DDR SDRAM & Athlon Specs

Mr. Wong writes "We've finally got some solid benchmarks of DDR SDRAM on an AMD Athlon. Rambus better watch out, DDR looks solid." As well, check out this submission: ph4t1dck writes "Anandtech managed to get a hold of an Athlon motherboard with support for DDR memory and then proceeded to put it through its paces. Looks like this could spell big trouble for Intel & Rambus since the Athlon with DDR seems to be destroying the P3 with RDRAM. It'll be interesting to see how the P4 fares in all this with its 400 mhz front side bus."

Comments

Maybe, Maybe Not...

[Christopher B. Brown]

Technical luddite that I am, I'm getting extra RAM for a Pentium Pro 200 box that I've got so that I can run SAP R/3 on it; I don't yet have anything Athlon or P-III-based. (I'm sort of waiting for SMP Athlon motherboards to come out, although not too seriously...)

Anyhoo, P-III is likely to have more relevance than Intel should be comfortable with because:

• It's available today; P-IV isn't.
• It is compatible with today's motherboards and cases; P-IV isn't.

  That's rather important; you can't get systems based on a CPU until you can get all the necessary components.

  The fact that StrongARM motherboards aren't readily-and-cheaply available is certainly a contributing factor to a lack of widespread use for "desktop-like" applications.

  The fact that Athlon SMP motherboards aren't available (yet?) is THE CRUCIAL factor that results in there being a dearth of SMP Athlon systems.

  And once P-IV is "released," there will be a time when it is still not usefully available due to the motherboards and cases not yet having gotten through distribution channels.

• It is compatible with today's cooling technology... P-IV... may require a lot of your fans...

amd vs. intel - oem market main sector

[popoutman01]

When dell finally swallow the losses of the cheaper processors and the halved advertising expenditure, and make amd pc's, the market will really swing over to amd, especially as the future intel chips get whipped by amd and dell cannot afford to risk the loss of the market share.

rambus rebates?

[photozz]

"Looks like this could spell big trouble for Intel & Rambus"
Wasn't there a story recently, either here or ZDnet, where Intell is now offering manufacturers rebates for their rambus stuff??

Waiting....Waiting....Waiting....

[LordLobo]

The end of the waiting is near. I've lusted for this combo for more than a year: Dual Athlon + PC2100 DDR Ram We're yet to see the full potenial of the EV6 bus. Sadly it looks like my system of 2000 will probably have to wait for 2001. Sigh.

Re:Waiting....Waiting....Waiting....

[Sloppy]

You still have it pretty darn easy compared to those of us who wanted POP boxes.

---

Re:P-III vs Athlon = Pointless

[Black Parrot]

> The P-III is at the end of it's life...

Yeah, me and all my friends run P.IVs, and they stomp the hell out of..., erm, um, huh? Oh, sorry, my alarm just went off and I have to wake up now.

--

Re:Almost there...

[sensate_mass]

If I had a nickel for every time someone misspelled 'nickel' while correcting the spelling of 'announcement,' I'd have a nickel.

Linux benchmarks?

[NetJunkie]

What are some good well-rounded Linux benchmarks?

i820 kinda sucked on those benchmarks

[crisco]

I didn't see a huge performance differece between the DDR and the PC133 Ram and using the DDR didn't put a 1 GHZ Athlon too far ahead of a PIII using PC133. But the surprise was how badly a 820 based board using RDRAM fared, even on Intel's benchmarks.

Also of interest was the reports of decent stability with the pre-production motherboard. P. Mmm, gotta love GHZ

Re:err...

[tak amalak]

They sure do only have 8KB of L1 data cache. Check it out: http://www.realworl dtech.com/page.cfm?ArticleID=RWT091000000000
--

the screaming cindy advantage

[_|()|\|]

So a comparision where AMD wins is more 'normal'?

Quake III is heavily optimized for SSE, but not for 3DNow. Unless the software you use is similarly optimized, the Q3 benchmarks are a little skewed in Intel's favor.

