No? When the former marketing team touted Bulldozer as *THE* chip for the server market, and three years ago the new marketing team comes in front of an investor audience and tells world + dog that ARM is better than x86 for servers, how do you interpret that? That Bulldozer was a huge success?
Nobody says Bulldozer was a "huge success".
8150 vs 1100T BE - http://www.anandtech.com/bench/Product/434?vs=203
When talking about advancements, Bulldozer was an advance on AMD's flagship at the time not a step backwards.
In the interests of fairness I'll be the first to admit that it wasn't much of an advance! It still was not a step backwards.
Nobody says Bulldozer was a "huge success".
8150 vs 1100T BE - http://www.anandtech.com/bench/Product/434?vs=203
When talking about advancements, Bulldozer was an advance on AMD's flagship at the time not a step backwards.
In the interests of fairness I'll be the first to admit that it wasn't much of an advance! It still was not a step backwards.
I am disappointed in the lack of performance of the last 2 intel generations, but at least they are not slower in close to half the benchmarks (no, I did not count them up, just a rough estimate) like in the comparison you showed.
That's all I meant - both Haswell and Piledriver are what Ivy and Bulldozer should have been a year ago.
Ivy wasnt a mega flop like Bulldozer. It was progress on all fronts.
And Haswell isn't near the increase Piledriver was. It's all shaken out to be quite similar in the end.
And Haswell isn't near the increase Piledriver was. It's all shaken out to be quite similar in the end.
Intel probably needs a completely new architecture to keep its CPUs IPC improving.
When talking about advancements, Bulldozer was an advance on AMD's flagship at the time not a step backwards.
This! 8 K10 cores on 32nm silicon destroys Bulldozer in every metric; in single thread, multi-thread, and power consumption; while incorporating the additional extensions introduced on Bulldozer.Step backwards in terms of architecture performance. Slightly faster than the phenom 2 x6 but uses more power despite the phenom being 45 nm vs 32 nm. If bulldozer was on 45 nm how would it have performed?
Die shrinked phenom x6 would probably have given better perf/watt (which really is the most important characteristic for servers).
Nobody says Bulldozer was a "huge success".
8150 vs 1100T BE - http://www.anandtech.com/bench/Product/434?vs=203
When talking about advancements, Bulldozer was an advance on AMD's flagship at the time not a step backwards.
In the interests of fairness I'll be the first to admit that it wasn't much of an advance! It still was not a step backwards.
This! 8 K10 cores on 32nm silicon destroys Bulldozer in every metric; in single thread, multi-thread, and power consumption; while incorporating the additional extensions introduced on Bulldozer.
It's easy to say that Haswell is theoretically derived from the basic architecture of the P6 and Banias, but there's been so much iteration on that stuff that they are barely recognizable as the same thing.
Just going to point out the obvious that you aren't comparing bulldozer to thuban there.
Rather, you are comparing an 8-core 16MB cache processor running at 3.6GHz to a 6-core 8MB cache processor running at 3.3GHz...hardly apples-to-apples.
Here is a comparison that is a little more even-keeled, both are running 8MB cache and are running at nearly the same clockspeed (albeit the clockspeed is still tilted in favor of the bulldozer chip which has also turbo-core).
Piledriver is over 50% faster than K10 in certain MT loads. The die is smaller than Thuban and power consumption remained quite similar
disappointing but not really unexpected. I mean as leader your usually better of to not led your competitor get a significant advantage in a certain area, like iGPU. if that takes of your busted. So as long as AMD CPUs suck intel can invest in a better iGPU.
Personally I would prefer a mainstream 6-core without igpu...with this I still have no incentive to upgrade my CPU at all. Display and dGPU will go first.
You've seen 8 K10 cores on 32nm have you? If it was that simple then AMD would have shrunk K10 by now.
Piledriver is over 50% faster than K10 in certain MT loads. The die is smaller than Thuban and power consumption remained quite similar.
When Anand reviews the 4770K we can compare the 2700K to the 4770K and see which company has made the most performance gains.
I feel like i've stated this a million times but AMD is not a motivator to intel whatsoever at this point, because desktop is becoming less relevant by the day. The real threat to intel is ultra portables and the mobile market, and that is the entire reason that intel has spent years trying to make their processors super efficient - intel will _not_ be relevant 10 years from now if they can't make waves in the ultra portable market (tablets, etc). That isn't to say that Haswell won't be used on the desktop / enterprise, but the uarch is designed from the ground up to be efficient. IPC matters much less these days because average consumers don't care, and enterprise cares more about efficiency than they ever have in the past. It is absolutely critical that intel improves efficiency as much as possible.
AMD has little bearing on intel's direction. Actually , scratch that. AMD has no bearing whatsoever on intel's current direction - the threat right now is ARM SOCs. On that note, Haswell is a very positive step in the right direction, as rumors are indicating 12 hours of battery life for their ULV parts. If that is true, intel will definitely create small waves this summer/fall in the mobile market, while next year should be a game changer. ULV Broadwell parts will be that much better than Haswell, and will have even better efficiency than ARM SOCs with significantly better performance.
Even with a conservative process scaling of 1,5 a Thuban shrink to 32nm would scale down to 230 mm². With a 1,75 scaling down to 200 mm².
If you want to see what K10 on 32nm looks like, take a look at Llano- and compare it to Trinity. Trinity only slightly increased the die size, but kept the same 100W TDP and significantly improved performance.
EDIT: For example:
If you want to see what K10 on 32nm looks like, take a look at Llano- and compare it to Trinity. Trinity only slightly increased the die size, but kept the same 100W TDP and significantly improved performance.