Here's something I've been wondering: what will it take for AMD to win this year with Phenom?
If my Phenom Preview is to be believed, AMD will improve performance around 15% clock for clock over the K8 with Phenom. At lower clock speeds/cache sizes, AMD and Intel end up offering similar performance as I showed in the Mainstream CPU Roundup. There is definitely the possibility that AMD could offer higher performance vs. Conroe/Kentsfield with Phenom, but the question is will it be enough?
For Socket-AM2 owners, the upgrade path will make a lot of sense, just swap in a Phenom and get Core 2-like performance. But for new system buyers or LGA-775 owners, the move is a little less clear. If AMD can offer better performance, but at a similar price, will that be enough for users to choose Phenom over Core 2?
Then there's the issue of overclocking. AMD hasn't had much success at 65nm, all of its highest clocked Athlon 64 X2s are still based on the older 90nm process. While I've been told that things will change with Phenom, I'm not very confident that Phenom will be able to out-overclock Core 2. So if AMD can provide Phenom that's clock-for-clock faster than Intel, at a similar price, but can't overclock as high will the new core change anything?
Intel also has an interesting upgrade path, as Penryn will eventually make its way down to even the cheapest Core 2 products, so as long as you have a Penryn-compatible board today you'll be sitting pretty come next year.
Then again, maybe I'm looking at things too cynically, maybe all AMD needs is to have more competitive performance at stock speeds and at competitive prices; let's leave it up to the sales 'n marketing folks to duke it out with OEMs and in the channel.
If my Phenom Preview is to be believed, AMD will improve performance around 15% clock for clock over the K8 with Phenom. At lower clock speeds/cache sizes, AMD and Intel end up offering similar performance as I showed in the Mainstream CPU Roundup. There is definitely the possibility that AMD could offer higher performance vs. Conroe/Kentsfield with Phenom, but the question is will it be enough?
For Socket-AM2 owners, the upgrade path will make a lot of sense, just swap in a Phenom and get Core 2-like performance. But for new system buyers or LGA-775 owners, the move is a little less clear. If AMD can offer better performance, but at a similar price, will that be enough for users to choose Phenom over Core 2?
Then there's the issue of overclocking. AMD hasn't had much success at 65nm, all of its highest clocked Athlon 64 X2s are still based on the older 90nm process. While I've been told that things will change with Phenom, I'm not very confident that Phenom will be able to out-overclock Core 2. So if AMD can provide Phenom that's clock-for-clock faster than Intel, at a similar price, but can't overclock as high will the new core change anything?
Intel also has an interesting upgrade path, as Penryn will eventually make its way down to even the cheapest Core 2 products, so as long as you have a Penryn-compatible board today you'll be sitting pretty come next year.
Then again, maybe I'm looking at things too cynically, maybe all AMD needs is to have more competitive performance at stock speeds and at competitive prices; let's leave it up to the sales 'n marketing folks to duke it out with OEMs and in the channel.
Facebook
Twitter
RSS
7 Comments
View All Comments
supremelaw - Wednesday, November 21, 2007 - link
When I did systems analysis many years agofor a minicomputer company, one of the most
valuable lessons we learned was to put our
wood behind the right arrowhead.
Here's a very simple example:
if a subroutine spends 90% of its time
in one set of code, and the remaining
10% of its time in another set of code,
where does a 2x speed improvement
make the most sense?
If we optimize the 10% code and
it goes to 5%, we've reduced run-time
from 100 units of time to 95.
No big deal.
If, on the other hand, we optimize
the 90% code and it goes to 45%,
then we've reduced run-time from
100 units to 55 units.
This IS a BIG DEAL!
The widespread industry focus on
CPU performance begs the following
question: Do you want a CPU that
idles 95% of the time, or 98%
of the time, especially when the
latter 3% margin costs $200-400 MORE?
That appears to summarize the choices
that Intel and AMD have routinely
presented to the marketplace, over
many years of recent history.
The above observations have now convinced
me that our industry as a whole should shift
focus to accelerating Input/Output --
ACROSS THE BOARD.
Here's another simple example:
by upgrading from 1GB to 2GB of RAM,
and by configuring a RamDisk Plus
at 512MB, our IE7 browser cache now screams,
and that change also off-loaded
the hard drive where IE7's browser cache
was previously hosted.
Another, more complex example is
a solid-state solution that we are
now designing here: Gigabyte's
newly released GO-RAMDISK-BOX-RH
aka "RAMBOX" is a very promising
device, especially if Gigabyte heeds
marketplace requests (mine!) and upgrades its
interface to 300MB/second using DDR2 DIMMs.
Lastly, and probably of greatest significance,
insofar as typical users are coming to rely
more and more upon the Internet, America as
a nation is now suffering from horribly slow
broadband speeds, by comparison with other
high-tech regions of this planet. Once again,
this is an I/O problem, NOT a shortage of CPU power.