Hardware reviews rely too heavily on the Quake benchmarks. AMD cleverly helped id and 3dfx optimize Quake II for 3DNow. A K6-2 and a Voodoo 2 with all the right drivers and patches made for a great Quake II system. The average DirectX game, though, revealed the weakness of the K6-2 FPU. Now the tables are turned and Intel uses SSE to make up for a FPU disadvantage in Quake III.

Personally, I'll take a solid FPU over SIMD extensions any day.

Re:the screaming cindy advantage

[Score Whore]

I wouldn't disagree that a solid general purpose FPU is a better component than a specialized parallel unit.

I checked with the man (JC) about SIMD optimizations in Q3. To paraphrase him, there isn't any in the id codebase (having to write it in asm, lacking portability, uglifying the code, and not enough speedup were his reasons.) But it's likely to be in the various OpenGL drivers.

I agree that Q3 is not the be-all end-all of benchmarks. It showcases a on set of CPU strengths/weaknesses. UT is similiar in that it will exercise different aspects of the CPU and it doesn't make any sense for a benchmark-er to say one is better (or in this case more "normal") than the other. They're just different and either one is a good indicator of general system performance. And to say that one CPU/mobo combination is better based on selecting benchmarks to show that is ridiculous. It's no different than Apple claiming that the 266 Mhz G3 in their iMacs is twice as fast a processor as the 500 Mhz P3 because of a very damaged benchmark suite (ByteMARKS.)

Unlil we start seeing differences > 10% as a general rule in a wide range of benchmarks it would be pretty much irrelevant to a typical user to go with any 1Ghz machine (if they want that kind of speed.)

What would be interesting to see is a null-OpenGL driver for Q3. Thus we could leave the graphics card/driver/agp/pci issues out of things and just see how fast this particular task can be solved by a particular CPU. And to tell you the truth if we're looking at tasks that can be optimized by use of specialized instructions then they should be used. Otherwise we aren't testing the boundaries of the chips involved. IMO.

Like I said before, I'd rather have 3DNow! than MMX anyday. But even before either of those I'd take another 500 Mhz of CPU clock and 256 MB RAM.

Re:the screaming cindy advantage

[Score Whore]

Uh oh, I'm replying to my own posts!!!!

Had another thought. Both UT and Quake are going to be hit by texture transfers. But UT will leverage the paletted textures of the Voodoo card which Quake doesn't. So if anything the benchmarking system is biased towards UT and requiring less memory bandwidth. Whereas Q3 will need a larger amount of memory bandwidth to move 32bit textures out to the card. What Anand has shown is that the DDR system isn't as good a performer as the RDRAM system. Bah.

These benchmarks are incomplete anyway.

[The Master Magician]

If you notice while they are able to run the processor using the 266mhz RAM speed, the final platform also provides for 266mhz bus speeds for the CPU (133x2 for EV6 bus = 266). They did not have a processor that would run at that bus speed, so this is not a test of the platform itself, it's more of a test of the difference between the Athlon running with PC133 and the Athlon running with DDR RAM.

This is really a very useless and misleading test. They should really have waited until they could utilize all the functionality of the new platform rather than showing us benchmarks using only half the added functionality.

Some issues with that thar P-4.

[InThane]

First off, as some people have pointed out, the P4 is running slower than the P3, clock for clock. In addition, by the time the P4 is due out, at current ramp-up for the Athlon, we should be seeing eqivalent-speed Athlons running on DDR 133mhz system busses, or roughly eqivalent to 266mhz. In addition, we're looking at using DDR SDRAM on the 266mhz system bus, which should give it a nice little kick in the pants versus RAMBUS, which probably won't even be able to fill half of the 400mhz system bus of the P4.

Now, Intel might still have some tricks up their sleeves, but IMO they've just plain dropped the ball. The Athlon came out of nowhere, kicked them in the pants, and they just weren't ready for it. They were investing major money in the Merced chip program, trying to milk some more money out of the P6 core, and so little RnD went towards upgrading their Pentium line.

Me? I'm bullish on AMD, bearish on Intel. I doubt that Intel is going to go out of business, but I can see a reversal of position and popularity in the cards if AMD doesn't fumble and Intel keeps on their current chosen path.

