2GHz Hammer at MPF

Kelledin · Oct 16, 2002

Nice thread on the Opteron demo:

<A HREF="http://www.hardforums.com/showthread.php?s=&threadid=512418" target="_new">http://www.hardforums.com/showthread.php?s=&threadid=512418</A>

Especially nice, as the Hammer's supposed to be released about 400-600MHz higher than that...

I can love my fellow man...but I'm damned if I'll love yours.

baldurga · Oct 16, 2002

Well, at least some info about Hammer. I was starting to think that it has died ;-)

Seriously, that sounds good. Just praying for no more delays, paper launches and bad news ...

DIY: read, buy, test, learn, reward yourself!

slvr_phoenix · Oct 16, 2002

Nice. So the Opteron (I'm assuming SledgeHammer core here, since last I read, most ClawHammers will be in single-CPU systems and thusly named Athlons instead of Opterons) doesn't even beat an Itanium in FP.

It had better come out 600MHz faster or have one heck of a low price tag, or else Opteron servers are going to be a joke. (Especially when Itanium servers are already considered to be a joke by most of the server market.)

And it also puts ClawHammer in a really bad light. I mean if Sledge can only slightly beat a 2.8GHz Northwood, imagine how Claw will compete against a Prescott!

It's sad, really. Had AMD released Claw earlier, it might have helped them regain the performance crown. It might have lived up to the hype. Now though, it looks like the Hammers will barely keep AMD in the game. That is, unless there is a significant improvement going into the final silicon that we just aren't seeing in these little demos.

And here I was hoping for a desktop chip that finally had great floating point power and SSE2.

<A HREF="http://forumz.tomshardware.com/community/modules.php?name=Forums&file=viewtopic&p=45775#45775" target="_new">Join the THGC LAN Party!</A>

imgod2u · Oct 16, 2002

I gotta say, those integer performance numbers are insanely impressive for an x86 chip. Especially considering it's running legacy code with none of the x86-64 enhancements enabled. For a webserver, it would be an incredible asset. Of course, the FP power leaves something to be desired since it is a high-end processor. So it may have a harder time getting into the CAD design market and rendering farms, etc. But it has huge potential for webservers.

"We are Microsoft, resistance is futile." - Bill Gates, 2015.

slvr_phoenix · Oct 16, 2002

I agree. The integer results are impressive and it would probably make a good web server.

That aside, such an integer score will do almost nothing for the SOHO market or for workstations of any kind. Considering that Claw and Sledge are nearly identical in ALU and FPU architecture and that the vast majority of sales will be for SOHO and workstation use, AMD would have been much better off putting more into the FPU than they did into the ALU.

It seems to me as though AMD designed the Hammers almost entirey around moving into the web server market. If you think about the x86-64 aspect as well, it seems even more likely. So one has to wonder just how useful these chips will actually end up being to us, the typical home and office users.

I severely hope that AMD isn't about to give up almost all effort at competition in the SOHO market just to break into the server market. That would be a pretty silly move. It's costing Intel a fortune to try and break into the server market.

<A HREF="http://forumz.tomshardware.com/community/modules.php?name=Forums&file=viewtopic&p=45775#45775" target="_new">Join the THGC LAN Party!</A>

spud · Oct 16, 2002

It had better come out 600MHz faster or have one heck of a low price tag, or else Opteron servers are going to be a joke. (Especially when Itanium servers are already considered to be a joke by most of the server market.)

Common silver be fair I haven’t heard any Itaimium owners complain. In fact Nexen Inc., Agricore United, and Husky Oilfield. All got Itaimium 2 servers, and a few SUN Blade workstations. Nexen needed them for a North Atlantic seismic analysis for oil reserves. Argicore needed them to compile oil flow and reasorce management. Husky Oil well that’s a not hard to understand since the expansion of the refinery has been approved here, they need the extra power to manage flow control, temps, burn-off and power consumption.

