[VRNews]

AMD have revealed a further key details regarding their next-generation processors. AMD's aging Stars architecture, though competitive by aggressive pricing, desperately needs a revamp.

Much delayed, and much anticipated, Bulldozer has finally taped out and will follow K10/Stars which itself was much delayed to 2007. Bulldozer will be fabricated at Globalfoundries 32nm, and is set to sample in 2010. The time from tape out to general release for CPUs is about 9 to 12 months, so Bulldozer is well on its way to a 2011 release. Meanwhile, AMD Ontario is set to release early - as soon as Q4 2010.

More next page.


Read More:
AMD Bulldozer tapes out, Ontario in Q4 2010

Share|

[raysusan]

great news....

[XiaoDongDong]

great name... lol... pray it bulldozed others and wont get bulldozed! Nice way to go!

[quantumfusion]

good . lets hope it can make profit for Q3 !

[riggnix]

i'm pretty curious about ontario... if it's as good as promised it'd be a nice engine for a media center PC

[power666]

As expected, Bulldozer is going to be released in the first half of 2011. I just hope it lives up to expectations in terms of performance.

[przemyslaw]

AMD Bulldozer tapes out, Ontario in Q4 2010 | VR-Zone | Gadgets | PC Enthusiasts

[psolord]

Originally Posted by power666

As expected, Bulldozer is going to be released in the first half of 2011. I just hope it lives up to expectations in terms of performance.

Yeah me too.

Although AMD's offerings are more than enough for home/office use and they offer truly great price/performance ratio, the not so good implementation of multithreading, is not doing them much good. AMD's performance per thread has quite some room for improvement. Their power consumption can go down as well.

So, although I own three Intel PCs, I really hope that AMD can bring some much wanted competition in the market.

My best wish, is that the small octacore model, will be 50% faster than the 980X! To explain it better, I'd like 25% more performance per clock per Mhz in addition to the 25% more cores. That 50% would only be possible in something like Cinebench, but this would suffice for me.

[power666]

Originally Posted by psolord

Although AMD's offerings are more than enough for home/office use and they offer truly great price/performance ratio, the not so good implementation of multithreading, is not doing them much good. AMD's performance per thread has quite some room for improvement. Their power consumption can go down as well.

AMD doesn't currently have multithreading and Bulldozer introduces a design that is neither pure symmetric multithreading or pure multi-core. AMD's approach does look like it'll be faster than a pure multithreading solution as it is giving dedicated execution hardware and cache to each thread. The real variable here is how fast is a Bulldozer module running a single thread? I predict that Bulldozer will be faster per thread than current Phenoms but the real comparison is against Intel's Sandybridge.

Originally Posted by psolord

My best wish, is that the small octacore model, will be 50% faster than the 980X! To explain it better, I'd like 25% more performance per clock per Mhz in addition to the 25% more cores. That 50% would only be possible in something like Cinebench, but this would suffice for me.

I'm hoping for a good increase in performance per core per clock alongside a decent increase in clock speeds. The PC processors really do need those attributes to change. With the current software base, there are limits to how much performance can be extracted from additional cores. I'd love to see AMD ship chips past 4 Ghz at stock speeds.

[NextGen_Gamer]

Originally Posted by power666

AMD doesn't currently have multithreading and Bulldozer introduces a design that is neither pure symmetric multithreading or pure multi-core. AMD's approach does look like it'll be faster than a pure multithreading solution as it is giving dedicated execution hardware and cache to each thread. The real variable here is how fast is a Bulldozer module running a single thread? I predict that Bulldozer will be faster per thread than current Phenoms but the real comparison is against Intel's Sandybridge.

What I am wondering though is what AMD is referring to when it says "Zambezi" will be an eight-core "Bulldozer" part. Does that mean it will have eight modules, each of which has two execution engines, for a total of 16 threads? Or does AMD consider each execution engine to be its own core, which would mean an 8-core "Bulldozer" part has only four modules for 8 threads total? If it's the latter, than "Sandy Bridge" has already won...

[power666]

Originally Posted by NextGen_Gamer

What I am wondering though is what AMD is referring to when it says "Zambezi" will be an eight-core "Bulldozer" part.

Anandtech asked that very question and the answer was that what AMD will market as an 8 core chip will contain four modules. Earlier road maps have this chip being released for socket AM3 for consumers (I'm predicting that it'll see release with socket FS1 which the Llano chip will introduce). AMD's socket G34 solution will contain two dies containing four modules for a grand total of 16 simultaneous threads. If AMD follows that same pricing structure as Magny-Cours, then the socket G34 solution could be affordable for the enthusiast segment.

I wouldn't say that Sandybridge has won yet. Sandybridge's own performance is a bit of a variable. Early performance figures were unimpressive. I was hoping for a bit of an increase in performance per clock but that hasn't materialized. The workstation/server variants of Sandybridge may boost performance due to wider/faster memory buses.

[haylui]

Originally Posted by psolord

Yeah me too.