Re:Some issues with that thar P-4.

[Paul Komarek]

"Now, Intel might still have some tricks up their sleeves, but IMO they've just plain dropped the ball. The Athlon came out of nowhere, kicked them in the pants..."

To pick nits, the Athlon didn't 'come out of nowhere', especially insofar as Intel is concerend. Intel knew full well what AMD was doing--you just can't miss that NexGen and Alpha architects are being hired by AMD. Intel completely blew it. They had been patching up the x86 processor for years, while other companies failed to execute properly. When the Athlon arrive, Intel was mired in its own internal disasters and AMD finally got their act together. Only after the Athlon 1GHz was announced did we start to really witness Intel coming apart at the seams.

It might be worth noting that the only thing x86 had going for it after about 1992 was backward compatibility. Intel knows a lot about manufacturing chips, but has never really been successful in the cpu-research department--which was fine so long as they were making the best x86 implementation available. When Intel did become more adventurous, they failed. Even with HP's help (which they didn't really like to acknowledge) on the Merced/Itanium, Intel still couldn't hack it. This platform may never make it to the market, though we might see its successor. But even once Intel's VLIW stuff does reach the market, they won't have backward compatibility, the only valuable part of their existing processor market.

Compare this to AMD's 64-bit strategy, or to Alpha's 64-bit legacy. If you want 64-bit x86-like code, go with AMD Sledgehammer. If you want super-duper 64-bit code and don't need backward compatibility, go with a damn fast, proven architecture like Alpha. And there are other 64bit architectures out there that are competitive with the Alpha, for instance the IBM RS-64 III.

VLIW as a technology may stand a chance in the future, but I'll venture that Intel doesn't have what it takes to make it work in any reasonable amount of time. In fact, my opinion is that Intel's x86 architecture is retarding the commercial development of new cpu architectures -- even within Intel.

I predict AMD will eventually forget about retail microprocessors once x86 is no longer desired in the market. Intel will stop torturing us with their designs and become the world's largest and best fabrication company. Microsoft will stop writing software and become the world's largest legal firm (do they already hold this distinction?). This is how we'll finally see the end of the x86 and Win32 architectural disasters that are anchoring computing technology in the 1980s.

-Paul Komarek

Re:RAMBUS is a memory bus bottleneck

[MarNuke]

Pentium 4 system bus is 64 bit 100 MHz QDR, 100*4*64 = 25600 kbit/s

That's still slow as heck.

Regular stories on Techweb

[Greyfox]

I've been seeing about a story a week on Techweb. Latest one has Micron and Hyundai (Yes, Hyundai!) filing suit trying to get the Rambus patents declared invalid. I gave up on submitting them to /. since they're always rejected, but Techweb has enough interesting stuff on it that it should be one of the sites you visit on a regular basis anyway.

Re:Here's why ....

[taniwha]

I'm sure it does, but I doubt it fetches partial cache lines so once it starts a fetch the bus and (more importantly, because before it was slower) the memory subsystem is busy untill a full cache line has been fetched.

Also if the bus doesn't have some form of flow control then the memory controller will have to wait until it has a full line almost ready to go before it can start sending data - this will add a lot to latency - but now with a bus/memory combination that run at the same rate these sorts of latencies will also go away.

A prediction - as 266MHz (2x133) DDRAMS start to appear the k7 bus will be pumped up to 133MHz (266 on both edges) and k7 speed grades will start to appear in 66MHz increments rather than the current 50MHz ones.

Re:Direct Links to Benchmark Pages

[ackthpt]

Not working. Perhaps they got a C&D from Intel to stop making someone's product look better than the upcoming P4. ;-)

Vote Naked 2000

Re:Direct Links to Benchmark Pages

[PraveenS]

Anandtech is running NT. Serves 'em right.

Re:the Slashdot effect?

[Psiren]

Yeah, but if you look down the list of companies they thank, one of them is absent... ;-)

Re:Come a long way

[cdipierr]

Geez, I love how moderators mod things up even though posts aren't based in reality.

