Posts by JF-AMD

Penguin Computing overclocks Opterons for Wall Street

Not really overclocked

Actually, they are not really overclocked. Those are special parts that we created (mainly) for the financial industry.

The AMD Opteron 4100 Series processors are designed mainly for web/cloud where low power is the key. We created a special part with a higher TDP specifically for these types of customer needs.

Penguin did a great job of identifying the types of customers where this makes sense and is targeting that solution specifically to them. But the clocks are all stock, nothing is overclocked. That means full warranty, no worry about early silicon death because of too much juice.

AMD laughs at Intel with Opteron Bulldozers

Actually, your math is wrong

That "12.8% improvement" has been debunked dozens of times on the web.

You can't do a "per core improvement" when looking at a fully utilized server. The performance numbers that we are quoting are for total throughput of 16 cores vs. 12 cores.

to give you an idea of how that scales, look at two examples:

AMD Opteron 4-core to 6-core - 50% more cores, 33% more throughput

Intel Xeon 4-core to 6-core - 50% more ccores, 33% more throughput

Compare to:

AMD Opteron 12-core to 16-core - 33% more cores, 50% more throughput

Obviously there is some architectural improvement that is boosting overall throughput. As for single threaded performance (the 12.8% often quoted), If you look at only 1 thread running on a 12-core and only 1 thread running on a 16-core, you will see significantly more performance increase than 12.8%.

When people try to make this comparison it is like trying to determine how long it will take you to get to the office at 3AM if it takes 30 minutes during rush hour. I can guarantee you that when traffic is lower at 3AM you can get there a lot quicker.

Wrong math

That "12.8% improvement" has been debunked dozens of times on the web.

You can't do a "per core improvement" when looking at a fully utilized server. The performance numbers that we are quoting are for total throughput of 16 cores vs. 12 cores.

to give you an idea of how that scales, look at two examples:

AMD Opteron 4-core to 6-core - 50% more cores, 33% more throughput

Intel Xeon 4-core to 6-core - 50% more ccores, 33% more throughput

Compare to:

AMD Opteron 12-core to 16-core - 33% more cores, 50% more throughput

Obviously there is some architectural improvement that is boosting overall throughput. As for single threaded performance (the 12.8% often quoted), If you look at only 1 thread running on a 12-core and only 1 thread running on a 16-core, you will see significantly more performance increase than 12.8%.

When people try to make this comparison it is like trying to determine how long it will take you to get to the office at 3AM if it takes 30 minutes during rush hour. I can guarantee you that when traffic is lower at 3AM you can get there a lot quicker.

IBM freshens cut-rate blade server

Virtualization anyone?

With a 1P-style workload that doesn't require a lot of resources, why would you buy a blade chassis and a bunch of 1P blades? Wouldn't a single server with virtualization be a better play?

Blades solve a density issue, but I can't see 1P workloads driving a need for blades unless a.) they were very ultra-dense blades (i.e. ~4U chassis height) or there was a legal requirement to have standalone hardware.

HP preps Tukwila servers for April 27

If you are going to be there....

There are actually 3 good sessions on AMD processors, BB20, BB21 and BB22. Of course that is if there isn't another volcano eruption to block my flight ;)

AMD draws x64 battle lines with 'Magny-Cours'

Looks good, actually.

If you look at SPEC Int, a 6172 (not top bin) against an X5680 (top bin) looks like this:

AMD

Performance: 386

Price: $1165

Power: 80W ACP

Intel

Performance: 381

Price: $1663

Power: 130W TDP

ACP and TDP are not direct comparisons. In the SPEC Power benchmark, at the wall at full utilization, a 95W TDP Intel X5570 was drawing more power than an 80W ACP AMD Opteron 6174. So I have to assume that the 130W TDP is over the 95W TDP.

For floating point, the performance delta was ~20% higher for AMD. This particular comparison, however, does not show the full performance, we went with a lower bin to show the value that customers will get. People are more concerned with the power and price than raw performance. This is borne out by the fact that only ~3-5% of the market buys top bin processors. 95-97% are looking for price/performance or perforance/watt.

MCM was OK, implementation was not.

We never said MCM was bad, we said "unconnected MCM" was bad. That creates off-package traffic that should stay on package between the two dies. Somehow the press mainly picked up "MCM is bad" and the rest of the story wasn't told.

Intel proved us right by dropping that design.

Now, the bigger questions is while all of that was happening, Paul Otellini was backing the strategy as a good idea, claiming better yield and manufacturability (2 claims that we acknowledged.) So, are they going to back down from that now?

AMD: OEMs primed for Opteron 6100s

wrong!

First, tier 1 support will be there at launch. Secondly, SuperMicro has been showing product since November. Thirdly, SuperMicro has been showing a 1p G34 for a while now.

Intel preps new Xeons for March 16 launch

I hope you like slow cooking ;)

If you are planning on turning a 4P server into a grill, good luck. At last week's analyst event we showed off the power consumption of our Magny Cours processors. Side by side, the 12 core was at about the same power consumption levels as the 6-core Istanbul at 20%, 60% and 100% utilization. It idle the 12-core was actually lower at the wall.

If you want to grill, that new 130W Beckton might be just the ticket. And with a memory buffer (like the old FB DIMM) soldered down on every memory channel as well, you can get a real good diner-style griddle going.

Hash browns anyone?