AMD unveils Opteron 6300 goodness

C

colinstu

2[H]4U
Joined
Oct 11, 2007
Messages
3,563
http://www.engadget.com/2012/11/05/amd-unveils-opteron-6300-hopes-to-put-servers-in-a-piledriver/

http://www.amd.com/us/press-releases/Pages/press-release-2012Nov5.aspx

AMD's advantage these days most often rests in datacenters that thrive on the chip designer's love of many-core processors, so it was almost surprising that the company brought its Piledriver architecture to the mainstream before turning to the server room. It's closing that gap now that the Opteron 6300 is here. The sequel to the 6200 fits into the same sockets and consumes the same energy as its ancestor, but speeds ahead through Piledriver's newer layout and instructions -- if you believe AMD, as much as 24 percent faster in one performance test, 40 percent in performance per watt and (naturally) a better deal for the money than Intel's Xeon. Whether that's true or just marketing bluster, there's a wide spread of chips that range from a quad-core, 3.5GHz example to a 16-core, 2.8GHz beast for massively parallel tasks. Cray, Dell, HP and others plan to boost their servers before long, although the surest proof of the 6300's success from our perspective may be that everything in the bacrkoom runs just as smoothly as it did yesterday.
Click to expand...

725H6.jpg


3.2GHz all-core turbo on that 16 "core" chip... pricey too but damn! Curious to see what these chips will bring to the folding table.
 
Last edited:
Not enough cores :rolleyes:
That HE with 16 cores might be the best deal but dropping used prices for high end MC chips is all we really need.
 
AMD has been saying to expect at best a 15% increase in performance over bulldozer.
 
When they get cheap they might be interesting, that is if SMP folding itself is interesting at that time.
 
Zink said:
Not enough cores :rolleyes:
That HE with 16 cores might be the best deal but dropping used prices for high end MC chips is all we really need.
Click to expand...
Yeah It's hard to get too excited about 16 fake cores when I'd probably rather have 12 real cores

that being said, with a 15% increase, these are getting more interesting
 
The 16 in BD is about on par with the 12 in MC. BD just used more power.

Now throw in 15% up.... with a QRB.... Very interesting.
 
As a point of reference, this chart compares average PPD on a few regular SMP projects between the FX-8350 and FX-8150.

FX-8350-89.jpg


These shown differences of +24%, +21% and +19%.

The catch is that these are achieved at stock clocks, where the FX-8150 has a full-load turbo of 3.9 and the FX-8350 has a 4.1 at full load. So if we account for clock speed, then the differences are actually +18%, +15% and +13% clock-for-clock.

So the 15% increase looks like it will be about right.
 
How does one get project 7020 units and similar? Can they be gotten through a linux vm? Does it need to be a v7 client?
 
Zink said:
Not enough cores :rolleyes:
That HE with 16 cores might be the best deal but dropping used prices for high end MC chips is all we really need.
Click to expand...

Ha! Thought the same thing in my little writeup today. I really want to see if I can make a 1U at 1A 110/120V server with one of those. The price is a bit high, but if you can put one of those in with 64GB of ram inexpensively.... might be interesting.
 
If the 15% performance increase was true, a stock Opteron 6386SE would do folding as fast as a Xeon E5-4650!
But I'm not so optimistic. PPD is not linearly dependent on performance, it should be converted into TPF first before we can compare the performance accurately, and this would reduce the difference. Furthermore, a bulldozer won't actually work at its all-core turbo in folding until we force it by a "tpc -psmax 1" cmd, so the real frequency of FX-8150 is unknown. Though the FX-8350 might not work at its all-core turbo either, but its stock frequency is much higher and close to its turbo frequency so the frequency loss would be smaller.

Dead Things said:
As a point of reference, this chart compares average PPD on a few regular SMP projects between the FX-8350 and FX-8150.

These shown differences of +24%, +21% and +19%.

The catch is that these are achieved at stock clocks, where the FX-8150 has a full-load turbo of 3.9 and the FX-8350 has a 4.1 at full load. So if we account for clock speed, then the differences are actually +18%, +15% and +13% clock-for-clock.

So the 15% increase looks like it will be about right.
Click to expand...
 
Last edited:
Agreed.

We need raw TPF comparison with chips operating continuously at single frequency.
 
Any signs of 63xx 'spicy' edition chips yet? I wonder if these will overclock even better than 62xx - so the real PPD increase with unlocked samples could be greater than 15%.
 
knopflerbruce said:
Any signs of 63xx 'spicy' edition chips yet? I wonder if these will overclock even better than 62xx - so the real PPD increase with unlocked samples could be greater than 15%.
Click to expand...

I don't see it happening any time soon.

We didn't see extra spicy MC's until IL's were close to launching. The same with extra spicy IL's, they didn't start showing up until BD was close to launch. So it all depends on where AMD goes in the future in the G34 server area.

I do agree with 402blownstroker and Zink, lower cost high end 6100 would be nice...so would lower cost MB's as Kendrak suggested. :D
 
Assuming tear is referring to this:

ebay dot com /itm/ws/eBayISAPI.dll?ViewItem&item=140900421214&item=140900421214
 
Must be one of the new ones... but i guess not so interesting when there was only one chip for sale. THe same guy has two 1.7Ghz MC engineering samples :)
 
So, has anyone gotten a hold of any 6300 series chips yet? Kinda tempted to look at upgrading my 2x6272 box but it has to be a pretty nice improvement to be worth it.
 
jebo_4jc said:
Yeah It's hard to get too excited about 16 fake cores when I'd probably rather have 12 real cores

that being said, with a 15% increase, these are getting more interesting
Click to expand...

Can someone explain to me what's fake about the 16 cores? I've noticed some references online to the 16 core AMDs not being "true" 16 cores, but I'm not if it's not real because it's two 8 cores slapped together ,or if it's some sort of quasi AMD HyperThreading thing.
 
tsuehpsyde said:
Can someone explain to me what's fake about the 16 cores? I've noticed some references online to the 16 core AMDs not being "true" 16 cores, but I'm not if it's not real because it's two 8 cores slapped together ,or if it's some sort of quasi AMD HyperThreading thing.
Click to expand...

There are 16 Integer cores and 8 Floating cores, there are 8 modules each has 2 Integer cores and 1 floating core and a load of cache.
Saying 16 cores or 8 cores is both wrong, it is just a different design.

Intel CPUs and older AMD CPUs had the same number of Integer cores as Floating cores.
 
Many pics on the interwebz:

1-amdsbulldoze.jpg


^^ single module, containing two CUs (compute units -- AMD's nomenclature);
  each CU is visible as one processor

There are 8 such modules in each of 6300 series.

FPU is double-wide. Initial AMD marketing suggested it would be able to perform like two
independent units (each at full-speed); we know today that's not exactly true... (for whatever
the reason).
 
tear said:
ZS258045TGG54
Click to expand...
Results from this unit:
Code: 
freq  nominal-voltage  temp  8102-tpf-1p  8102-tpf-4p-estimated
 3.5           1.1250    41       24m45s                  7m07s
 3.7           1.1875    43       23m31s                  6m46s
 3.8           1.2250    46       23m09s                  6m39s
 4.0           1.2875    54       22m04s                  6m21s

Board (GL) overregulates voltage by ~notch so 1.1250 is more like 1.1375, 1.1875 is more like 1.2000 and so on...
 
You must log in or register to reply here.
Back
Top