Slashdot Mirror
First Benchmarks of AMD Hammer Prototype
porciletto writes "As seen on Ace's Hardware, this article features Quake 3 benchmarks comparing an 800 MHz ClawHammer sample to Athlon MPs at 800 MHz and 1667 MHz, as well as a Willamette Pentium 4 (256 KB L2, 400 MHz FSB) at 800 MHz and 1600 MHz. The benchmark results indicate a 40% performance increase over an Athlon MP for the ClawHammer. Additionally, the 800 MHz ClawHammer manages to tie (actually outperform by 1 FPS) the 1667 MHz Willamette Pentium 4."
They tested some software which had been compiled for 64 bit mode. With the large number of 64 bit registers the hammer has there should be some significant speed improvement.
ps -- where is the obligatory Beowulf cluster commentary on this??? I am shocked and appalled at this apparent oversight by my fellow /.'ers...
...we are from the government - we are here to help...
If you manage to get through the slashdotting, the story in the tecchannel web pages is amazing. The prototype Clawhammer, while limited to 800 MHz, performed shockingly well on the few, but varied, benchmarks they subjected it to. It's interesting that both Intel and AMD teach the same lesson, that MHz doesn't determine performance. Unfortunately for Intel, they demonstrate it by the P4 not running as fast as the MHz would imply, where the AMD chips run far faster than MHz would imply.
I can't wait for these chips to get out there.
thad
I love Mondays. On a Monday, anything is possible.
No, it outperforms the 1600 by 1FPS. Still quite the feat. If this thing releases at even 1200mhz, you're looking at something comparable to a P4 2.4ghz. The site does say they stated they were aiming for 1.6ghz (nice!), but we'll see if that actually happens.
It's nice to see that the industry isn't playing too much of the "more is faster" game, at least as much as they used to. When an 800mhz part is comparable to a 1600mhz, you've got to wonder what intel isn't doing to optimize.
It's obvious computing needs and trends are changing rapidly. PC are no longer luxury items as they once were. They are now taken for granted and no one cares about the technology behind it anymore. With all the cheap PC's out there, the PC is becoming less high tech and more every day. Back when PC first started, it was new and different, but now it's just another appliance. With that kind of perception taking over the PC industry, no one gives a damn how many ghz a pc has.
Sure there's always gamers and programmers who want the newest/best system. But that will change in another year or two. Perhaps that's why Intel is starting to focus more on telecommunications and less on consumer CPU.
What I'd like to see is the performance of the gcc x86-64 backend. Especially against ICC, since it's been beating gcc in benchmarks.
If you had read the article (and many others) the industry expects that if they can work out some silicone problems the opteron will debute at 1.6Ghz, or twice what the demos are running at. Since the current 800MHz parts are matching the 1.6GHz p4, Intel would need to be at 3.2Ghz to match the Opteron at release, since quatity shipments of the Opteron won't happen till Q2 03 and Intel's roadmap says that's when the 3.2GHz p4 will begin production I would say it is likely we will have the same situation we have today where about once a quarter based on production schedules one of the manufacturers will take the speed crown from the other just to have it retaken a few weeks later. It looks like unless the marketing muscle of Intel can misinform people into believing that just because they come with a bigger number attached that the p4's are SO much better that the Opteron should do well. Add in the 64bit icing on AMD's cake and it things start to look good for AMD in the low to mid range x86 server portion of the market.
There are 4 boxes to use in the defense of liberty: soap, ballot, jury, ammo. Use in that order. Starting now.
AMD will be doubling the speed and stepping, with a targe tof 1600 to 2000MHZ. They are also doubling the onboard cache of ClawHammer.
Intel will have a pure 64bit chip out with 1/2 the 32bit performance of todays PIV (their words mind you). Since it will take years for 32bit apps to disappear, this is a big deal. Their first 64bit chips will come out at 2.4 - 3GHZ, but with half the 32bit performance and only 1/3 the FPU performance since intel worries more about MHZ than individual operation efficiency.
The ClawHammer should ber 20 - 40% faster out of the gate initially, and AMD has plans to ramp up the speed to 3GHZ by mid summer. Intel has no plans to ramp up speed on their 64bit chip beyond 3.5GHZ in 2003.
It's all speculation, but get your facts right at least.
Try looking at it one of these ways: (A) This isn't news, it's entertainment. Everyone's cheering for the underdog AMD, partly 'cause it's easier for little ol' us to identity with than giant Intel and partly because the fun will end if Intel kills AMD. (B) We don't want to pay either of 'em $2000, so we shift our support to whoever's the underdog in order to prolong the price/performance war.
It's easy to make up & spread cool- and credible-sounding stuff. Finding & checking hard facts is hard work.
