ClawHammer to Perform 30-50% Better Than Athlon XP at Same Clock Speed

[AGodspeed]

Jerry Sanders said one salient thing yesterday at the Morgan Stanley Technology Conference. He said that the Hammer will perform 30-50% better than the Athlon XP at the same clock speed running 32 bit applications.

Link

 

[Athlon4all]

If this is the case, then the inital Hammer should be able to beat a Pentium 4 clocked as high as 3.5GHz, and should blow away what it will be facing initally (a 2.8 p4). Do you think that that is really possible? I don't think Jerry would say that unless he really meant it, but is that really possible?

 

[dullard]

Has anyone given an official initial speed for the Hammer? The 30%-50% boost will be great if the Hammer actually runs at around 2 GHz, but it will suck if the first few Hammers are 1.5 GHz.

My thoughts: Clock for clock an Athlon is about 25% faster than a P4. When the P4 moves to 133 MHz fsb this will be reduced to a 15% clock for clock speed boost. Now multiply that speed boost by 40% (an average of that 30% to 50% range). So the Hammer should have about a 60% boost clock for clock over the Intel P4. Thus a 2GHz Hammer will be about the same as a 3.2 GHz P4. However a 1.5 GHz Hammer will only be able to compete with a 2.4 GHz P4.

 

[Mavrick007]

Wow, the Hammer is really looking sweet from the standpoint right now. Alot of it's hype until I see the real world benches.

 

[MadRat]

Rumour has it the initial silicon was targeted for only 1.6GHz, meaning it isn't quite ready for the big time yet.

 

[ElFenix]

damn... thats a huge improvement. how do they do that and make the mhz more scalable? more pipes? better branch prediction? wider path to cache? make the L1 a trace cache so stuff doesn't have to be decoded every time?

 

[johndoe52]

I'm gonna look for it but does anyone know when Hammer is going to be released? It's still going to be socket A too right?

 

[PlatinumGold]

johndoe

hehehe, where you been man. no way the hammer is socket a. do some research check out the pin out on these babies.

 

[spanky]

stop... hammer time!

one thing is... is amd going to mass produce the solo board, or license the amd8000 chipset for others mobo manufacturers to use?

 

[AGodspeed]

A lot of hardware analysts (including Anand himself) have speculated that ClawHammer will debut at around 2.0GHz, which is certainly attainable with .13-micron SOI and a 12 stage integer pipeline (compared to the Athlon's 10 stage integer pipeline and the P4's 20 stage integer pipeline). It?s also been speculated that ClawHammer will have an 800MHz FSB.

 

[BD231]

If AMD says it will be 30 to 50% faster than an XP at the same clock then you can trust that, they have been really good about delivering what they promise.

 

[dullard]

<< A lot of hardware analysts (including Anand himself) have speculated that ClawHammer will debut at around 2.0GHz >>



That is my problem, it is just too much speculation right now. Madrat claimed 1.6 GHz was a target. I've seen others state that 1 GHz will be the initial speed.

According to my rough (extremely rough) calculations above, then the Hammer must start at 1.75 GHz to match the current P4 available at that time. However since I think Hammer will be expensive and will require a new MB - it really needs to be 2.0 GHz or more to have much success. With such a wide variety of estimates on its initial frequency, the initial Hammer's fate is still in a cloud of mistery.

Some of the roadmaps floating around show that the Hammer will quickly ramp up in the first few months after its release. If these are true (who really knows) I'd have to say it is a lot easier to start low and ramp fast, than to start high and ramp fast.

 

[AGodspeed]

That is my problem, it is just too much speculation right now. Madrat claimed 1.6 GHz was a target. I've seen others state that 1 GHz will be the initial speed.

I'd like to see any credible web site that claims that ClawHammer will possibly debut at 1.6GHz or lower. I've followed Hammer very closely and I've never seen or heard "1GHz" or "1.6GHz" as the possible introduction clock speed for ClawHammer. Heck, AMD's ~30 day old ClawHammer silicon showcased at IDF last week was quoted as running no worse than other high-end 64-bit processors (800MHz-1GHz presumably, as Anand said himself).