Now how do I know well I know the head tech at Nexen and Agricore and Husky was proud of their new farm so it was open to the public for a couple days. They have nothing but positive things to say about the Itaimiums, but like the SUN boxes better since their power consumption is fairly low. But number crunching seismic samples and radiation levels in fuels to gas density they never have had anything so powerful. Things that took at times days to compute take a matter of hours now. I’m sure other companies like Ford motor vehicles or Lockheed Martin have good things to say too. The Itaimiums a great cpu and does what it does well, saying the industry thinks its a joke well that’s a just about a lie since not everyone loves everything.

-Jeremy

Just some advice from your friendly neighborhood blue man

Edited by spud on 10/16/02 03:27 PM.

imgod2u · Oct 16, 2002

You have to realize that Hammer was never mainly targeted towards the end consumer, i.e. Joe Slack. It was designed as a "one-size-fits-all" solution because, frankly, AMD doesn't have the resources for 2 independent core designs. The majority of improvements over the Athlon is the improved integer performance due to the expansion of the GPR's, the 64-bit extensions and the packing stage (quite simply, your average code can maintain at best 1 FP op per clock, so packing instructions really doesn't help much with that). This was to help it compete in the webserver and 1U rack mounts that are still in high demand these days among webservers. This is a very profitable market and one that can be entered in a lot more easily than the high end server or CAD markets. x86 MPU's have far surpassed RISC-based, proprietary solution in price/performance for years and yet movie industries have only recently begun to switch to x86-based cluster farms. That just tells you how stubborn these guys are.
The main benefits of Hammer for the consumer is the slightly higher clockrate and of course, fine-tuned FP performance which in its own right is still impressive, just not earth-shattering.
On a side-note, remember when that guy "texas_techie" claimed that Opteron was scoring somewhat around 100%-300% better than the P4 in SpecInt? Whatever happened to that guy anyway? As it seems it isn't true, it's more like 20% and, let's just say if they used an optimized x86-64 compiler, 50% better than the latest P4 at 2.8 GHz (I think his number was in reference to a 3 GHz P4?).

"We are Microsoft, resistance is futile." - Bill Gates, 2015.

slvr_phoenix · Oct 16, 2002

Common silver be fair I haven’t heard any Itaimium owners complain.

That's probably because there aren't many Itanium owners.

Heh heh.

Okay, seriously though, Itanium2 is winning people over. The first Itaniums though ... well, Intel had a hard time selling people on them. Even Itanium2 is a hard sell. Most companies don't go out looking for Itanium. They usually 'settle' for Itanium if they use Itanium at all.

I'm not knocking Itanium though. I think it's a great concept that hopefully will catch on and get some cool improvements. It sounds like Intel is pretty devoted and it just may have a bright future. Itaniums are good chips.

That aside, most pros are hesitant to consider Itanium.

<A HREF="http://forumz.tomshardware.com/community/modules.php?name=Forums&file=viewtopic&p=45775#45775" target="_new">Join the THGC LAN Party!</A>

spud · Oct 16, 2002

Well I never said anything about the hot plate known as the Itaimium 1. Hehe I know I would loose any discusion on that bad boy. For first gen I guess it was alright though. Personally the Itaimium 3 would be a good starting point for the core IMO.

-Jeremy

Just some advice from your friendly neighborhood blue man

slvr_phoenix · Oct 16, 2002

You have to realize that Hammer was never mainly targeted towards the end consumer, i.e. Joe Slack. It was designed as a "one-size-fits-all" solution because, frankly, AMD doesn't have the resources for 2 independent core designs.

That's the problem though. I'd have thought that AMD would at least have made them more attractive to the majority of their customers...

This was to help it compete in the webserver and 1U rack mounts that are still in high demand these days among webservers. This is a very profitable market and one that can be entered in a lot more easily than the high end server or CAD markets.

My understanding though was that most people in this market are looking for low cost solutions. (Quick and easy pop-ins.) Opterons aren't likely to be all that low of a cost.

The main benefits of Hammer for the consumer is the slightly higher clockrate and of course, fine-tuned FP performance which in its own right is still impressive, just not earth-shattering.

It just concerns me that Intel could not just meet, but exceed AMD's efforts here. Then where would AMD be?

Not good. AMD is taking way too long to give us this and the impressiveness drops yet lower every time Intel releases a new chip.