In the real world, where people dont usually just run the few micro-benchmarks Motorola's CPUs excel in, you need high speed execution of long strings of dependent serial instructions ... trying to improve performance by architecture alone runs into diminishing returns awfully fast, only MHz helps there in the end. Thats AMD's bread and butter, and Im pretty sure they realise that.
...
As everything in life you can go overboard of course, Intel might have gone a little overboard to one side (too many pipeline stages) but Motorola is too far overboard on the other side to play a part in the high performance processor buessinuess (which is why Apple will probably have to switch to IBM wholesale in the near future, especially with Motorola giving up trying to keep their semiconductor processes competetive for high speed logic).
AMD probably wont close the MHz gap, but they do not intend to lag it to the same extent as before
Still, I'm eagerly awaiting the ClawHammer release. Every x86 box I've built for the last 5 years has been pure AMD, and I've been quite happy with them.
i just find it weird for the community to really compare the new hammer with the p4 product line of intel. if the main reason behind the hammer is to directly compete in the server line, then it should be the hammer vs itanium2 vs sparc vs pa-risc vs alpha vs powerpc. if you are going to compare it with p4, a professional will not even take you seriously.
why use a some low form benchmark. although i understand that the current systems are in prototype, the benchmark should reflect something of the server world including but not limited to tpc, spec, etc. i would really love seeing the performance of hammer in a oracle/sql/db2 or other database benchmark. i would love seeing the hammer handling ssl transactions and others.
with regards to amd using x86 with compatibility to 32bit, would it be dumb if you would run some non native applications? this means that amd anticipates that companies will not optimize their software to run on pure 64bit platform. this may be an indication that the initial design is not intended for the server product line. running 64bit does not make you compete in the server arena!!!!! the server market is a very different ball game compared to the consumer - cpu is not the prime reason.
and x86 is obsolete. it is not the efficient out there so it is time for a major change in the hardware world.
Live your life each day as if it was your last.
Clawhammer and Willamette (using PC800) have similar amounts of memory bandwidth 2.7GB/s vs 3.2GB/s. The Clawhammer, with its ondie memory controller, will probably have half the latency of the P4+Rambus. Hammer is also supposed to have higher IPC, putting it way over the top.
Because of the number of games using the Q3A engine, I don't think it is a bad benchmark of gaming performance.
The rest of the year should be interesting: If AMD hits 1.6 and that scales to the speed of a 3.2GHz Willamette, then it should be very close performancewise with Intel claiming to be ramping Northwood up to 3Ghz for year end.
I think it is very honorable (can't find a better word) that you are supporting your company, and I am fully aware that yes, given the choice, AMD will over-charge for their processors too -- BUT:
1) Intel's tactics, previously *and* recently, has not been too wise. one of the most important of this in RAMBUS: Intel basically said that (back in the first days when RAMBUS came out) we are going forward w/ RAMBUS. everyone knows that it's a M$-like move, everyone knows that it's b/c intel is getting a huge chunck of $$ from RAMBUS, and nobody liked it. Did Intel care? no; in fact, Intel stopped licensing VIA for making SDRAM / DDR chipsets; People like me looked around -- who is supporting DDR? well, AMD is...
2) Technically, Hammer makes more sense: sorry but from a user point of view (user as in programmer, IT admin, etc) -- Hammer makes like easier than IA-64; i mean, if you are gonna make such a drastic move away from IA32 (which, i admit, sucks bad), then do it slowly! What is the difference between migrating from IA-32 to IA-64 vs. say, alpha / solaris / whatever? not a whole lot -- pretty much everything needs to be redone. i mean, i still get the Intel sticker, but what's the difference there?
3) sorry but people has an tendency to cheer the under-dog; fact of life... i remember back then everyone cheered for NVIDIA when 3dfx was "king", now-a-days people still respects NVDA, but does not cheer for them anymore because they are now a status-quo -- they are expected to be one top -- but any move of the underdogs: ATI radeon / matrox paphelia (sp? heck, where they come up w/ this anyway) / Kryo(?) gets recognition. it's like a fight - will the new comer defeat the champion? we are geeks -- not much soap opera in our life, so this is the subsitute.
So... yes given the opportunity, would AMD do the same as Intel (rambus, etc)? maybe -- but i actually think AMD CEO is wiser than that -- he is a cool guy, btw -- visit any AMD facility and you can see his indiana-jones parody poster -- it's pretty cool. but we never know; but one thing is for sure -- if AMD ever gets on top, there will be probabbly a lot of cheer for intel for making attempts at the "de-throning"
last point, AMD is not TEXAS based. they are in SAN JOSE, CA. addr is something like 1 AMD lane or something.... get it straight people.
again, I would like to see your counter-arguments, if there are any. it's not fun writing without knowledge that there would be some reply.
-LQ
My life in the land of the rising sun.
But while temping at Intel I got the impression that AMD was no good, not even in giving Intel employees huge discounts on P3's (which at the time was their flagship product) - I even got into an argument with a technician who couldn't understand how AMD chips reduced the prices on Intel chips.