I'm guessing it'll take AMD another ~ 6 months of debugging, validation, etc. before final ClawHammer's are sent to OEMs and through the retail channels. This would give AMD enough time to work on getting acceptable yields to ramp up Claw and Sledge Hammer supply and it would be enough time for OEMs and such to have ClawHammer systems ready to go. In fact, if Claw and Sledge yield well enough, they may even be introduced around November, just in time for the X-mas season.

 

[Eug]

So what am I talking about here for my next desktop upgrade at a reasonable price?

Clawhammer 2.2 GHz in Sept 2003 with 512 KB L2 cache and 1 GB single channel DDR. How do the FSB numbers work, since supposedly the DDR is no longer dependent upon the FSB per se anymore right? 400 MHz memory bus x 2 x 8 = PC3200, but will FSB speed be the same? What about Sledgehammer and dual channel DDR, and how do those numbers work?
AGP 8X with ??? card
Serial ATA
Firewire 800 Mbps
USB 2.x I don't care all that much about USB 2 though.
No more floppy drive or serial port!
DVD+/-RW drive

Despite it being 1.5 years from now I'd probably still use a couple of ROM and hard drives from my current machine.

 

[AGodspeed]

Do you think that that is really possible? I don't think Jerry would say that unless he really meant it, but is that really possible?

Yes, I do. Looking at the Compaq's EV7, it seems that the premiere technology companies (except Intel) are taking the same approach to CPU architecture as AMD has taken with the K8.

 

[Anubis]

agodspeed: It?s also been speculated that ClawHammer will have an 800MHz FSB.

ok if thats true im gonna crap my pants. i cant even imagin how fast that freekin thing would be

 

[SexyK]

Sounds great, but I don't care how "closely" anyone's been following the Hammer, but there is no gaurentee that these will come out at anywhere near 2.0Ghz. It seems like AMD is having enough problems with the .13u process on the Tbred core... hopefully that will be straightened out by the time hammers are coming out, but I'd say 2.0GHz is far from a definite figure.

Kramer

 

[MGMorden]

<<
No more floppy drive or serial port!
>>



Whether or not you like them, they need to stay there. Whenever you bring up whether or not a floppy is useful it always turns into an argument. The fact that there is an argument means the port should stay (afterall why piss off half of your potential customers just to not add something that people who don't like it don't have to use). Serial ports are also still used a lot (TI calculators still largely use Serial port graphlinks for PC's. Last time I checked the USB links didn't even work on PCS, only Macs).

This doesn't even affect me as I don't use serial stuff and my floppy drive is IDE based, but it would me mindless for them to remove these connections.

 

[dullard]

<< quoted as running no worse than other high-end 64-bit processors (800MHz-1GHz presumably, as Anand said himself). >>



See it isn't quite clear. Does that mean the current Hammer is only at 800 MHz to 1 GHZ? That is a lot of scaling left to do if they want a 2 GHz model in 6 months. And 2 GHz will only barely beat the competition. They really need more speed than that for a great launch.

Don't get me wrong, Hammer may be great by Sept 2003 when Eug wants one. However, I've yet to see anything that shows it will be great in Dec 2002.

 

[Rand]

<< agodspeed: It?s also been speculated that ClawHammer will have an 800MHz FSB.
>>


Well now technically speaking it doesnt HAVE a FSB, at least in the classical definition of the term.


Ealier statements have for quite some time indicated a 25-30% per clock performance boost over the Palomino core. Given the revised statements of a 30-50% per clock performance boost I think it reasonable to guess that Jerry Sanders is revising past statements made that were based on software architectural simulations of the Hammer's performance... he also seemed to hint at this a couple times in the past.
The newer statements being based upon the performance of the actual early engineering sample silicon that has recently become a reality.



<< how do they do that and make the mhz more scalable? more pipes? better branch prediction? wider path to cache? make the L1 a trace cache so stuff doesn't have to be decoded every time? >>



AMD's tech docs on the Hammer architecture are still sitting at their site open for anyone that cares to study them.



