Wow, this could be big for AMD, or a big let down for us speculators. I'll just forget I heard this and wait for the real deal to be tested here at XS.
Printable View
Wow, this could be big for AMD, or a big let down for us speculators. I'll just forget I heard this and wait for the real deal to be tested here at XS.
Very nicely said.We don't know what freq. Penryn will hit and we don't know how "K8L" performs(we have a vague idea after AMD's press conference).Quote:
Originally Posted by alucasa
I think AMD will be more than competitive this year.Next year will be tough for both Intel and AMD.
gOJDO, you should change the thread title.
AMD doesn't make that claim to attribute to them.
Nowhere do they say they'll be 40% > Clovertown.
Quote:
Originally Posted by uOpt
I think AMD's Randy Allen said it(and I don't think he would be lying about it :) )
Quote:
"We expect across a wide variety of workloads for Barcelona to outperform Clovertown by 40 percent," Allen said
Quote:
Allen said there's an 80 percent scale over dual core processors; AMD's own benchmarks put it well ahead of the Xeon 5300 in terms of performance.
"We took their best punch with Clovertown and this is our response," he said.
Need to type quicker .....
deleted.
40% more/GHz...
sounds great to me.
that puts a 3.0GHz Barcelona against a 4.2GHz clovertown.
if it scales any better, it will full out dominate
Quote:
Originally Posted by gOJDO
It's true, but only for multi-socket systems. K8L should perform a bit better than Clovertown by itself, but 40% seems far fetched. It's definitely more than possible in your 4P or even 2P systems.
proof please!
These comments and the "wide variety of workloads" use highly multithreaded workloads where the Hypertransport, the integrated memory controller and NUMA come into play.
On a per-core performance they don't say how much.
But they say that going dual-core to quad-core gives them 80% speedup. They don't say how many CPUs that test system has but you can assume 2 since they mention Clovertown which is 2-only. So going 4 cores to 8 cores raises performance on their benchmark by 80%, which implies a highly scalable multithreaded benchmark.
At the same time they say that there is a 60% per-watt gain AND that there is the same wattage for dual and quad core. Those numbers don't mix with the 80% gain.
We can speculate that per-core per-MHz performance must raise somewhat, otherwise you cannot beat Clovertown no matter how well your cherry-picked benchmark scales.
Overall AMD is NOT saying that it is 40% faster than Clovertown and the thread title should be changed.
@uOpt
I agree with you. If K8L is a 3 issue core then it has no theoretical chances to perform 40% faster than Clovertown. Not even close to that number.
IMO clock for clock, K8L will perform similar like Core2. One will be marginaly faster than the other. According to the data and the logical speculations, I think that Penryn will be faster clock for clock than K8L.
Penryn will be clocked much higher than K8L and it will clearly outperform K8L CPUs on platforms with 1 or 2 CPUs.
K8L will dominate on MP platforms with 4 or more CPUs.
I think that Server is the market where AMD will place thier bets with K8L.
The desktop/workstation market will be owned by Intel.
Anyway, Allen claimed that K8L will be 40% faster than Clovertown. So, I think that I should'nt change the topic of this thread.
Agreed.Quote:
Originally Posted by uOpt
Actually, they say that for floating-point calculations there will be an 80% speedup per core at a given frequency:Quote:
Originally Posted by uOpt
"A faster floating-point engine performs mathematical calculations--long an Opteron strong suit, though not as important a part of the chip as that for integer operations. At a given clock frequency, a Barcelona core outperforms a current Opteron core by a factor of 1.8. By going quad-core, a Barcelona chip overall will provide a boost factor of 3.6, Allen said."
OK, let's pick some more: [ETA: didn't see oldblue's reply before I wrote this]
So what he seems to say is that the new chip at the same clock rate is 80% faster (3.6 / dual = 1.8 = +80%) for floating point.Quote:
"We expect across a wide variety of workloads for Barcelona to outperform Clovertown by 40 percent," Allen said. The quad-core chip also will outperform AMD's current dual-core Opterons on "floating point" mathematical calculations by a factor of 3.6 at the same clock rate, he said.
There's no question that some architecture changes are coming, but nothing too exiting has been announced, ever, except the widening of the SSE instructions.
