HAMMER BENCHMARKS 800MHz!!!!! 256K L2 Cache vs Athlon MP 800!!!!

[Gstanfor]

Well, if this turns out to be true, intel had better start pouring serious resources into imitanium (yamhill) quick smart. Clawhammer is going to absolutely murder the P4.

AMD will be perfectly capable of running the P4 as it stands clean off the roadmap even with this early silicon. Look at the performance you are getting with only 800mhz and if clawhammer is roughly as easy to manufacture as the althlon AMD should be able to scale up to 1800mhz (at least 3600+) any time they feel like it.

There is also a rumour that AMD may be capable of launching clawhammer as early as august if it so chooses (chipsets to support the CPU will be available by then).

Greg

 

[Athlon4all]

Doesn't it seem odd that a CPU with a longer pipeline smokes one with a shorter one at the same speed? A 40% improvement in Q3A?

Well for one, remember, Hammer features a vaslt improved branch predictor and thus It should offset any penalites for the longer pipeline. Also, (and especially in Quake 3 Arena) the memory latencies are being so much lowered by the integrated memory controller that it easily overtakes any performance loss suffered by the 20% increase in pipeline. I really think this is valid. Tecchannel.de has a history of being correct when benchmarking pre-released products.

Well, if this turns out to be true, intel had better start pouring serious resources into imitanium (yamhill) quick smart. Clawhammer is going to absolutely murder the P4.

AMD will be perfectly capable of running the P4 as it stands clean off the roadmap even with this early silicon. Look at the performance you are getting with only 800mhz and if clawhammer is roughly as easy to manufacture as the althlon AMD should be able to scale up to 1800mhz (at least 3600+) any time they feel like it.

I don't doubt AMD's capability to get the Hammer up to higher clock speeds, but what I do doubt is their ability to get it to high clock speeds (2.0GHz specifically) in a timely fashion. To be quite honest, AMD may end up having a very difficult decision to make coming September. They may need to decide. Do they push the Hammer launch back to early 2003 so they can get 2GHz+speeds, or do they release Hammer at lower clock speeds (I'm thinking 1.6-1.8GHz). Now they could do fine and laucnh a 2GHz Hammer in Oct. but still, I do not like what I've been hearing about the .13 micron process progress. We shall see.

There is also a rumour that AMD may be capable of launching clawhammer as early as august if it so chooses (chipsets to support the CPU will be available by then).

I highly doubt that. Plus, there is also a rumor that it will be pushed back to 2003 lol. Gotta love these rumors!!!

 

[Gstanfor]

Personally, I very much doubt there is anything wrong with AMD's 13 micron process at all. The first silicon for hammer came out ages ago and was good enough for AMD to show off. Also Sanders said the 13 micron transition was ahead of schedule and smooth around the time AMD put out their earnings statement.

It's simply a matter of priorities and hammer validation/production setup has probably been far more important to AMD than worrying about getting Thoroughbred out the door. If the 13 micron process were problematic I don't think we would be seeing the XP2500+ rumours that circulating at present.

Greg

 

[Athlon4all]

Yea I suppose so. I just to be honest still am doubtful. I will not be suprised to see AMD launch Hammer at 2Ghz. I just still lets put it this way, I'm holding my breath. I will not be suprised to see it either way, ya know? We shall see.

 

[RaynorWolfcastle]

Originally posted by: Athlon4all

AMD will be perfectly capable of running the P4 as it stands clean off the roadmap even with this early silicon. Look at the performance you are getting with only 800mhz and if clawhammer is roughly as easy to manufacture as the althlon AMD should be able to scale up to 1800mhz (at least 3600+) any time they feel like it.

I don't doubt AMD's capability to get the Hammer up to higher clock speeds, but what I do doubt is their ability to get it to high clock speeds (2.0GHz specifically) in a timely fashion. To be quite honest, AMD may end up having a very difficult decision to make coming September. They may need to decide. Do they push the Hammer launch back to early 2003 so they can get 2GHz+speeds, or do they release Hammer at lower clock speeds (I'm thinking 1.6-1.8GHz). Now they could do fine and laucnh a 2GHz Hammer in Oct. but still, I do not like what I've been hearing about the .13 micron process progress. We shall see.

There is also a rumour that AMD may be capable of launching clawhammer as early as august if it so chooses (chipsets to support the CPU will be available by then).