<< If AMD says it will be 30 to 50% faster than an XP at the same clock then you can trust that, they have been really good about delivering what they promise >>



Sorta like the K6 core's supposed superior FPU performance over the P6 core. Or the K6-2 300's comparable FPU performance to the PII 400MHz?
If he's incorrect in his statements it wouldnt be the first time AMD has been less then 100% accurate.



<< That is my problem, it is just too much speculation right now. Madrat claimed 1.6 GHz was a target. I've seen others state that 1 GHz will be the initial speed. >>



AMD's original indications of well over a year ago hinted towards a 2GHz release they were aiming for, since then they have been quite reluctant to even hint at a release speed however. I have not seen any credible statements that have hinted at any lower clockspeed then 1.6GHz though, and most indications are strongly pointing towards the 1.8-2.2GHz clock range.
The only thing we know for a fact is that it is still scheduled to be released with an initial model rating of 3400+.



<<
I'm guessing it'll take AMD another ~ 6 months of debugging, validation, etc. before final ClawHammer's are sent to OEMs and through the retail channels. This would give AMD enough time to work on getting acceptable yields to ramp up Claw and Sledge Hammer supply and it would be enough time for OEMs and such to have ClawHammer systems ready to go. In fact, if Claw and Sledge yield well enough, they may even be introduced around November, just in time for the X-mas season. >>



I'm still wondering just how in the hell they expect to get the Hammer in release by the end of this year. Given they've only recently attained A0 stepping silicon that gives them less then 8 months for debugging, testing, validation etc etc.
Such a pace would undeniably be an incredibly quick intro, far faster then anything yet attempted on such a large scale.
In the past one has seldom seen such a pacing of less then 1.5 years let alone a mere 8 months, it will truly be amazing if AMD does hit their targeted release date though for all intents they certainly appear confident they will hit a Q4 2002 release.
Personally I'm still betting on it being delayed until Q1 2003.



<< seems like AMD is having enough problems with the .13u process on the Tbred core... >>



Quite a few problems indeed from all indications.

 

[AGodspeed]

<<

<< quoted as running no worse than other high-end 64-bit processors (800MHz-1GHz presumably, as Anand said himself). >>



See it isn't quite clear. Does that mean the current Hammer is only at 800 MHz to 1 GHZ? That is a lot of scaling left to do if they want a 2 GHz model in 6 months. And 2 GHz will only barely beat the competition. They really need more speed than that for a great launch.

Don't get me wrong, Hammer may be great by Sept 2003 when Eug wants one. However, I've yet to see anything that shows it will be great in Dec 2002. >>

I'm still a little confused with your logic dullard. Bare with me for a second.

First, take into account the following two facts:

1. ClawHammer will be able to scale higher than the Athlon XP in pure frequency due to an integer pipeline that is 20% longer (K7 = 10 stages. K8 = 12 stages).

2. The Athlon will reach 2GHz before ClawHammer arrives, and is in fact scheduled for 2.2GHz in Q4 this year.

Having talked to other people who work in the semiconductor/fab industry, 6 months is about the average time it takes to get today?s processors fully validated and up to spec (e.g. of ?up to spec?: 2GHz clock speed). Acceptable yields might be attainable during this time as well. ClawHammer being "only" at 800MHz-1GHz is nothing to sneeze at with just 30 day old silicon (which last I heard is damn impressive). AMD has a lot of time to get ClawHammer up to spec, and I'm betting they'll be able to do it for a Q4 release.

Now, take into account what Jerry Sanders about ClawHammer being 30-50% faster clock for clock than the Athlon XP. Lets say this statement is true and, worst case, ClawHammer is 30% faster clock for clock than the Athlon XP. If ClawHammer debuts at 2GHz (as the majority of hardware web sites have speculated), ClawHammer will be equal in performance to a 2.6GHz Athlon XP (30% of 2.0GHz = 600MHz (+ 2.0GHz) = 2.6GHz), which would coincidently give this theoretical Athlon XP a 3400+ rating, which is exactly what ClawHammer is scheduled to debut at later this year.