Then, if you cherry-pick a benchmark that uses these SSE instructions (most of them are floating-point) and that scales well in the presence of AMD features (hypertransport, NUMA, integrated memory controller), then the math looks as follows:
Per core:
- for general applications K8 is about 20% slower than Core2 for most workloads, but much more some some
- for highly optimized SSE workload it's worse, probably 50-60%
- if you gain 80% in these SSE workloads, you end up 20-30%% faster
- then you add 10-20% on top from better scaling
That doesn't change the fact that the 40% claim is only for highly scalable multiprocessor benchmarks making heavy use of SSE.
The per-core performance will probably only go from a 20% disadvantage to a 15 or 10% disadvantage.
Not that any of the AMD numbers given are even remotely compatible with any of the other AMD numbers given.
Now,how on earth would you know that???Quote:
Originally Posted by gOJDO
How can die shrink of MCM known as Clowerton ,with larger L2(which in allendale vs conroe comparo gives an~5% advantage to conroe) be faster than yet unknown CPU which haven't been available on the market nor tested by anyone outside the AMD?
And wouldn't you think that AMD knew about Merom some years ago when they started working on post K8 designs??One would think AMD was aware of MCW assault by intel and was preparing its own "weapons" to respond.
So AMD will be late 1 year and will bring out the underperforming CPU at the most crucial moment for the company's future?Umm,i don't think so.
without getting into a debate, i hope this performs at least a wee bit better then C2D systems so intel's inflated C2D pricing will come down.
it would be sweet to be blown out of the water again like on A64, but intel already blew us out with C2D, so the odds of that happening again so soon i feel like are a bit lower.
It depends on the application. There are a lot of factors that affect performance other than how wide a core is, and the actual number of instructions C2D can decode is limited by its 16 byte instruction fetch.Quote:
Originally Posted by gOJDO
Here's a good X-bit Labs article on the subject.
I think that Hans De Vreis once(at Aces' Hardware board) discussed the "issue" of the 3-issue design of "K8L" and said that it is not the problem at all.I can't find the thread,it seems it's well digg up in Ace's archive(i will try some more digging).Quote:
Originally Posted by oldblue
All in all,like oldblue said,the problem is not in theoretical numbers but in the capabilities of the instruction fetching and related units.
I hope AMD does one-up intel to keep intel from becoming stagnant again like they did with the P4. Each one on-upping each other makes for faster stuff at a lower price :D
I just don't think AMD would be able to keep up in a game of one-upping... I think they just need to press on with whatever plan they have
well, what i said was far from that as wellQuote:
Originally Posted by LOE
I don't think Intel besting AMD for a year is going to put AMD in the ground, far from it
all I'm saying is I don't believe AMD has the fabs to keep up with Intel in a game of constant 1-upping..... yet...
Am I the only one who thinks comparing a next gen AMD cpu with a Intel CPU that is out for months by now doesnt say that much?
From what I recall K8L is gonna hit the market in late Q3, maybe Q4, so it wouldnt have to compete vs Clovertown. And are they talking about one K8L vs one Clovertown. Or one K8L vs 2 Clovertowns or 2 vs 2?
I hope AMD will have something really fast with K8L. Thats only good for the competition.
But I still dont trust anything some employees say. Its a whole different story if the CPUs can be orderd and unbiased people post numbers and articles.
K8L is still very far away to hold the money back.
If you always wait for something faster that has been announced, you will never get something. As soon as that promised ultra fast stuff is out, there are already upcoming faster things on the horizon.
Again, I hope K8L will be fast, but its way too early to compare it with CPUs that are on sale righ now. There is too much stuff and new inventions in the IT market for that kind of comparissons.
No, ur not the onlyone. But I think K8L will be launched at 22. April and thats early Q2.Quote:
Originally Posted by Fr3ak
Read the thread from the beginning,AMD's New Core will come very soon,in May/June for the s1207(QuadFX probably in the same timeframe) and also in May/June for uniprocessor systems in the form of AM2 version.This was said many times in AMD presentation and now we have confirmations that AMD is ramping up Barcelona http://www.theinquirer.net/default.aspx?article=37154Quote:
Originally Posted by Fr3ak
...32byte instruction prefetch...
:P~~
->
Watch out! A nice 64bit performance bumb heading your direction!
OK, so I take everything back of what I said.
I had Q3 07 in mind and Q1 08 for AM3. Guess those dates are wrong then.