On a side-note, remember when that guy "texas_techie" claimed that Opteron was scoring somewhat around 100%-300% better than the P4 in SpecInt? Whatever happened to that guy anyway? As it seems it isn't true, it's more like 20% and, let's just say if they used an optimized x86-64 compiler, 50% better than the latest P4 at 2.8 GHz (I think his number was in reference to a 3 GHz P4?).

Frankly, I've never really trusted any of his information. He doesn't give enough specifics and at times doesn't even sound like he knows what he is talking about. Not that he's a bad guy, just that I don't put any stock in his 'secret' information. Of course, I only believe what I have proof of.

Sounds like yet another perfect example to me.

<A HREF="http://forumz.tomshardware.com/community/modules.php?name=Forums&file=viewtopic&p=45775#45775" target="_new">Join the THGC LAN Party!</A>

imgod2u · Oct 16, 2002

That's the problem though. I'd have thought that AMD would at least have made them more attractive to the majority of their customers...

Designing an MPU improvement isn't something you can just do with a snap of a finger. There are limitations to how well you can do things. AMD is currently stuck with the original design they got from Next Gen. They don't have enough time or resources to build another core from the ground up like the P7 nor can they afford to just keep going with the Athlon. So they improve it a little and add x86-64 to try to get more profits from another market to perhaps fund the designing of a new MPU.

My understanding though was that most people in this market are looking for low cost solutions. (Quick and easy pop-ins.) Opterons aren't likely to be all that low of a cost.

Compared to other processors that are used in that market segment, it wouldn't be too expensive. It'll be up against the Xeons, which, in their own rights, aren't cheap. And the performance boost possible would be worth it.

It just concerns me that Intel could not just meet, but exceed AMD's efforts here. Then where would AMD be? Not good. AMD is taking way too long to give us this and the impressiveness drops yet lower every time Intel releases a new chip.

You gotta remember those numbers are probably for the x87 FP performance. Hammer will have SSE2 as well. There's no telling how well the SSE2 unit on Hammer will work. It may indeed surpass that of the P4 (in fact, it's very likely).

"We are Microsoft, resistance is futile." - Bill Gates, 2015.

slvr_phoenix · Oct 16, 2002

Designing an MPU improvement isn't something you can just do with a snap of a finger. There are limitations to how well you can do things. AMD is currently stuck with the original design they got from Next Gen. They don't have enough time or resources to build another core from the ground up like the P7 nor can they afford to just keep going with the Athlon. So they improve it a little and add x86-64 to try to get more profits from another market to perhaps fund the designing of a new MPU.

Yeah, but Hammer hasn't exactly been designed at the snap of a finger either. AMD has been working on that ah heck for years. Hell, it's been delayed for years. Heh heh. (If not litterally, it at least feels like it.)

But yeah, I see your point too. Maybe AMD's next core (whenever they get around to that) will be a significant redesign or new core all together.

Compared to other processors that are used in that market segment, it wouldn't be too expensive. It'll be up against the Xeons, which, in their own rights, aren't cheap. And the performance boost possible would be worth it.

That's true. Intel has kind of been running their Xeon business a little funny. But then, they've been slow to put the Itaniums on the best process too... Come to think of it, Intel has a kind of funny attitude towards their server chips. I mean it makes sense, but it's still kind of funny. I wonder what their server chips were like if they were to give them the same priority and treatment as their desktop chips...

You gotta remember those numbers are probably for the x87 FP performance. Hammer will have SSE2 as well. There's no telling how well the SSE2 unit on Hammer will work. It may indeed surpass that of the P4 (in fact, it's very likely).

That's true. Hammer supposedly will have SSE2 anyway.

So maybe that'll help. I guess it all depends on how well AMD implements SSE2.

One thing that worries me though is that people have been expecting Intel to fix their FPU performance, well, ever since Willy. Most people thought it would be in Northy. Then nada from Intel. Now most people think it'll be in Pressy. If it actually is ... AMD could be in trouble.

Though maybe that's why Intel hasn't done it. Heh heh.