Although AMD's offerings are more than enough for home/office use and they offer truly great price/performance ratio, the not so good implementation of multithreading, is not doing them much good. AMD's performance per thread has quite some room for improvement. Their power consumption can go down as well.

So, although I own three Intel PCs, I really hope that AMD can bring some much wanted competition in the market.

My best wish, is that the small octacore model, will be 50% faster than the 980X! To explain it better, I'd like 25% more performance per clock per Mhz in addition to the 25% more cores. That 50% would only be possible in something like Cinebench, but this would suffice for me.

don't put so much expectation to bulldozer architecture
even Intel not able to do 25% IPC from generation to generation(forget about P4 to Core transistion pls)
but if you need FP performance badly, then BD might be able to live up to your expectation

[psolord]

Originally Posted by haylui

don't put so much expectation to bulldozer architecture
even Intel not able to do 25% IPC from generation to generation(forget about P4 to Core transistion pls)
but if you need FP performance badly, then BD might be able to live up to your expectation

Why not mate?

If you look at this sum up of this huge cpu review, you will see that the Q9550 gets an average rating of 125.5 while the equivalent in Mhz, the Core i7 860, gets a rating of 165.8. That's 32% faster, although granted, the i7 has the Turbo mode to help it as well as HyperThreading.

If you want to take HyperThreading out of the picture, you can compare the Q9400 with the Core i5-750, but the 750 still holds the advantage of its turbo mode. Still the scores are 115.7 for the 9400 and 145.30 for the 750. A healthy 25.6% difference again.

If you look at AMD's results, direct comparisons get to be just as tricky if not even more so. The problem with AMD is that are stuck with K10 architecture too long now and K10.5 is not really a step forward in terms or processing advantages. So the correct way to do a comparison is a K8 with a K10 CPU. So if you compare an Athlon X2 6000+ (3.0Ghz) with an Athlon 2 X2 250 (3.0 Ghz), you will also see a nice 22% advantage. The problem here is that the Regor core used for the 250, is completely deprived of its L3 caches, thus losing some of its next gen edge. With its L3 present, it would be easily another 10% faster.

See how I didn't compare, say, the Q9400 to the Q6700 since they are essentially based on the same architecture.

Now the problem is that asked 25% over the i7 architecture, which seems kind of too much but as I said, AMD has been stuck with K10 performance level for like three years now and I believe it's time for them to deliver.

[haylui]

Originally Posted by psolord

Why not mate?

If you look at this sum up of this huge cpu review, you will see that the Q9550 gets an average rating of 125.5 while the equivalent in Mhz, the Core i7 860, gets a rating of 165.8. That's 32% faster, although granted, the i7 has the Turbo mode to help it as well as HyperThreading.

If you want to take HyperThreading out of the picture, you can compare the Q9400 with the Core i5-750, but the 750 still holds the advantage of its turbo mode. Still the scores are 115.7 for the 9400 and 145.30 for the 750. A healthy 25.6% difference again.

If you look at AMD's results, direct comparisons get to be just as tricky if not even more so. The problem with AMD is that are stuck with K10 architecture too long now and K10.5 is not really a step forward in terms or processing advantages. So the correct way to do a comparison is a K8 with a K10 CPU. So if you compare an Athlon X2 6000+ (3.0Ghz) with an Athlon 2 X2 250 (3.0 Ghz), you will also see a nice 22% advantage. The problem here is that the Regor core used for the 250, is completely deprived of its L3 caches, thus losing some of its next gen edge. With its L3 present, it would be easily another 10% faster.

See how I didn't compare, say, the Q9400 to the Q6700 since they are essentially based on the same architecture.

Now the problem is that asked 25% over the i7 architecture, which seems kind of too much but as I said, AMD has been stuck with K10 performance level for like three years now and I believe it's time for them to deliver.

BD supports AVX instructions set, FP performance at 25% gain over K10.5 I believe is not a problem.
But INT performance increase I'd remain conservative due to current design. Throughput and multi-threaded performance are BD design direction.
In engineering, everything and anything is related to cost. AMD could be as good as Intel if they have financial strength as Intel. But sadly, reality is AMD is just 1/10 of the size of Intel; on the flip side(i know shouldn't be compare it this way) AMD is just lagging <25% behind...lol

[power666]

Originally Posted by haylui

BD supports AVX instructions set, FP performance at 25% gain over K10.5 I believe is not a problem.
But INT performance increase I'd remain conservative due to current design. Throughput and multi-threaded performance are BD design direction.

Bulldozer has the ability to execute four integer instructions per clock with up to an additional 4 instructions via the shared FPU per thread. The K10.5 core is limited to a total of three instructions of any type. The ability to execute more integer instructions is where the additional integer performance will come from. This is one of the big changes that made Intel's Core 2 Duo so much faster than its predecessors even on integer based workloads.

There also changes in Bulldozer to the cache hierarchy and memory controllers that will affect performance on all workloads.