Istanbul EE launches today

[BaronMatrix]

Yep, the long-sought EE brand of Opteron has been released to day by the folks in Sunnyvale. It is said to operate at 1.8GHz with a 40W ACP\60W TDP.

It's supposedly called the 2419EE and should eb available pretty quick as several sites are reporting it. Check it out at fudzilla.

[Game_boy]

It's official.

http://blogs.amd.com/work/2009/08/30/ho ... an-you-go/
http://www.techreport.com/discussions.x/17502

Per a comment on TR, I'd like to see performance and power consumption figures versus the Xeon L5530. The Xeon is cheaper and has a 33% higher clockspeed versus the Opteron having 50% more cores.

[JF-AMD]

If you are going to compare to the L series, you REALLY need to look at the max power for the parts:

TDP Max power
L Series 60W 132W 1066MHz DDR3, 5.8GT/s QPI
L Series 38W 89W 800MHz DDR3, 4.86GT/s QPI

You can get to a 38W TDP on those parts, with the lowest speed memory and lowest speed QPI.

The EE parts have an ACP of 40W. Our max power is 60W.

FYI, in-flight wifi is pretty cool but slow. Remember when they told us that having wifi enabled would bring the plane crashing to the ground? Apparently the solution to that problem was to charge $12.95.

[muziqaz]

http://www.fudzilla.com/content/view/15286/35/

[BaronMatrix]

FYI, in-flight wifi is pretty cool but slow. Remember when they told us that having wifi enabled would bring the plane crashing to the ground? Apparently the solution to that problem was to charge $12.95.

That was funny. I mean how smart do you have to be to figure out that Power 96 FM doesn't interfere with Lite 105? Besides, I doubt if anyone would try to jump in on the signal from the plane they're FLYING IN.

[vsary6968]

If you are going to compare to the L series, you REALLY need to look at the max power for the parts:

TDP Max power
L Series 60W 132W 1066MHz DDR3, 5.8GT/s QPI
L Series 38W 89W 800MHz DDR3, 4.86GT/s QPI

You can get to a 38W TDP on those parts, with the lowest speed memory and lowest speed QPI.

The EE parts have an ACP of 40W. Our max power is 60W.

FYI, in-flight wifi is pretty cool but slow. Remember when they told us that having wifi enabled would bring the plane crashing to the ground? Apparently the solution to that problem was to charge $12.95.

Today 6 core Istanbul EE 2419 release was 1.8Ghz. Are their more higher clock speed higher than 1.8Ghz with 40W?

[abinstein]

If you are going to compare to the L series, you REALLY need to look at the max power for the parts:

TDP Max power
L Series 60W 132W 1066MHz DDR3, 5.8GT/s QPI
L Series 38W 89W 800MHz DDR3, 4.86GT/s QPI

This is very true. It is so sad that many people didn't have the ability to make simple comparison:

• AMD ACP ~= Intel TDP
• AMD TDP ~= Intel MaxPower

Also many people like to use "but Xeon xxx consumes only yyy watts in some zzz benchmark." As I pointed out in another thread, Intel actually patented a cheat-and-deceit tactics of power efficiency specifically for the benchmarks.

I happen to find it very dull to run the same program+data over and over in my servers. Maybe some people out there think it's useful?

BTW, JF-AMD, there's an inaccuracy in your slide (see below):

AMD_Intel_PowerFeatureComparison.jpg (170.52 KiB) Viewed 3952 times

Xeon L5530 does NOT have Turbo Boost. Xeon L5520 does.

[scientia]

When it was introduced back in 2003 the fastest Opteron was 1.8Ghz. So, 6 years later we have an Opteron that runs at the same speed but with 6 cores, large L3 cache, RAS features, Virtualization, and 128 bit SSE, yet it draws less power. I would also guess that the yield is better as well.

[JF-AMD]

Today 6 core Istanbul EE 2419 release was 1.8Ghz. Are their more higher clock speed higher than 1.8Ghz with 40W?

No, that is the product that we launched.

[Datsun]

Do you know about ACP? This is only marketing term that used by AMD. The processor VRM could get much power during its operations and exceed the power consumption even as designed with 65 watt TDP.

[Game_boy]

Do you know about ACP? This is only marketing term that used by AMD. The processor VRM could get much power during its operations and exceed the power consumption even as designed with 65 watt TDP.

If you're going to say that, it can exceed the TDP as well (just very rarely, and not under real workloads). The question is which is the most useful number, and the one people want is typical power consumption under 100% load for my kind of workload. ACP fits that better than either definition of TDP.

The real problem is people using TDP to say a chip consumes more or less than another. Even if AMD and Intel's definitions were consistent, TDP tells you almost nothing about the actual power it will use when in your machine. A given Pentium 4 may have the same TDP as a Core 2, but the Core 2 consumes a lot less on average.

[abinstein]

Do you know about ACP? This is only marketing term that used by AMD. The processor VRM could get much power during its operations and exceed the power consumption even as designed with 65 watt TDP.

Do you have any proof of whatever BS you just said?

According to AMD, ACP is measured by the power consumption of an array of CPU-intensive server benchmarks under maximum load to all cores. I'd say it is a useful parameter to estimate server power usage. It is very engineering in nature.

AMD TDP is the maximum amount of power possibly consumed by the processor. It is defined as max current values times max voltage value. Note however, the average current*voltage is always less than the max (any digital signal is a waveform alternating between ~0 and max), so there's no way that AMD TDP can be exceeded. The VRM may contribute to about 5% of the TDP.

OTOH, Intel's TDP is calculated in a similar way as AMD's ACP, so it's likely for Intel TDP be exceeded in real-world operations.