<A HREF="http://forumz.tomshardware.com/community/modules.php?name=Forums&file=viewtopic&p=45775#45775" target="_new">Join the THGC LAN Party!</A>

Dark_Archonis · Oct 16, 2002

Indeed... if those benches are true, then Opteron has some pretty interesting performance considering it's running on x86. But remember, Spec benches don't tell you alot when it comes to real-world applications. Still, though, those integer benches sure are impressive, when compared to Itanium 2. If you add optimized compilers, and if the opteron runs in 64-bit, then the performance may increase about 30-50%. Also FYI, clawhammer is supposed to be released at about 2.4 - 2.6 Ghz. Opteron, as far as I know, will not be released at such high frequencies. I'll be surprised to see Opteron released at a frequency higher than 2Ghz. Indeed, according to these benches, the Opteron performs similarly to what I predicted. It still though won't match the performance of Madison, and clock for clock, Itanium 2 can still run circles around Opteron in terms of performance. Of course, Itaniums have several problems (cost, heat, availibility, support, etc.). I wasn't really expecting Opteron to perform extremely well. Considering the fact that Opteron is an x86 CPU, I knew that there would be a few performance "limitations". Accroding to these benches, clawhammer no longer seems to be a threat to Prescott. According to texas_techie, clawhammer's performance would be "about" equal with Prescott. Personally, I don't believe that, and these benches confirm my beliefs. I'm also afraid of how Opteron will fare against Prescott.

texas_techie made several optimistic claims with regards to the performance of the hammer. Most people on these forums (except the AMD fanboys) don't believe them. I for one do not believe his performance claims of hammer.

imgod2u, your info seems pretty accurate, and the most important thing for racks is heat. Opteron needs to have good heat dissipation, as well as a fairly low heat output, or it won't do good in that market. BTW texas claimed that the "hammer" (he did not mention which hammer) had 100% better performance than a 2.8 P4 in Specviewperf, not Spec INT.

Unfortunately, the clawhammer's future doesn't look too good. It seems that it won't be able to take the performance crown away from Intel in the desktop market. In 64-bit, Opteron will have a very hard time competing (performance-wise) against Intel's IA-64 juggernaut.

If AMD can offer decent prices on Opteron, and if Opteron has good heat dissipation/low heat output, then it just might be a success.

In short, it seems that AMD has a bigger chance of succeeding with Opteron than with clawhammer.

- - -
All good things must come to an end … so they can be replaced by better things!

Tapout · Oct 16, 2002

That's the problem though. I'd have thought that AMD would at least have made them more attractive to the majority of their customers...

AMD needs NEW customers. The success that AMD has enjoyed over the past few years was becuase of new customers buying their product. What percentage of those people do you really think are waiting for Hammer so they can buy one? My bet is maybe 10% of the FanB... erm Enthusiasts. The rest of the world is just going on with business. Need a PC? Buy one. Need a server... buy Intel.

Enter Opteron.

I think the true goal is to break up the Xeon monopoly on the workgroup server end. After researching servers, hardware-acquisition types will see the price/performance advantage of AMD, and then hopefully, Athlon will finally be embraced by corporate IT.

Anyway, I'm probably way off, but it just seems to me that what AMD needs most is new customers.

<This space left intentionally blank to avoid the embarrasment of not having a sig>

eden · Oct 16, 2002

After doing some speculative calculations, the Opteron is as AMD promised, or near that: It is about 60% better per clock than the same AthlonXP. Do some calcs, and you'll see a 2GHZ AXP performing as much as this weak compared to it.
Speculatively, these are not real world apps, and the FP benches don't even use SSE2. It still managed to beat the 800MHZ higher beast. Now imagine SSE2 apps, I'd say like imgod2u, it is very likely to do destructive damage. I'd love to see Opteron at 2.8GHZ, WITH SSE2 in SSE2 tests, that would for AMD, (despite being early yet, and a bit overdone) boost the morale of many followers and would definitly tell people to hold a bit. I don't expect Prescott to be a big boost, Intel themselves made it clear it won't be a big core improvement. The extra FSB will allow about 10% better IPC, and the extra cache, well it depends if some apps have had enough of the 512K L2, so we'll go with 8%.
Overall, from 15-25% better IPC, Prescott will definitly not be the be-all end-all of Desktop chips.
[enable optimistic mode]
AMD theorized Opteron will have about 50% better IPC, though here it has 60%, I'd say CH will have 30% better IPC than AXPs, add SSE2 and clock it around 2.6GHZ, not to mention the chance of a good chipset from nVidia or VIA for Hammer platforms, and there is still a chance for CH to be up there against P4s and Prescott. With Prescott included there should be around a 5% tradeoff between both in benches.
[/end optimistic mode]
Of course any of this is theoretical, and results can be worse or better, depending on how the outcome is. Intel can have higher Prescott speeds, AMD can end up with lower CH speeds. Who will outpace who?
I'd say stop trying to be pessimistic, but also don't hype, simply stay neutral like me, be impressed and comment at benches thrown but DO NOT expect yet.

