Maxtor storage

The New Amd Opteron Processors

Recommended Posts

How well do you guys think these chips will perform?

33W = 1.4GHz

50W = 2.0GHz

Do you think the 2.0GHz models will have the same performance as the old 2.0GHz Opterons? which use about 89W. In my opinion there might be a little performance loss of about 5~10%.

Share this post


Link to post
Share on other sites

Why would there be a performance hit?

My understanding is that the new chips benefit from an improved (refined) manufacturing process and (I would think) an improved chemical process

Otherwise it's the same circuits at the same clock speed.

Dogeared

8^)

Share this post


Link to post
Share on other sites

AMD has stated several times that the low-power Opterons operate at full speed.

There is no reason to doubt this, as DogEared points out, these low-power chips are simply a result of improved manufacturing processes combined with lot sorting by power consumption. They are identical in all respects to their more power hungry brothers except in their appetite for electricity.

I think if AMD can ramp up production of these types of chips they could crush the Xeon in the cluster, HPC areas.

These new 64-bit Xeons are a joke compared to the Opterons in any area where density is a concern. In a lot of situations, the whole point of using cheap x86 chips over other server solutions is that you can use a lot of them. You can't, practically, use a lot of Prescott Xeons in a dense environment. Unfortunately for Intel this is pretty much the entire 64-bit x86 market at the moment.

Do well.

Jonathan Guilbault.

Share this post


Link to post
Share on other sites

In fact, if one thinks about it, if Intel was going to spend the time to make an x86-64 compatible version of one of their processors, they should have used the Pentium M core. Such a chip would have been tremendously competitive in the relevant markets. It would have been very possibly superior.

It seems to me that they simply added x86-64 simply to quote it in the feature set of their Xeons without any consideration for targetting the chip to a market. I guess in workstations it could be useful, if irritating to workers in the volume department.

Do well.

Jonathan Guilbault.

Share this post


Link to post
Share on other sites

I think that the lower power consumption only means taht the chips are egetting better because they are more efficient at what they do, and less thermal losses.

Share this post


Link to post
Share on other sites

If the chip processes the same amount of information for less energy required, then yes, it would be more efficient! :)

Actually, there's a direct connection between how much a power a processor draws and how much heat a processor kicks out. Draw more power, kick out more heat.

Share this post


Link to post
Share on other sites

Forget servers and high end workstations for a second - will these have any impact on the quiet computing market? Low heat output without underclocking, on a powerful chip like a 2GHz Opteron... will these be a practical alternative for gamers on a noise budget, or will they cost too much in comparison to a regular chip with expensive quiet cooling?

Share this post


Link to post
Share on other sites
Forget servers and high end workstations for a second - will these have any impact on the quiet computing market? Low heat output without underclocking, on a powerful chip like a 2GHz Opteron... will these be a practical alternative for gamers on a noise budget, or will they cost too much in comparison to a regular chip with expensive quiet cooling?

Well, people are putting the P-4 Emergency Edition in notebooks today. A low power Opteron would get even more CPU horsepower at a lower power consumption, if AMD has a chipset fully up to the notebook task,

Share this post


Link to post
Share on other sites
Actually, there's a direct connection between how much a power a processor draws and how much heat a processor kicks out. Draw more power, kick out more heat.

Computers do almost no work in a Physics sense. If it consumes X watts input, then it produced X watts of heat.

Share this post


Link to post
Share on other sites
In fact, if one thinks about it, if Intel was going to spend the time to make an x86-64 compatible version of one of their processors, they should have used the Pentium M core.  Such a chip would have been tremendously competitive in the relevant markets.  It would have been very possibly superior.

The Pentium M processor hasn't shown many signs of scaling performance up so far. Is it possible it's already max'd out?

Yamhill supposedly really did exist and supposedly really was cancelled a couple of years ago. Suppose it was kept alive as a skunkworks project. Pentium-M was a skunkworks of sorts and wouldn't have had any Yamhill. Last year when Intel finally decides they must have AMD64 compatibility Yamhill may have looked much closer to reality on a Xeon platform than Pentium-M, to an executive anyway.