I highly doubt that. Plus, there is also a rumor that it will be pushed back to 2003 lol. Gotta love these rumors!!!

2GHz Paper launch anyone?

-Ice

 

[Insomniac]

Originally posted by: ragiepew

Doesn't it seem odd that a CPU with a longer pipeline smokes one with a shorter one at the same speed? A 40% improvement in Q3A?

You make a good point about legitimacy, however you cannot assume that a CPU w/ a longer pipeline will not have a higher performance per clock than one w/ a shorter one... there are other things that help performance out, especially when you consider that hammer will have an advanced improvements aside from just its traditional core. Anyway, I too am a bit skeptical, hell I think everyone is, but it doesn't make these benches any less interesting... hell, if anything, it gives us something to yap about for a while.

While I agree optimizations can be made to speed up the pipeline, enough for that kind of improvement at the same clock speed? Why haven't they optimized the Athlon pipeline since it came out? Or lengthen it and optimize so you can scale speeds and smoke Intel at the same clock rendering the "MHz Myth" pointless? And no matter how good your branch prediction unit is, it can't always be right. I don't know what a dual Athlon MP system at 800 MHz would do, but I bet they would be close. Do you think their next genearation part is that fast? Look at history.

 

[Insomniac]

Originally posted by: icecool83

2GHz Paper launch anyone?

-Ice

I'd love to see the spin if that happened!

 

[classy]

Originally posted by: Insomniac
Am I the only one who questions the legitamacy of this? Especially with the string in the bottom right of WCPUID grayed out?

Doesn't it seem odd that a CPU with a longer pipeline smokes one with a shorter one at the same speed? A 40% improvement in Q3A?

This all seems like BS to me. Why has AMD only been showing simple demos if it performs that well already? Wouldn't they show more intricate demos like Intel does?

Edit:
I'm not trying to say the Hammer won't perform, I just question what techchannel is claiming.

It was no different with the Athlon when it was in its early stages. They did the exact same thing. But when the smoke finally cleared, well we all know the story . Remember this early preveiw of the Athlon,
Athlon in the Beginning. I think this should sum up some of the specualtion around Hammer. I believe its real and its going to be here sooner than people think. As a matter of fact I can hear the shaking of the Intel "crumbling empire". Its coming................Its coming

 

[Athlon4all]

2GHz Paper launch anyone?

lol. I seriously doubt that would happen. lol

It was no different with the Athlon when it was in its early stages. They did the exact same thing. But when the smoke finally cleared, well we all know the story . Remember this early preveiw of the Athlon,
Athlon in the Beginning. I think this should sum up some of the specualtion around Hammer. I believe its real and its going to be here sooner than people think. As a matter of fact I can hear the shaking of the Intel "crumbling empire". Its coming................Its coming

Agreed. This article proves that Hammer is for real. The only caveat as I have said is can they mature their .13um SOI process, and I think that they will and thus HAmmer will launch at the speed of the highest clocked XP (likely 2.0GHz) and that 2GHz CPU, if this article is any indication (which I believe it is), this 2GHz Hammer will blow away whatever Intel has at the time and will easily compete for a long time with the Pentium 4.

 

[Rectalfier]

Originally posted by: Insomniac

Originally posted by: ragiepew

Doesn't it seem odd that a CPU with a longer pipeline smokes one with a shorter one at the same speed? A 40% improvement in Q3A?

You make a good point about legitimacy, however you cannot assume that a CPU w/ a longer pipeline will not have a higher performance per clock than one w/ a shorter one... there are other things that help performance out, especially when you consider that hammer will have an advanced improvements aside from just its traditional core. Anyway, I too am a bit skeptical, hell I think everyone is, but it doesn't make these benches any less interesting... hell, if anything, it gives us something to yap about for a while.

While I agree optimizations can be made to speed up the pipeline, enough for that kind of improvement at the same clock speed? Why haven't they optimized the Athlon pipeline since it came out? Or lengthen it and optimize so you can scale speeds and smoke Intel at the same clock rendering the "MHz Myth" pointless? And no matter how good your branch prediction unit is, it can't always be right. I don't know what a dual Athlon MP system at 800 MHz would do, but I bet they would be close. Do you think their next genearation part is that fast? Look at history.