--
What made you choose your THG Community username/nickname? <A HREF="http://forumz.tomshardware.com/community/modules.php?name=Forums&file=viewtopic&p=19957#19957" target="_new">Tell here!</A>

Arrix · Oct 16, 2002

Dont forget that those SPEC numbers were for 32-bit code, they expected about a 20% increase in those numbers for 64-bit native code. Lets see a Itanium 2 get those numbers running in 32-bit X86 mode. NOT! Anyway, Opteron is not going to compete directly with Itanium, at least not early on. Its going mainly after Xeon server systems.

sonoran · Oct 17, 2002

Was it just me, or did anyone else notice that the official presentation material (<A HREF="http://www.amd.com/us-en/assets/content_type/DownloadableAssets/Fred_Weber_MPF_2002_Presentation.pdf" target="_new">link to presentation</A>) shows the numbers as "Estimated score = ..."?

Did they present actuals at some point, or were the figures purely estimates?

* Not speaking for Intel Corp *

imgod2u · Oct 17, 2002

I'm not too sure on that. They specified the specific compiler they used so I'm guessing it should be actual scores. Another thing to note is that Prescott will not just be core enhancements towards IPC, it will also be able to scale much higher. With 90nm transistor size and strained silicon technology, it should be able to scale quite high along with this rumored "SSE3", which, hopefully, will add the x87 flexibility to the new extensions. If it completely replaces x87 and software developers embraces it, AMD will be in a world of trouble because their one huge advantage currently is their strong x87 FPU.

"We are Microsoft, resistance is futile." - Bill Gates, 2015.

Dark_Archonis · Oct 17, 2002

Correct me if I'm wrong, but spec FP benches for the P4 also do not use SSE2. According to my knowledge, spec CPU 200 doesn't take SSE2 into account. Obviously, you can't be comparing the 2.8 Ghz P4 with the Opteron, I mean that's just not fair. It's quite obvious the Opteron can beat the 2.8 P4, but not by a whole lot. Prescott should narrow the gap. Of course, I never stated that Prescott would be a very big core improvement, I simply stated that CH would have a hard time competing with Prescott, according to these spec benches. I agree that it'll be roughly close between Prescott and Clawhammer, in terms of performance. Tejas, though, is set to be a bigger core improvement than either Prescott, or Northwood. AMD has worked on the K8 for over 5 years, and AMD isn't expected to make another core revamp as big as this until 2005. 2005-2006 is when K9 is slated to be released, but that's an early release date by AMD's standards. I'm worried about what AMD's going to do in 2004 and early 2005, because that's when Intel will release Tejas, a major revamp of the P4, and in early 2005, Intel is set to release the mysterious "P8" or Nehalem. This is all speculation, but it seems everyday that Intel is accelerating their production and release schedule. For example, the 3.06 P4 was set to be released in Q1 of 2003, but Intel moved it up to Q4 of this year. Intel has basically been "delay-free" for over a year now. Intel has really started to work full steam on a whole whack of CPU's. A major blow for AMD will be Banias, which is simply superior to any other mobile CPU out there. Also, who knows how hyperthreading will affect performance in P4's? I'm still not sure of how AMD will continue to survive. AMD barely has any cash left (they only have about 1 billion US left in actual company money) and their profits are really thin. Intel's profits are also thin (about 700 million) compared to AMD, which has less than 200 million), but Intel has invested a HUGE amount of cash in R&D this year, and Intel announced that revenue went up in Q3 of this year, though it didn't meet investors' expectations. Anyways, I'm speculating alot because I REALLY want Intel or AMD to come up with a radical new core. Intel has done this with Banias, but it's for the mobile market. All Intel has to do is to create an amazing core like Banias, but for the desktop market. Unfortunately, AMD deosn't have the resources to invest into a creative and new core like Banias.