It seems to me that they simply added x86-64 simply to quote it in the feature set of their Xeons without any consideration for targetting the chip to a market.  I guess in workstations it could be useful, if irritating to workers in the volume department.

The announcement may have been a lot more important to keeping Dell in the house than actually shipping working silicon, at least in the short term.

In other words, Intel may not feel they need 64-bit Xeon but rather feel they need to slow down and inhibit customer defections to Opteron, which is a very different thing in the short term.

Share this post


Link to post
Share on other sites
The Pentium M processor hasn't shown many signs of scaling performance up so far.  Is it possible it's already max'd out?

Dothan (P-M v2) is coming out this April/ May... 2.0GHz + 2MB L2 Cache afaik

Share this post


Link to post
Share on other sites

Everyone,

We will see Pentium-M based CPU's with 2Meg of onboard L2 cache and clocking in around the 2Ghz mark very shortly. These are also rumoured to contain the ia32e and SSE3 extensions as well as a bunch of further optimizations on the archetecture to get branch prediction misses down and IPC up. Also they will move to the 800Mhz bus opposed to the current 400Mhz for the Pentium-M. These will use about 30W of power it is rumoured.

If a Pentium-M at 1.7Ghz with 1Meg L2 cache can match a P4-2.66 what will happen when you boost speed 20%, boost bus speed 100%, add 100% more L2 cache, and do a bunch of small changes that would help about 10% or so?

It's very likely that the Pentium-M will end up being the replacement to the P4 when they run out of headroom.

Add to this that the Pentium-M is moving to 0.09 micron very soon and dual core has been talked about and you have to wonder what Intel is really up to.

As for the LP AMD Opteron's they are nothing more then hand-picked out of current production lots and run at lower voltage then the average chip. But you pay heavily for this - they are $1500 for a 240 LP version. We have a few of the 240 LP's here in a very high density blade based cluster we're testing and they perform well - but are definately NOT cheap compared to paying $300 for a normal 240 90W hotplate.

SG

Share this post


Link to post
Share on other sites

We're testing out a Octigabay system that uses 12 LP 240's in a very small case.....

Check them out at www.octigabay.com -- they just got bought by CRAY.

Where I come in is the disk backend behind that HPC cluster (and others) - and it's a work of art!

We went with a Topspin IB to FC gateway which plugs right into their IB fabric. Something makes you wet when you see 3.1GB/s of sustained *random* I/O performance to a rack of 12 CPU's.

Too bad I don't have that sort of disk bandwidth at home -- Office would load real fast!!!

Rethinking that it's too bad I just don't have cash the disk backend costs -- I could buy myself a very nice Ferrari and Brittney Spears for that amount!

SG

Share this post


Link to post
Share on other sites
The Pentium M processor hasn't shown many signs of scaling performance up so far. Is it possible it's already max'd out?

Old laptops use to use fans to cool the processors. So desktops continued getting faster and laptop processors staying around 1ghz. Then breakthrough technology came through. The heat pipe. All current laptops have a heat place with thermal paste over the the processor. The heat is transfered through a pipe to a heatsink on the edge of the laptop with an attached fan blowing hot air out cooling the heatsink so more heat will transfer from the heat plate. It's possible the laptop processors are getting too hot again. My laptop which has an athlon xp 2000+ seems to run nice and cool though but that may be the problem with the mobile p4s.

Share this post


Link to post
Share on other sites

Pentium M wasn't designed to reach high clock frequencies unlike P4. IIRC there was a very good article on Ars or Ace's about the Pentium M.

Opteron doesn't produce 89W. The 89W is a design spec that all mainboards must adhere to.

Share this post


Link to post
Share on other sites

The Athlons will basically need that low a consumption for future purposes. Remember AMD has dual core K9s planned.

If each of the cores dish out 90Watts, you're gonna have some bad blood.

