AMD limits Clawhammer to 800MHz

[MadRat]

"The demos we're currently showing are anything but high-performance. They're pretty low-performance right now and we don't want to invite questions about just how fast they are running."

"low performance" <> "poor performance"

Your source has taken the entire quote totally out of context. What they said is that the demoes are not showing a heavy workload, not that Hammer is suck.

 

[Nemesis77]

Originally posted by: majewski9
I am alittle bit wary about the onboard mem controller. A lot of people are criticizing it for limiting memory bandwidth. I know it has a DDR controller intergrated and this could limit mem bandwidth to 2700 with DDR333. will different chipset allow for DDR 400 or even DDR II or can they work with hammer's mem controller? Is memory bandwidth all that important? I mean RDRAM has tons of memory bandwidth, but we know that DDR SDRAM and the right chipset can perform better even though it has less memory bandwidth. Another point is onboard graphics are going to suffer from the onchip mem controller.

If you have onboard graphics, you can always use external frame-buffer with it (like SiS does with it's Hammer-chipset).

As to the compatibility with future memories. There's no point yet to build DDR-II mem-controller in Hammer. DDR-II is not available, it would just eat more transistors and it would drive up the costs. When Hammer get's its first die-shrink (in second half of 2003) it will get DDR-II controller. And while Hammer has been designed to work with DDR333, I wouldn't be surprised if it supported DDR400 as well.

As to the bandwidth. That is up to the design of the CPU. P4 has been designed to work with massive amounts of mem-bandwidth. Not so with Athlon. Athlon is competetive even with lesser amounts of bandwidth, whereas P4 is crippled by it. Same thing applies to Hammer I guess. And you must remember that Hammer still has more bandwidth that Athlon does, and what's more important, it has less latency.

 

[etalns]

*Begins to breath normally* Ooh boy, can you next time make the title a tad bit nicer 😉

 

[neuralfx]

dullard, I see what you're saying about your comments on the article ..

I'll comment on the deal though .. Ok basically what the guy was doing, correctly so, was what's called "CYA". They gave locked chips to mobo manuf's so they wouldn't overclock it (ie change the performance) in some way and report X performance .. Which if for whatever reason Retail/Released chips didn't meet X performance .. then well he would probably get fired ..
On the PR 3400+ theory .. I don't know how well that holds water, the Hammer series is a different architecture .. so how could you really compare Hammer's PR to XP's PR for a direct comparison to an AMD to Intel Pentium4 PR, which should be an AMD to Itanium PR, but which would you choose .. confused yet?
eh but I would agree with you about the release date .. I'm still waiting on Mustang =) ..
-neural

 

[Bluga]

Originally posted by: ElFenix
still faster than itanium 😉

 

[Sunner]

Not that it has a whole lot to do with this discussion, but I'd like to know where Dullard came up with the 1350 SpecFP number for a 1 GHz Itanic(which isn't available AFAIK)?

The 800 MHz Itanic with 4 MB L3 achieved around 700 in both peak and base IIRC, and I know some people have estimated the McKinely to score around 1300-1400 upon introduction.

 

[dullard]

Originally posted by: Sunner
Not that it has a whole lot to do with this discussion, but I'd like to know where Dullard came up with the 1350 SpecFP number for a 1 GHz Itanic(which isn't available AFAIK)?

The 800 MHz Itanic with 4 MB L3 achieved around 700 in both peak and base IIRC, and I know some people have estimated the McKinely to score around 1300-1400 upon introduction.

You are right that it isn't yet available, but 1 GHz Itanium 2 is supposed to be available any time now. Rumors from earlier in the year were for a late June release. Intel says it will start production mid-2002. So unless we are further updated, June/July looks like the release.

I got the 1350 number from a link smack dab in the center of Intel's home webpage. I've been surprized that no one has mentioned it at all on these forums. That score will break the record by 12%.

 

[MadRat]

He got it from this statement:

Intel also expects the Itanium 2 processor to deliver exceptional processor and system performance:

Memory bandwidth: Single-processor Itanium 2-based systems using the Intel E8870 chipset are expected to deliver 3.7 GB/s on the Stream* benchmark, which is more than 2.5 times that of a comparable Itanium-based system and more than four times the performance of a comparable UltraSparc III system at 0.891 GB/s.

Floating point: On the SPECfp2000* benchmark, single-processor Itanium 2-based systems are expected to achieve industry-leading performance with a base result of 1350, almost doubling the base result of 701 on comparable Itanium-based systems.

Integer: Intel estimates that on the SPECint2000* benchmark, single-processor Itanium 2-based systems will have a base result of 700 or higher, approximately twice that of the Itanium processor. Other published SPECint2000 results can be found on www.spec.org.

