good for latency).

I still think a hierarchical solution like Horus makes a *lot* more sense for 8+ way.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
June 21, 2005 8:34:55 PM

I know they have 2x4 boards, but they link with 2 HT busses because each chip only has 3 HT busses. @ to link to each CPU and 1-2 to link to the chipset(s). With these new chips you take a standard 4-way board and 2 of the 4 HT busses on each chip will link to other CPUs and one of the chips needs to use more more to link to a chipset. That would leave 7 HT busses open. That would mean you would have 7 plugs on the back to connect to 7 racks. Each of the other racks will connect to other racks. It will make a web of links so you can get from any CPU to another without taking more than 4 steps. Now you could make it as a big plug and connect with a slab of PCBs with all the plugs. Designing simple PCBs like this is pretty cheap and strait forward.

Yes I know the Athlon64 X2 is rated for 110w, but it actually only draws less than 90w. Intel on the other hand rated their P-EE 840 at 130w and it actually draws 150w. To make this worse I ran across this chart on HardOCP. It was part of a pressentation from HP I think. It shows the Dempsey at 150w and the Tulsa at 175w. If this follows their current dual cores than you are looking at about 180w and 220w. Eeek! Even if those are the actual max power draw, that is still way too much. :p 

<A HREF="http://www.hardocp.com/image.html?image=MTExOTM1OTg0OUR..." target="_new">http://www.hardocp.com/image.html?image=MTExOTM1OTg0OUR...;/A>
June 21, 2005 10:15:18 PM

>Designing simple PCBs like this is pretty cheap and strait
>forward.

Ah, well, that is your assumption. I don't think there is anything simple let alone cheap about routing that many time critical paths on a PCB, using a protocol that isn't designed or specced for use over connectors, using variable length traces to several different daughterboards...

= The views stated herein are my personal views, and not necessarily the views of my wife. =
June 22, 2005 3:13:04 AM

I actually talked with someone on my old board about something else like this as he did this for a living and said it is actually pretty easy since the software does most of the work. The only think that would really complicate things is if the number of layer got rediculous, but it would not, or if the trace lengths had to match like it does for the Athlon64 memory traces, which it likely does not.

If I recall correctly the HT bus only has about 30 traces. At 7 cable x 8 servers, that would be about 1500 traces. Make it 200 wide and 8 layer deep... no big deal. Heck it would likely be only about 2" wide or so.
!