Insomiac. These numbers seem right on to me. The P4 1600mhz has a 20 stage pipeline, but look how closely it competes with the Athlon MP at 1600mhz. Quake 3 is very memory performance\bandwidth limited. The Clawhammer's on die memory controller gives it a huge boost in performance. In benchmarks, you can see that the P4 competes with the Athlon clock for clock in memory bandwidth and SSE enabled applications. Clawhammer will remain dominant in FPU apps, as the Athlon core always had, and will surpass Intel in it's current benchmark wonder's, such as those that use or require SSE, High Memory Performance, and 512 L2 cache.

 

[Insomniac]

Originally posted by: Rectalfier

Originally posted by: Insomniac

Originally posted by: ragiepew

Doesn't it seem odd that a CPU with a longer pipeline smokes one with a shorter one at the same speed? A 40% improvement in Q3A?

You make a good point about legitimacy, however you cannot assume that a CPU w/ a longer pipeline will not have a higher performance per clock than one w/ a shorter one... there are other things that help performance out, especially when you consider that hammer will have an advanced improvements aside from just its traditional core. Anyway, I too am a bit skeptical, hell I think everyone is, but it doesn't make these benches any less interesting... hell, if anything, it gives us something to yap about for a while.

While I agree optimizations can be made to speed up the pipeline, enough for that kind of improvement at the same clock speed? Why haven't they optimized the Athlon pipeline since it came out? Or lengthen it and optimize so you can scale speeds and smoke Intel at the same clock rendering the "MHz Myth" pointless? And no matter how good your branch prediction unit is, it can't always be right. I don't know what a dual Athlon MP system at 800 MHz would do, but I bet they would be close. Do you think their next genearation part is that fast? Look at history.

Insomiac. These numbers seem right on to me. The P4 1600mhz has a 20 stage pipeline, but look how closely it competes with the Athlon MP at 1600mhz. Quake 3 is very memory performance\bandwidth limited. The Clawhammer's on die memory controller gives it a huge boost in performance. In benchmarks, you can see that the P4 competes with the Athlon clock for clock in memory bandwidth and SSE enabled applications. Clawhammer will remain dominant in FPU apps, as the Athlon core always had, and will surpass Intel in it's current benchmark wonder's, such as those that use or require SSE, High Memory Performance, and 512 L2 cache.

The P4 at 1.6 GHz does not compete with any Athlon at 1.6 GHz. It gets smoked any way you look at it. Clock for clock, an Athlon is faster than a P4. That is why AMD is using model numbers.

Do you agree that increasing the length of the pipeline decreases the IPC?

I am not saying the on board memory controller doesn't give a huge boost in performance, bit enough to overcome the lenghtened pipeline and make Q3A run 40% faster?

How about the fact that Via has been making a better memory controller for Athlons than AMD can (KT266A vs. 760)? You think AMD is going to have the best and most optimized memory controller in there right now?

Even with the most optimized pipeline and memory controller, I have BIG doubts about the ClawHammer being 40% faster in Q3A at the same clock speed.

BTW--Why did the gray out the string in the bottom right? Didn't want anyone to see that they photoshopped the info into WCPUID maybe?

 

[Rectalfier]

I agree, overall the P4 does not have a chance clock for clock. I was merely saying that in certain SSE2 and Memory bandwidth benchmarks, the P4 is able to beat the P4 clock for clock, and tha P4 has a 20 stage pipeline, while the Claw will have 12(only two over K7). Quake 3 is rumored to have SSE2 enhancements, and is known to be bandwidth intensive. I believe the Quake3 benchmarks, as quake 3 is very memory intensive, and likely has SSE2 enhancements. %40 percent is a huge increase, but does not give an accurate overall picture of Clawhammer's performance. I would say on average, Clawhammer will be 25-30% faster that K7.

 

[BDSM]

Insomniac.. The P4 actually outperforms the Ahtlon in certain well optimized apps. And the p4's pipeline is twice as long as that of the Athlon, the hammer is just 20% longer if I am not missinformed. And the hammer is a completely new design. Not incredible at all if it is better.. I think

 

[CTho9305]

Originally posted by: joohang

Originally posted by: icecool83
I don't want to rain on anyone's parade but isn't quake3 one of the most memory dependant of the commonly used benchmarks? Surprise surprise, you take a Clawhammer, give it DDR333 using the integrated memory controller and it performs great despite the low speed

Did anyone expect otherwise?