Compare to:
Precision WorkStation 730, 800 MHz Itanium: 314/314
AlphaServer ES40 Model 6/833: 518/544
PowerEdge 1500SC, Intel Pentium!!!-T 1.26GHz: 611 623
Precision WorkStation 530, Intel Xeon 2.0GHz: 642/663
PowerEdge 1500SC, Intel Pentium!!!-T 1.4GHz: 648/664
Epox 8KHA+ Motherboard, AMD Athlon XP1900+: 677/701
Epox 8KHA+ Motherboard, AMD Athlon XP2000+: 697/724
Epox 8KHA+ Motherboard, AMD Athlon XP2100+: 720/749
Precision WorkStation 340, Intel Pentium 4 2.0GHz: 640/653
IBM eServer pSeries 690 1.1GHz: 680/708
Precision WorkStation 340, Intel Pentium 4-A 2.0GHz: 738/759
Precision WorkStation 530, Intel Xeon 2.2GHz: 788/810
Precision WorkStation 340, Intel Pentium 4-A 2.2GHz: 790 811
IBM eServer pSeries 690 Turbo 1.3GHz: 804/839
Precision WorkStation 530, Intel Xeon 2.4GHz: 824/851
Intel D850EMV2 motherboard , Intel Pentium 4-B 2.4GHz: 852/864
Intel D850EMV2 motherboard , Intel Pentium 4-B 2.53GHz: 882/896
Precision WorkStation 340, Intel Pentium 4 2.53GHz: 893/922


(NOTE: Results = Base/Peak)
Source:Spec CINT2000

 

[dullard]

Since we are talking about SPECfp, I decided to list them (although your SPECint scores do show the Itanium?s weakness - if you can call about equal in speed to current Athlon XP?s and P4s a weakness).

Compare to:
Precision WorkStation 730, 800 MHz Itanium: 645/---
AlphaServer ES40 Model 6/833: 590/658
PowerEdge 1500SC, Intel Pentium!!!-T 1.26GHz: 422/440
Precision WorkStation 530, Intel Xeon 2.0GHz: 743/765
PowerEdge 1500SC, Intel Pentium!!!-T 1.4GHz: 437/456
Epox 8KHA+ Motherboard, AMD Athlon XP1900+: 588/634
Epox 8KHA+ Motherboard, AMD Athlon XP2000+: 596/642
Epox 8KHA+ Motherboard, AMD Athlon XP2100+: 613/660
Precision WorkStation 340, Intel Pentium 4 2.0GHz: 715/734
IBM eServer pSeries 690 1.1GHz: 1017/1075
Precision WorkStation 340, Intel Pentium 4-A 2.0GHz: 744/764
Precision WorkStation 530, Intel Xeon 2.2GHz: 779/802
Precision WorkStation 340, Intel Pentium 4-A 2.2GHz: 779/801
IBM eServer pSeries 690 Turbo 1.3GHz: 1202/1266
Precision WorkStation 530, Intel Xeon 2.4GHz: 803/825
Intel D850EMV2 motherboard , Intel Pentium 4-B 2.4GHz: 840/873
Intel D850EMV2 motherboard , Intel Pentium 4-B 2.53GHz: 861/873
Precision WorkStation 340, Intel Pentium 4 2.53GHz: 878/901

I will stand corrected, looking at the IBM pSeries 690 Turbo 1.3 GHz, the Itanium 2 will only beat it by 12% not 30% - sorry I haven?t been keeping up with scores.

 

[Sunner]

Until it's available, it's a moot point, and it is intel's own estimates.
Not saying I think anyone is lying, Im just saying, numbers aren't worth jack until you can prove them.

I seem to remember nVidia estimating that the TNT would be roughly as fast as V2-SLI once a long time ago.

 

[dullard]

Originally posted by: Sunner
Until it's available, it's a moot point, and it is intel's own estimates.
Not saying I think anyone is lying, Im just saying, numbers aren't worth jack until you can prove them.

True, numbers like that should be taken with a grain of salt. However, the only thing Intel can do in the next month is minor optimizations - it isn't like they have time to completely redesign the chip. These optimizations are why they have to publish estimates and not cold hard data - since the numbers will surely change a few percent in the coming weeks. It really doesn't matter at this stage if the score is 1330 or 1370 instead of 1350.

However if I were to trust anyone at this point in time, it would be Intel. Why? Because they have invested so much money into the Itanium, the last thing they need is some exaggerated claim 1 month before release that will cause tons of bad publicity. Plus these estimates are only double what the original Itanium had and with the architectural improvements, optimizations, and hefty 25% clock speed boost a 100% performance boost seems quite reasonable. If they claimed a 1000% boost, it would have been another story.

I'm not too interested in the Itanium 2 myself - but I think the Itanium 3 vs Opteron battle will be quite interesting to watch (since they will be released at about the same time).

 

[Pariah]

The starting costs for an IBM pSeries690 is $450,000 for an 8-way system (1.1GHz). The cheapest system using the Power4 CPU is the pSeries670 starting at $175,000 for a 4way system (1.1GHz). You want 1.3GHz? How about a cool million for a 16way system? In comparison, a quad Itanium 800 can be had for under $50,000. The Itanium 2 will cost more, no doubt, but not $125,000 more for a quad system. Doesn't really sound like a fair fight. Comparing an Opteron to an Itanium makes even less sense as one has been out a year the other won't be out until next year. What kind of comparison is that?