Assuming everything goes to plan for Intel and AMD for the 4th quarter this year, AMD will have a ClawHammer processor that's equivalent to a 2.6GHz AXP (3400+ rating equivalent) and Intel will have their 2.8GHz Pentium 4 (533MHz FSB). If ClawHammer is pushed to Q1, then it'll be a 3400+ processor versus a 3GHz P4 (or a 3.2GHz P4 if Intel really pushes the envelope).

It?s quite logical once you do the math.

 

[formulav8]

Where are you getting thse so-called indications that AMD is having a hard time with .13 on Thrbrd? I haven't heard anything from AMD that said they are having problems with their .13 process? Have any links or anything like that?

 

[dullard]

<< Assuming everything goes to plan for Intel and AMD for the 4th quarter this year, AMD will have a ClawHammer processor that's equivalent to a 2.6GHz AXP (3400+ rating equivalent) and Intel will have their 2.8GHz Pentium 4 (533MHz FSB). If ClawHammer is pushed to Q1, then it'll be a 3400+ processor versus a 3GHz P4 (or a 3.2GHz P4 if Intel really pushes the envelope).

It?s quite logical once you do the math. >>


Just finish out your math though. The AMD PR rating works quite well when compared to a P4 Northwood at 400 MHz fsb. Thus a 1800+XP is really really close in speed to a 1.8A GHz P4. However the 533MHz fsb Northwood seems to do quite well compared to the 400 MHz fsb Northwood. Lets give it a very conservative 10% boost. Thus the 2.8GHz Northwood with a 533MHz fsb should be given a PR rating of 2800 * 1.1 = 3100+.

Thus pretending that both AMD and Intel are right on track then the Hammer will be just under 10% faster than the P4:

3400+ / 3100+ = 1.097

In my opinion, anything under 10% is barely noticible.

 

[Rand]

<< Where are you getting thse so-called indications that AMD is having a hard time with .13 on Thrbrd? I haven't heard anything from AMD that said they are having problems with their .13 process? Have any links or anything like that? >>



The information isnt from AMD... it's simply gathered from rather clear indications from all the news reports about the issues AMD is having getting .13u on on time, and scaling in a decent manner, with appropriate thermal characteristics.
You wouldnt expect AMD themselves to admit their having issues would you?
Once upon a time .13u from AMD was scheduled to be available in Q4 of last year.



<< Having talked to other people who work in the semiconductor/fab industry, 6 months is about the average time it takes to get today?s processors fully validated and up to spec (e.g. of ?up to spec?: 2GHz clock speed). Acceptable yields might be attainable during this time as well. ClawHammer being "only" at 800MHz-1GHz is nothing to sneeze at with just 30 day old silicon (which last I heard is damn impressive). AMD has a lot of time to get ClawHammer up to spec, and I'm betting they'll be able to do it for a Q4 release. >>



I suggest you talk to those friends again, to go from initial silicon to full release in less then 1 year would be an incredible feat indeed.
Also, to quote what someone else said that is employed by an unnamed microprocessor manufacturer...
"You can't debug, validate, and attain yields at the same time. To debug and validate consists of changing the process. How can you work on improving yields when nothing is consistant?"

Whether AMD will hit release this year is debateable, they certainly seem confident they can do it.
I doubt anyone with any experience in the industry would debate the fact that AMD is setting an almost unheard of timeline to go from initial engineering sample silicon to full release though in under 12 months.

 

[Sohcan]

I'd just like to butt in here and point out flaws with extending a performance rating system to architectures and MPUs at frequencies not due for quite some time. Namely, average memory access time gets worse as clock frequency gets higher, and thus performance does not increase linearly with clock speed. It is also not possible to assume that performance will scale at the same rate with two different microarchitectures at different clock frequencies. It is thus FAR too early to even try to compare the performance of MPUs due 9-12 months from now based simply on an existing performance rating.