Not to mention such discussions were met with violent disagreement. If you said you had an AMD system in an interview you could kiss that job good bye.
I also got the impression talking to people that if Intel executives had this master button that said "get rid of amd" that no-one would even hesitate to push it. Despite the fact that AMD pushes Intel to move faster (and visa versa). Just compare all the time between the P90 (which I recall coming out in 92 and costing over 900$), PPro, and P2.
Why the worry?
There are geeks like me who have been screwed by intel's buggy processors (somewhere around here I have a 500$ P90 that has an fdiv bug in it - after much haggeling on the phone I never did get it replaced), and their high prices. But I still have Intel systems floating around here, as well as a few sparc, and a few amd machines.
AMD is the underdog - they made X86 compatible CPU's without reverse-engineering (yes its true! do the research), they made intel's cpu's for a very long time and I for one think its kinda neet that a company like AMD despite all the flack Intel has given them has made chips that perform comparibly with Intel's chips - which is probably why slashdotians favor them in discussion boards.
I am a long time system designer
Another reason why i tend to prefer Amd is the cynical marketing processor known as the P-4. The vast majority of benchmarks show that unless your running software thats heavily SSE-2 optimized, the Athlon's spank the P-4. Yet the P-4's are much more $$$$ due to all those wonderful Intel commercials with dancing morons in bunny suits, or some smucks painted up like a martian with a bad head cold. Instend of wasting all that money marketing, use it to improve your designs! Amd spends virtually nothing on marketing, and yet whenever they have a good design, their products sell extremely well. And dont get me started on intel's late ddr support, or the earily 845 chipsets that were sdram only, which had PATHETIC performace.
I guess the point of my whole rant is......I use Intel or Amd, or whoever, as long as they give me a good value for my (or my customer's) dollar. Give me a nice industry standard design. Dont foist some new marketing propierty design on me. If its gotta be propierty, it better be for one of two reasons: Considerably cheaper, or considerably faster. Intel in the past few years has NOT focused on giving the customer value. Amd has. Give me a 1000 dollars, and I can build either an Intel box, or an Amd box thats 20% faster then the Intel box, and just as stable. (I dont buy the Amd isnt stable arguement, it all comes down to knowing your hardware and how to configure it properly for stable operation.)
When Intel returns to delivering a product that is worth the price Intel charges for it, I'll use Intel again. Until then, I'll continue to laugh at ridiculous marketing schemes and do my research on which product is the fastest for the least money.
Lawyers, MBA's, RIAA? A jedi fears not these things!
I think people sense that Intel went the wrong way. Even Intel can't fight physics.
Energy dissipation of processor is in vast majority due to tranistor switching. So if you switch tranistor N times more frequently you generate N times more heat. In case at hand P4 has to switch it's transistors roughly 2 times more often than the Hammer for the same amount of work to be done.
Why is this important, who cares about power usage ?
The problem is that too much heat on a very small spot can rise temperature (localy) to the point it damages that part of processor. This is why both athlon and P4 are in the end thermicly bound.
Thought it might seam that going to smaller technology helps it really makes matters worse. Just think of it, now you have transistors that generate heat closer together, so cooling becomes even bigger problem. Yes you can lower the voltage (till the noise gets too big), but it turns out that this effect is not enough. (Meaning is you have a CPU core thermicaly bounded at 180 nm it will be stay thermicaly bounded at smaller technology)
So I think P4 extreme frequencies will prove quite a problem for Intel in the long run... Asuming that AMD and Intel will have comparable technology P4 woun't be able to keep up with Hammer.
First, Intel literally SCREWS users over. AMD hasnt went that far. Making what they or Intel can off a product is one thing. Changing the PII to Slot I because it is sooo superior to Socket 7 and similar socketed designs (all while hiding the standard type chip in a neat case) and *advertising this* to help cripple AMD's slow creep into your market was absurd. It hurt the market, and it was a lie - as we all (who werent the techie type and already knew) shown when Intel changed back to the socket design.
Intel released buggy CPU after CPU just to make deadlines and be able to yell "me first!" even when the defects were known (online articles, lawsuits and statements concur).
As mentioned, "flavor of the month" seems to also be Intel's biggest favorite thing to do to make some extra money off everyone involved... new chipsets all the time, new socket/slot/socket/slot designs (often for the same damn chips). Exhorbitant pricing (while you are claiming AMD would/will do the same but is consistently cheaper on an equal performance level).
Then there's Eternium... do I even need to go any farther with that buggy, half decade delayed piece of still in the works crap?
So, nothing personal, but why do you bother? You apparently dont know much about the company you work for. I've worked for some big IBM VARs and seen too much of what Intel really does that I am not permitted to mention here.
WebMaster:
BinFeeds
XXX Thumbnailed Image Newsgroups but