 

[dullard]

Originally posted by: Pariah
The starting costs for an IBM pSeries690 is $450,000 for an 8-way system (1.1GHz). The cheapest system using the Power4 CPU is the pSeries670 starting at $175,000 for a 4way system (1.1GHz). You want 1.3GHz? How about a cool million for a 16way system? In comparison, a quad Itanium 800 can be had for under $50,000. The Itanium 2 will cost more, no doubt, but not $125,000 more for a quad system. Doesn't really sound like a fair fight. Comparing an Opteron to an Itanium makes even less sense as one has been out a year the other won't be out until next year. What kind of comparison is that?

That is the point of the Itanium - to perform the same as the proprietary chips at a fraction of the cost. I really have no clue what the pricing will be on the Itanium 2. The Itanium costs $1000-$2000 for processors with 2 MB of cache and $4000 for processors with 4 MB of cache. Obviously the cache is the major part of the total cost. The Itanium 2 sits in the middle with 3 MB of cache. So if I had to guess, I'd say the Itanium 2's extra features and less cache will balance out and keep the price at $4000 per processor.

People keep saying that Opteron and Itanium won't compete - but I think the exact opposite is true. Opteron probably makes more sense in the 2-4 processor situations while the Itanium will probably be better in the 8+ processor situations. However Opteron can still compete in that situation. According to AMD, "The processor architecture directly supports 1-8 AMD Opteron processors in a glueless (requiring no chipset support) multiprocessor configuration. A platform can support an unlimited number of AMD Opteron processors in a multi-way configuration with external/additional chipset support." Thus it may be possible to see a 100 processor Opteron platform competing with a 100 processor Itanium platform. For businesses that need 64-bit capabilities, these will be direct competitors. As far as I'm concerned the others (IBM, Sun, SGI, etc.) are out of the picture unless they can reduce prices dramatically.

Opteron realistically won't be here until the end of the 1st half of 2003. Itanium 3 is rumored for June 2003. To me those release dates are so close that they will make a great comparison. Comparing Itanium 1 to Opteron would be stupid I agree.

 

[Sunner]

I dont see Opteron competing with the Itanic simply cause the two companies are aiming them at different markets.

AMD is aiming the Opteron at the mid-high end x86 server/workstation market.
Intel is aiming the Itanic at the enterprise market.

I doubt the Opteron will have features such as hot pluggable CPU's, hardware domains, etc, while Im quite sure the Itanic will have these(haven't looked into it much, just guessing, but if they don't they can kiss the enterprise goodbye).

Of course they might compete where the two markets meet, just like a high end x86 box might compete with Sun's "SPARC pc servers", such as the SunFire 280r, 420r, and V880, but I wouldn't exactly call that head to head.

 

[spanky]

Originally posted by: MrGrim
That title is going to scare a lot of ppl. 🙂

 

[Mrburns2007]

Geez this is just FUD, there probably 2.0 Ghz units that are underclocked so that Intel can't tell what the real performance will be when they released. It's smart to just ease them out and not let your competition know what kind of performance they can acheive.

 

[Piano Man]

I just about crapped my pants until I read the whole post.

 

[CedarTeeth135]

I think AMD will deliver, but 800 mhz. If they underclock it that much, why lock it. Most AMD buyers are computer enthusiasts and would love to unlock the potential of the upcoming Clawhammer.

Even if it debuts with a rating of 3400 the Pentium 4 3.4 will most definantly overclock very nicely, making it better, in my opinion.

 

[Adul]

Originally posted by: CedarTeeth135
I think AMD will deliver, but 800 mhz. If they underclock it that much, why lock it. Most AMD buyers are computer enthusiasts and would love to unlock the potential of the upcoming Clawhammer.

Even if it debuts with a rating of 3400 the Pentium 4 3.4 will most definantly overclock very nicely, making it better, in my opinion.

you are assuming that a P4 3.4 will be out a that time.

 

[grant2]

Most AMD buyers are computer enthusiasts and would love to unlock the potential of the upcoming Clawhammer.

I think you're wrong about that, most computer buyers of ANY processor do not overclock.

Especially in business/server environments where AMD is hoping to make gains with this chip!

 

[CedarTeeth135]

I think you're wrong about that, most computer buyers of ANY processor do not overclock.

Well, ok. I know that most people buy AMD are either buying a Compaq or other premade computer and have no computer knowledge or they are building their own computer. Just look at Anandtech for an instance, what cpus does everyone most likely have, AMD, and most of us know quite a bit about computers. I just think it would be a silly move to lock the Clawhammer.

If Intel releases a cool 3.4 around the same time AMD releases it's locked 3400+, and they are both reasonably priced, I think I would have to move back to intel. I love my Athlon right now, but I just can't see how AMD can lock this Clawhammer. I would just love to push it as far as I can.

 

[WilsonTung]

http://www.theinquirer.net/06060203.htmMore bad news - Looks like the hammer family has a few problems and mobo makers aren't too happy.

Although, as we've reported earlier, you can find boards at many places at the Computex show, two separate sources at chipset companies have told us that the Clawhammer, for example, is far from ready and may be delayed even beyond December. One said that realistically we were talking nine months from now

Intel is going to have a field day if this turns out to be true!