-Ice

So your point is?

Although I did wish to see more benchmarks.

I want to know if the performance WILL actually double at double the clock speed, or other things will become bottlenecks.

 

[BlvdKing]

I have a feeling that DDR SDRAM will be a bottleneck at PC2700. That is until AMD/SiS/VIA/ALi produce a dual channel DDR SDRAM chipset for the Hammer like the one Intel is producing (Granite Bay).

It will be interesting to see how much memory bandwidth affects performance now that the CPU has an onboard memory controller running at full CPU speed.

 

[BDSM]

Mr King.. If I am not gravely misstaken the Hammer features a 128 bit integrated memory controller.

128 bit DDR 333 equals 5.4 GB/sec which is more than even Dual channel 1200 mhz RDRAM can do (4.8 GB/sec)

I could be wrong though. But I've seen pics of mobo's for the clawhammer with big text showing "dual channel ddr" so I am pretty confident that I am right about this.


Ctho9305.. Ofcourse performance will not double. In order to double the performance you would have to increase all the other components in the system to match the increase in core clock frequency. however in certain apps that run almost exclusively out of the on die cache, the performance will almost double!

What we do know for sure is that even in this early stage the Hammer is performing extremely well in the tests we've seen.
And performance will surely be tuned even more before final release. Now if AMD can release the hammer at 2 GHZ+ speeds we should have a real killer on our hands. But it is ofcourse as always a matter of how long it takes before release.

 

[MadRat]

SLEDGEHammer does in fact use an integrated 128-bit memory controller, giving a theoretical 5.4GB/sec bandwidth. You do have to pair up DDR sticks to get the benefit of the 128-bit controller otherwise it cuts your bandwidth in half. There has been no confirmation that CLAWHammer will not be crippled in the way of its memory controller.

 

[socketman]

I strongly feel AMD will release at a lower speed rather than miss an October launch. I dont see the T-bred being able to hold its own against the P4 if the launch is delayed. AMD simply cant afford to miss launch. On the upside, it appears even at lower clock speeds ( say 1.8 gig) the Hammer would still be a very competitive chip.
Since the northbridge is no longer a "feature" item with Hammer chipsets, It will be interesting to see what Mobo makers do with the southbridge. WIth SATA, Bluetooth, Hypertransport, PCI-X and other features... i think our motherboards are about to get alot more interesting

 

[MadRat]

If they can get the Thoroughbred up to 2.2GHz then there is no hurry for HAMMER. Better to release the new processor as advertised then to admit its underperforming.

 

[Athlon4all]

If they can get the Thoroughbred up to 2.2GHz then there is no hurry for HAMMER. Better to release the new processor as advertised then to admit its underperforming.

Agreed, but not a T-Bred, but rather a Barton with 512k L2 Cache. A Barton 2800+ (2.2GHz) will definately compete nicely with a P4 2.8GHz and definately could compete with a 3.06Ghz P4.

Since the northbridge is no longer a "feature" item with Hammer chipsets, It will be interesting to see what Mobo makers do with the southbridge. WIth SATA, Bluetooth, Hypertransport, PCI-X and other features... i think our motherboards are about to get alot more interesting

Yah. This is also one reason why I predict that we will see nVidia gain a significant amount of market share once Hammer becomes mainstream provided they can keep the cost of the MCP-D down.Not to mention VIA could come lower because manufaturer's will for one easily be able to slap the nForce MCP with any of the North Bridges out there with the exception of VIA's K8HTA. And really, I seriously doubt that anyone will be able to provide on-board sound at the same level as nVidia's MCP, and who knows what else they'll integrated into it? It will definately have at least ATA/133, USB2 and Firewire (nForce 2 will have these for the Athlon). Something else that I think we will begin to see is a less diversification of the chipsets used by mobo makers. Right now we have multiple chipset makers on a single mobo company's mobo line (Ie Asus has SiS, VIA, and nVidia Socket A boards). I feel that we will see companies choose one chipset maker for the North Bridge and equip that with the appropriate South Bridge (I honestly feel that nVidia's features with the MCP-D combined with SiS's low cost for the Northbridge will be a deadly combo). I know some of this is speculation, but it will be interesting to see how things unfold, and I honestly feel that VIA is really setting themselves up for disaster by insisting on using VLink rather than HT.

