AMD Bulldozer Core Patent Diagrams

Yeah but setting it all up is handier the more cores I have. Maybe hyperthreading would do it though. Then I could get it set up for 8 cores and the 4 instances will finish while the hyperthreaded instances will stall until the 4 are finished. Not sure if it would work like that or just be slow as tar.

 

For a quad-core SB with HT, run 8 instances at once. Each instance will take longer to complete than if you were to just run 4 at once, but it will finish all 8 faster than if you were to run two sets of 4 instances, usually about 20% faster. You can think of the 2600K as either a 4-core processor at 3.5 GHz or an 8-core processor at 2.1 GHz.

 

http://i152.photobucket.com/albums/s175/BallaTF/fcf67475.png

Apparently someone tested BD under PassMark and submitted a bunch of results in to their score database.
Download PerformanceTest 7.0 and see for yourself.

 

Heres comparison with my PII940

 

Looks good ... a lot faster than my Phenom "B50" quad core.

 

Lazy question: why is it winning in FP math and losing in INT math to the i5-2500? I would have (naively, ignorantly, etc.) thought the opposite.

Edit: Perhaps related to this 32-bit vs. 64-bit post.

Edit2: Thanks, fellix. Interesting. Taking DeF's PII X4 940 as baseline, that i5-2500K is 4x faster at INT math, and that FX-8150 is 3x faster (with 2x the INT ALUs).

 

From the description of the sub-tests in this benchmark, it appears that the FP portion is x87 only (single and double precision). Apparently, BD has retained some extra legacy FP processing capabilities along with the dedicated pair of MMX pipes, beside the main FMAC units.

 

Bulldozer has 2 128-bit FMA's. If both of those can do ether 80-bit x87 add or mul, one BD module can do either
1) MUL+MUL
2) MUL+ADD
3) ADD+ADD
in one clock cycle

Sandy bridge has 256-bit adder and 256-bit multiplier. If both can do one 80-bit x87 operation, one SB core can do just ADD+MUL in one clock cycle.

So bulldozer has "more freedom";in cases where the amount of mul's and add's is not equal bulldozer should do much better

 

According to that one submission from Passmark...

 

And my guess is that BD scores so low on INT because of front-end limitations and too low ILP?

 

Looks like "Piledriver" will save the day, but definitely won't be tomorrow.

 

Is that supposed to be a future platform with USB 2.0 and PCIe 2?

It's a good thing the x86 performance is projected to increase. I suppose that means AMD hasn't improved on Bulldozer's ARM performance.

Also, a VGA out on the southbridge?

 

Why should that save the day (assuming the slide is real)? If BD is really too slow a mere 10% increase is not going to help much (that is an increase in the same order SNB->IVB is going to be).
Apart from little tweaks (FMA3 etc.) the cpu seems to be nearly unchanged still with the same cache subsystem (minus the bugs, presumably). Though might be a decent improvement on the lower end (with IGP) over Llano hopefully.

 

Perhaps the thought (hope?) would be a 10% increase per-clock, along with better silicon yields so they can move the clock back where they originally wanted it?

The A8 Bulldozer Passmark result doesn't seem that bad, considering the pricepoint, however I'm not convinced that I would ever buy one for my desktop. I'm still far more convinced that I want the IVB version of i7-2600k, which is really what I'm saving all my pennies for right now

However, from my direct experience with an HP-manufactured Llano laptop, I'm quite happy with AMD in the mobile space.

 

Well a 10% clock increase along with a 10% per-clock increase would start to look better (if they can keep power consumption the same or lower). But given how unspecific that "10% improvement" is I've got some doubts that's really what is meant.

AMD will sell it at a competitive price point but it's not a good position they are in: much larger die which can only compete with the smaller intel chip if intel intentionally cripples it... Aside from obvious lower margins that probably also has some effect on power consumption (not necessarily on all products since intel doesn't exactly sell the chips at the lowest voltage possible but you get the idea). And from that slide I've unfortunately seen nothing which would indicate piledriver would fix that (a 20% performance improvement would not be enough it seems).

Yes at least it's got decent idle power consumption (and it's not too bad at full load power neither though at the cost of performance) and of course compared to SNB a kickass IGP. IVB will close the gap somewhat there though supposedly (at least I'd expect IVB IGP performance to be a little closer to Trinity than SNB was compared to Llano) but I guess Trinity will still make sense there. Though actually I'm not entirely sure a 4-core (2 module) Trinity will really be all that much faster on the cpu side than Llano, maybe AMD can get clocks up significantly (potentially with Turbo Core 3.0 when that finally actually starts doing something...).

 

Sure but Intel is talking about 50% power reduction with 22nm or 50% higher performance at same TDP.

 

Yes, I wonder if they can really pull it off (at the chip level). Could give IVB a really big advantage.
In any case it still looks like AMD has two problems:
1) the cpu design seems a bit too complex for the performance it achieves
2) they are way behind in manufacturing tech (well that's actually a GloFo problem but that doesn't matter...)

 

VGA is or was universal.
people do buy those 1366x768, cheapest displays, they are VGA only, monitor vendors still are saving a bit by putting one single interface and you'll have trouble finding a computer with DVI-D only.

I for one like to choose mobos with a parallel port, even. for ability to connect a laser printer that outlives most PCs, or just for the sake of knowing I can wire red LEDs right into it.

now boasting about USB 2 and PCIe 2 is ridiculous, I'll admit.

 

Yeah my Laserjet 4P has been used on about 15 PCs. PCL5 has had glorious longevity. So I like my parallel port, but I suppose I could get one of those USB adapters...

 

I rescued an old AXIS parallel print server device; it "bolts" directly to the LPT port on the printer and then converts to ethernet. I got it from my office about six years ago as they were throwing big piles of stuff away; it's an older version of this: http://www.axis.com/products/axis_office_basic/

Now I use it from every machine in the house