Share this post


Link to post
Share on other sites
As for the LP AMD Opteron's they are nothing more then hand-picked out of current production lots and run at lower voltage then the average chip.  But you pay heavily for this - they are $1500 for a 240 LP version.  We have a few of the 240 LP's here in a very high density blade based cluster we're testing and they perform well - but are definately NOT cheap compared to paying $300 for a normal 240 90W hotplate.

I can't confirm or deny if they are just bin splitting, but AMD did respin the processor before releasing these chips.

The other thing that must be mentioned it AMD doesn't say what the thermal properties of a 240 is. They have said that max thermal envelope for the processor family is ~80W. AKA this is what the cases and heatsinks must handle. Kevin McGrath in this presentation presentation sad they respin every 6 months and perf is higher and power lower.

They garentee the lower power on the LP Opteron, but we don't know how much lower that is than a 240.

Share this post


Link to post
Share on other sites
We will see Pentium-M based CPU's with 2Meg of onboard L2 cache and clocking in around the 2Ghz mark very shortly.  These are also rumoured to contain the ia32e and SSE3 extensions as well as a bunch of further optimizations on the archetecture to get branch prediction misses down and IPC up.  Also they will move to the 800Mhz bus opposed to the current 400Mhz for the Pentium-M.  These will use about 30W of power it is rumoured.

If a Pentium-M at 1.7Ghz with 1Meg L2 cache can match a P4-2.66 what will happen when you boost speed 20%, boost bus speed 100%, add 100% more L2 cache, and do a bunch of small changes that would help about 10% or so?

It's very likely that the Pentium-M will end up being the replacement to the P4 when they run out of headroom.

Add to this that the Pentium-M is moving to 0.09 micron very soon and dual core has been talked about and you have to wonder what Intel is really up to.

The intriguing thing here is that I have always been told that the Pentium M was a skunkworks project of sorts, not part of the corporate plan. After it showed spectacular results it was formally presented to corporate management, far past the point where entrenched US designers could kill it as a rogue project.

If true it does not speak well for Intel management that there are depending on others for the architectures they ship this year (AMD64 and Pentium M).

As for the LP AMD Opteron's they are nothing more then hand-picked out of current production lots and run at lower voltage then the average chip.  But you pay heavily for this - they are $1500 for a 240 LP version.  We have a few of the 240 LP's here in a very high density blade based cluster we're testing and they perform well - but are definately NOT cheap compared to paying $300 for a normal 240 90W hotplate.

Yikes! On the other hand, perhaps it says AMD has very high confidence in their 90nm process - they expect to get these results into the future, which means without going to such extreme sorting.

Share this post


Link to post
Share on other sites

I still don't get it, how did they manage to lower power concumption by that much? the transsistos are still in 0.13 micron "SOI" just like the old high power Opterons :huh:

Unless if they used different materials to build the chip?

I've been too busy to read everything about them so I'm behind their new improvements, i should get up to date by tonight.

Share this post


Link to post
Share on other sites

Maybe they are measuring by the minimum speed of the chip. For example my athlon xp 2000+ mobile can run at 933mhz, 1.16ghz and the full 1.67ghz. Maybe they were testing with the lower clock enabled.

Share this post


Link to post
Share on other sites
I still don't get it, how did they manage to lower power concumption by that much? the transsistos are still in 0.13 micron "SOI" just like the old high power Opterons  :huh:

Unless if they used different materials to build the chip?

I've been too busy to read everything about them so I'm behind their new improvements, i should get up to date by tonight.

Transistors vary quite a bit.

The fastest ones use alot of power, slower ones use much less. Only a small subset of the transitors limit the chips performance. This is the "critial path." If you do perf and power optimizations on a core you can save alot. Initial designs typically worry more about yield and correctness. You also shut down the sections that are not being used.

One thing that did is they shut down almost the entire chip when it is halted. Check out the vid for more info.

SOI can get wonderful power vs perf numbers if your transistor design is correct.

Share this post


Link to post
Share on other sites
Maybe they are measuring by the minimum speed of the chip.  For example my athlon xp 2000+ mobile can run at 933mhz, 1.16ghz and the full 1.67ghz.  Maybe they were testing with the lower clock enabled.

No this it at the full rated clock speed.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now