I strongly feel AMD will release at a lower speed rather than miss an October launch. I dont see the T-bred being able to hold its own against the P4 if the launch is delayed. AMD simply cant afford to miss launch. On the upside, it appears even at lower clock speeds ( say 1.8 gig) the Hammer would still be a very competitive chip.

Agreed. I would lean towards that happening provided 1.8GHz will perform at least at the levels of the fastest Barton, prefferably faster by 10-15%.

 

[Degenerate]

NOt spoiling things here but even though this is from the inquirer, just take this in mind.

Hammer delay

 

[TwilightZone77]

^Isn't the Inquirer the one that recently reported that clawhammer could launch as early as August?

 

[ST4RCUTTER]

This is also one reason why I predict that we will see nVidia gain a significant amount of market share once Hammer becomes mainstream provided they can keep the cost of the MCP-D down.Not to mention VIA could come lower because manufaturer's will for one easily be able to slap the nForce MCP with any of the North Bridges out there with the exception of VIA's K8HTA.

I agree with the general gist of your argument, but there will probably be significant hurdles to integrated video on the K8's MCP as indicated here from Anand's day one coverage of Computex 02':

"
What's very interesting about NVIDIA's chipset is that there doesn't seem to be any on-board frame buffer for integrated video. The reason this is such an important feature that is missing is because we have been hearing all along from chipset manufacturers that the latency introduced by having to use the Hammer's on-die memory controller effectively kills integrated graphics performance. Remember that with previous chipset designs with a unified frame buffer the memory controller was physically right next to the graphics processor (or a part of it), but now the graphics must go all the way to the CPU in order to get access to the memory controller and perform any memory reads/writes. Unless the graphics cores are modified significantly with much deeper buffers to take into account this change in latency, the performance of conventional integrated graphics solutions on Hammer will suffer. NVIDIA's lack of any on-board frame buffer on their reference CK8 board indicates one of two things:"



NOt spoiling things here but even though this is from the inquirer, just take this in mind.


AMDzone has this to say about MM's claims:

According to Damon Muzny at AMD the roadmap for Hammer remains unchanged. Now if you go by strict positioning it translates to as early as October, or as late March. I'm sticking to earlier than later as my prediction. Also the .13 micron process isn't having issues according to AMD, and this has been echoed throughout their conference calls the past few months by Jerry and Hector. When AMD was having process problems with the K6 series it was revealed in the conference calls.


I suspect that the truth lies somewhere between these two bits of information. I think AMD is fairly sucessful with .13um, but is perhaps having yield problems, especially at the higher frequencies. As for SOI, it's such a new technology that that's anyone's guess. I wouldn't be surprised if their yields suck at this point with SOI on .13um. If anything was evident at Computex 02' it's that AMD and motherboard makers in general have a long way to go to optimize the performance of Hammer based systems. Moving from traditional FSB technologies to a packetized Hypertransport link is not an easy task. I also don't think AMD would introduce 1.6Ghz Clawhammers as it would be seen as missing that "magic" 2Ghz mark. Barton is already sampling though so I have no doubts it will tow the line for AMD until Hammer is cleaned up and ready to lay down the smack...

 

[christoph83]

I think this should sum up some of the specualtion around Hammer. I believe its real and its going to be here sooner than people think. As a matter of fact I can hear the shaking of the Intel "crumbling empire". Its coming................Its coming

Guess what else is coming? Prescott. We could go on and on about future chips coming out. Anyway....if you think intel's empire is going to crumble think again. Intel has enough cash on hand to buy AMD outright over 3 times. With this logic intel's empire would have crumbled when the first p4 came out and the tbird was smoking it. But guess what, Intel lost small percentages of market share but still the company never crumbled. AMD has to be able to produce large amounts of this hammer processor to take intel down. Even if intel has a crappy chip that doesnt perform well they can still supply millions of it. Until AMD can produce as many chips as intel, intel will always be the market share leader.

 

[Insomniac]

Originally posted by: BDSM
Insomniac.. The P4 actually outperforms the Ahtlon in certain well optimized apps. And the p4's pipeline is twice as long as that of the Athlon, the hammer is just 20% longer if I am not missinformed. And the hammer is a completely new design. Not incredible at all if it is better.. I think

Even at the slower clock speeds when it was first released? I thought the Athlon had it clock for clock (except maybe a benchmark that relied heavily on SSE2). My bad.