[JF-AMD]

ACP is 40W.

To get ACP we start with "hot" parts out of the fab. These will be parts that would draw more power and kick off more heat (assume they fall into the "worst case scenario" for parts a customer would see.

Then we take the 5 or 6 biggest server benchmarks that represent real workloads. We run these at 100% utilization and measure the power for every half second for a set period of time (several hours I believe.) We pull the average of the 100% utilization, not the average from 0-100%.

From that we get ACP.

TDP is the actual max power the part could ever consume. That is very different from our competitor who has a TDP and a max power (which they do not publish readily.)

Think of TDP as the top reading on the speedometer. ACP is how fast you drive to work when you are really late.

[Datsun]

Do you know about ACP? This is only marketing term that used by AMD. The processor VRM could get much power during its operations and exceed the power consumption even as designed with 65 watt TDP.

Do you have any proof of whatever BS you just said?

Yes, I am BS so every member in this forum as well. But, you know, every company have a marketing bureau so I think this is one of terms created by AMD's marketing despite it's more technical or more economical reasons.

[JF-AMD]

Actually ACP came from our customers. They were having issues because our old 95W TDP parts rarely drew more than 50W. So we had customers making decisions that our 95W TDP, which rarely ever got above 50W was the same as our competitor's 95W TDP. Dempsey, for instance, had a 95W TDP and A 174W MAX POWER.

If someone wants to say ACP is an imaginary number, it is no more imaginary than our competitor's TDP at this point. We release information on how ACP is calculated, has anyone seen how their TDP is calculted? They probably don't want that to be known because too many people think that is max today. It really isn't. By a long stretch.

[abinstein]

Yes, I am BS so every member in this forum as well.

It'd have made you look better if you've just admitted that you were wrong.

But, you know, every company have a marketing bureau so I think this is one of terms created by AMD's marketing despite it's more technical or more economical terms.

As JF-AMD said, the concept of ACP really came from AMD's customers. It is a very useful parameter for datacenters to estimate how much power & cooling they need. If you run a datacenter housing hundreds of racks with tens thousands of processors, you really don't want to spend 50% more cooling or allocate 50% more power budget to the floor. ACP helps these customers to easily estimate the power & cooling budget they need per CPU.

OTOH, if you want insist that "ACP" is a marketing term, then you should know that almost everything you hear are marketing terms. That'd include x86, IA-32, AMD64, HyperThreading, Turbo Boost, HyperTransport, QuickPath, Netburst, Nehalem, Barcelona, Bulldozer, Itanium, EPIC, ... What's important is not who came up with these terms (in many cases nobody knows or cares), but what are the concepts behind them.

[Datsun]

Yes, you are right, I have changed my previous post to more technical or economical reasons. My previous post have been corrected by JF-AMD after my post. My statement in that post is wrong so JF have corrected with his post as an AMD insider.

[The_Ghost]

Yes, you are right, I have changed my previous post to more technical or economical reasons. My previous post have been corrected by JF-AMD after my post. My statement in that post is wrong so JF have corrected with his post as an AMD insider.

wouldn't it be nice if we had some one from intel here that would rell us how they measure their TDP?

[JF-AMD]

I could tell you, but then I'd have to shoot you.

[T800-101]

I could tell you, but then I'd have to shoot you.

I might survive, and will tell the world of Intel's most guarded secrets.

[AussieFX]

I could tell you, but then I'd have to shoot you.

I might survive, and will tell the world of Intel's most guarded secrets.

I have a secret...
JF's gun is a water pistol.

[JF-AMD]

Well, having grown up in Chicago, and now living in Texas, you should be careful about questioning whether I am armed However, I'm not a big gun guy, so I guess you called my bluff.

As I remember it back a few years ago, the way they determine TDP is a de-rating from max power. Max power is the maximum amount of power that a processor can pull, if EVERY transistor is powered up. For us, that is TDP, for them, that is "max power". How do you test max power? There is a thing called a "thermal virus" that you can run with certain math libraries which powers up the processor and gives you the max power reading. The thing about the thermal virus is that nobody will EVER get to max power because that enviroment (running a thermal virus with the worst mathlibraries) will never occur.

So, then, you use a "de-rating" to reflect the delta between the amount of power the thermal virus. This de-rating is supposed to reflect what the "typical workload" would be. So, you can set your de-rating based on your "estimate" of the delta between the thermal virus and where you believe "actual" workloads are. That derating can change based on how hot your processors are. It used to be, back 7 or 8 years ago that the derating was ~80% or so. So, if you had a TDP of 95W, your max power was ~118W or so.

But, if your processors are too hot, then your TDP pushes up. But, you fix this by changing your re-rating. Instead of saying it is 80%, maybe it becomes 60%. Let's look at the progression downward:

So let's look at some intel processors, their TDP, max power and the approximate de-rating.

Prestonia 2GHz: 58W, 66W = 87%
Nocona 3.2GHz: 103W, 127W = 81%
Irwindale 3.2GHz: 110W, 146W = 75%

Xeon 5080: 130W, 174W = 74%
Xeon 5160: 65W, 130W = 50%
Xeon E5240: 65W, 99W = 65%
Xeon E5345: 80W, 134W = 59%
Xeon E5440: 80W, 133W = 60%
Xeon E5540: 80W, 173W = 46%

So, what did we learn today?

1. There is a delta between max power and TDP for Intel.
2. That delta continues to get bigger and bigger with each generation
3. The delta really began to deviate from the traditional levels about the same time that AMD hit the market.

I would love to have someone from Intel come here and give us th actual calculations, but that probably won't happen.

Discuss amongst yourselves.