Hey arrix, let's see Opteron get those benches running at 1ghz. It's not going to happen. The performance of Itanium 2 is a testament to the wonderful things that the IA-64 ISA can do. It's a real bummer that AMD decided to go x86-64. I really hope that Intel's desktop CPU's go IA-64 as soon as possible, because there are alot of wonderful benefits if they do. x86 is over 20 years old, and frankly, I'm sick and tired of it. It's such a big performance bottleneck for modern CPU's. Anyways, Itanium 2 emualtes x86, so it'll never be able to run it natively, since it uses a different ISA. Also, you're correct, Opteron is targetted mainly at P4 Xeons.

imgod2u, Indeed, Prescott will be able to ramp extremely fast, and that may be a big problem for AMD and it's K8. I never knew that SSE3 was supposed to help x87 on Prescott ... can you elaborate?

- - -
All good things must come to an end … so they can be replaced by better things!

juin · Oct 17, 2002

But it has huge potential for webservers.

Webserver are manly built in california so the need of low power cosumation ir critical.Some may have read google story with there itanium 2 (true or not) yes itanium performance is much more that does from SUN but power consumation is 1/2 of Itanium.SO google is locking for blade server or sun.The same happen to IBM power cosumation is just too high right the market look for power and low cosumation.I think banias SMP version serais un tres bon server like P3-S in U1 rack.Opteron need to find what the market fit for it.Must be a linux base market need for 64 bit adressing and high performance ALU and medium power cosumation.

Now what to do??

juin · Oct 17, 2002

You gotta remember those numbers are probably for the x87 FP performance. Hammer will have SSE2 as well. There's no telling how well the SSE2 unit on Hammer will work. It may indeed surpass that of the P4 (in fact, it's very likely).

on the bench this can be only for the peak score they got be optimaze with spike feature or extensive set of extention for each part of the bench like SUN and many others like to do.Also acording to intel SSE 2 give 5 % increase in SPEC FP over X86.

Now what to do??

juin · Oct 17, 2002

there score are estimation when reach release clock speed.

Now what to do??

eden · Oct 17, 2002

So you're saying they have not yet gotten to 2GHZ?
I don't believe it, most news and media, had said it has reached 2GHZ by now.

--
What made you choose your THG Community username/nickname? <A HREF="http://forumz.tomshardware.com/community/modules.php?name=Forums&file=viewtopic&p=19957#19957" target="_new">Tell here!</A>

juin · Oct 17, 2002

Does it official no i think there are at 1.2 big max if not the 800 mghz on normal heatsink

Now what to do??

imgod2u · Oct 17, 2002

on the bench this can be only for the peak score they got be optimaze with spike feature or extensive set of extention for each part of the bench like SUN and many others like to do.Also acording to intel SSE 2 give 5 % increase in SPEC FP over X86.

Where does Intel state this? I don't remember this and that 5% is certainly very low in terms of the results we see. Besides, that is Intel's implementation of SSE2. We don't know how Hammer is implemented and given the fact that AMD wants it to be a powerhouse, I'd say they may implement it significantly different than how Intel does it. Maybe even give the issuing to the SSE/SSE2 units a full 128-bit data path.

"We are Microsoft, resistance is futile." - Bill Gates, 2015.

2GHz Hammer at MPF

Distinguished

Distinguished

Splendid

Distinguished

Splendid

Distinguished

Distinguished

Splendid

Distinguished

Splendid

Distinguished

Splendid

Distinguished

Distinguished

Champion

Distinguished

Distinguished

Distinguished

Distinguished

Distinguished

Distinguished

Distinguished

Champion

Distinguished

Distinguished

Share this page