Let's start shifting our collective focus
to achieving bright and effective improvements
in storage Input/Output speeds.
The Internet will definitely benefit, and
so will we who depend on it more and more
in our daily lives.
Sincerely yours,
/s/ Paul Andrew Mitchell
Webmaster, Supreme Law Library
http://www.supremelaw.org/">http://www.supremelaw.org/
NullSubroutine - Thursday, November 8, 2007 - link
If I remember correctly you did your benchmarks with B0 engineering samples, while what was produced and bought was BA samples which are reported to have "conservative performance increase estimate of 5%". Given that there is reportably another 2-4 revisions since B0 it could be 25% performance increase clock for clock over K8. While this might not win awards, AMD does not care about awards. It wants to sell volumes of processors, if it is a great deal price/performance for the entire system, what does it matter if they arent speed demons?While I would like to see AMD produce damn fast processors, the price war means us comsumers win, but it doesnt mean AMD sucks because they dont own the $1000 peformance crown.
razor2025 - Wednesday, October 31, 2007 - link
I'm not sure if Anand's comment on AMD's 65nm process "not good enough" is correct. From what I've read online, the Black Edition X2 5000+ is a 65nm product and has no trouble ramping up the clock to 3.2-3.3ghz. Albeit the architecture of X2 is quite mature, and thus easy ramp-up is expected, I hopeful that AMD's Phenom will prove to be competetive with Core 2.As far as motherboard upgrade-paths, either platform has pros and cons. As long as Intel keeps the CPUs pin compatible with socket 775, I can foresee even the 945 chipset supporting Penryn (courtesy of ASRock). AMD still has the best IGP platform, and promise of AM2 backward compatibility may help to keep some of the loyalist from migrating.
I am very much hoping that AMD will be alive when they release their Fusion-based products. They've hedged their company life in acquiring ATI and entering the GPU market, so I'm interested in seeing exactly what they want to do with Fusion.
FelixDeKat - Sunday, October 28, 2007 - link
AMD dropped the ball again. I hope Intel doesnt drive them out of business this time. Competition is good. Celerons for everyone!Tanclearas - Thursday, October 18, 2007 - link
I honestly believe that in the enthusiast community, AMD's biggest "product" slip was AM2+, and not so much Phenom. I had used an Opteron 165 (OC'd to 2.4) for over a year and a half. For me, that is nearly unheard of. Pretty much the only component that will stick around that long for me is my monitor.I knew that I was getting close to upgrading, because that's what I do. I didn't necessarily need more than the 165 was giving me. As far as gaming goes, I push the limits of the card (currently 8800GTS), and the CPU is rarely (if ever) the bottleneck. Even in those odd situations where the CPU is the bottleneck, it's in a game where I'm already getting insanely high framerates. I don't do much in the way of CPU-intensive computing, and where I do, I don't care if it can finish in 2 minutes vs 3 minutes.
Anyway, I had an opportunity to sell my complete system, and I took it. When I looked at what to build myself, Intel seemed like the only choice. Phenom was still at least a few months away, and though I didn't have any specific expectations for it, I was making (I think rather safe) assumptions that it would be faster in most applications than my 165. Beyond that, there were a few reasons I thought AM2+ would be a good choice.
1) I could buy a cheap X2 5600 (2.6GHz + 1MB cache per core) that would be slightly faster than my overclocked 165, and with a moderate overclock could give me as much as an extra 400MHz. Perhaps even as much as 600MHz more than I had my 165 running at.
2) It would be ready to take Phenom, and support dual power plane and higher HT.
3) Long legs that extend to AM3 processors. AMD is saying that AM3 processors will work in AM2+ boards.
That meant that a decent investment in a new motherboard and good DDR2 memory could potentially carry me for the next 2 to 3 years. If that had have happened, AMD would have also ended up selling not just 1 CPU, but 3. Without any AM2+ boards available, I decided to go with Intel, and I can pretty much guarantee that my next CPU will be Penryn-based in another year or so.
crimson117 - Wednesday, October 17, 2007 - link
The upgrade path is key.How long is AM2 going to be around? Manufacturers know how to make a peryn compatible motherboard, so they can design a PC around it and just drop in new processors down the line. How can investing in AM2 make sense, if phenom is only just catching up to Intel's Core 2?
kleinwl - Wednesday, October 17, 2007 - link
I think you are being to cynical Anand. AMD does have to be an enthusast company anymore... they have a direct line to the mainstream thru dell. All AMD has to do with Phenom is be competitive with Core 2 and have slightly lower prices and they should do well in the mainstream. Even if AMD chips don't overclock at all, AMD should ship enough chips to survive.It's just the crazy overclocking community that will be disappointed... and how big is that? maybe 0.01% of the total market? Yes we are opinion leaders... but still we maybe influence 1-4% of the market? Not enough to make or break a company.