3d-Now! is not merely an MMX extension. It was the first SIMD extension for floating point in the x86 architecture (Intel's SSE came later). It operates independently of MMX.

As for not being a substitute for your graphics card...well duh. But keep in mind that your CPU isn't doing nothing when pushing Quak3 out to the monitor. There's still the whole texture and lighting stage that must be done on the CPU (yes, Nvidia and all can get around this, but not all software uses those extensions).

As for "Is 3dNow used?", the answer is YES! Most modern video drivers actually incorproate 3dNow instructions in them as does some software. However, a lot of software prefers SSE since it's backed by the big guy. For reference though, the Athlon supports SSE prefetching just like the PIII.

RAMBUS is a memory bus bottleneck

[MarNuke]

It'll be interesting to see how the P4 fares in all this with its 400 mhz front side bus

However unlikly, if this bus is the same 400mhz, 8 bit bus of rambus, it's going to be slow.

Rambus uses a 8 bit bus where DDR SDRAM uses 32 bits at 200 mhz. That's 8bits*400mhz=3,200 and 32bits*200mhz=6,400.

How about an alpha? It uses a 64 bit bus at 200mhz. 64bits*200mhz=12,800.

How fast will the P4 fare in all of this? Well, once the info get there, it's going to be fast, but it might take a while, oh, and don't try to push it.

Funny how everything has moved forward, but Rambus and Intel is trying to drop a 8 bit on our heads.

Sorry, I won't buy a P4, just like I haven't bought anything after the p233 from them.

Re:RAMBUS is a memory bus bottleneck

[Anonymous Coward]

You are misinformed ...

DRDRAM (rambus) memory bus is 16 bit 400 MHz DDR, 400*2*16 = 12800 kbit/s

DDR SDRAM is 64 bit 133 MHz DDR, 133*2*64 = 17024 kbit/s

Athlon system bus is 64 bit 100 MHz DDR, 100*2*64 = 12800 kbit/s (and soon 133 MHz DDR)

Pentium 4 system bus is 64 bit 100 MHz QDR, 100*4*64 = 25600 kbit/s

Re:Corporate PC Purchases Rule

[MrBogus]

But Guess What?!? All the OEMs see AMD and think, "Hmmm...what kind of cheap 5hit components can I throw together and make a profit selling?"

Yes, it's very strange that none of the major vendors are targeting AMD-based systems at the lucrative corporate market, instead putting nice fast chips in low-end Office Depot small biz machines, and bells-and-whistles home machines like CPQ Presarios.

The only reasons for this I can think of are:

1) Corporations want motherboard components to change very slowly so they can use the same disk image for longer than a couple months.
2) AMD doesn't have the capacity to fill a desktop line which might last a year or more.
3) The vendors know that AMD motherboards are not as reliable as Intel ones, and are afraid of having to do a major recall if their more saavy customers found out.

(A few years ago, I saw a study which said that the average larger corporation spends something like $20,000/year/desktop to support a PC user. So, you're right, the marginal hardware cost difference between AMD and Intel is irrelelevant.)

Re:ha

[ackthpt]

I've read some interesting english on Tom's website and I think they run it through babel fish, as no sane human translator would make some of those gaffs.

Who knows, maybe they have trained a carp. ;-)

Vote Naked 2000

PC1600 used in the benches

[Keeper]

Pretty impressive numbers considering that the benchmarks used an 'alpha' motherboard (rev 1) and the slowest possible DDR SDRAM available. AND it was being compared to the fastest rambus ram available and many many benchmarks which favor the PIII. Not only that, but the game benches were done using a card over the PCI bus (no AGP support was available yet).

Should be interesting to see the kind of bandwidth improvements provided by PC2100 DDR SDRAM (~133mhz clock).

It should also be interesting to see when support for PC2600 DDR SDRAM (~166mhz clock) becomes available and what kind of additional performance it will provide. I'm not entirely sure that support for this will be provided in the 760 chipset.

Why should it make a difference

[meadowsp]

Why should it make any difference? If something does better on the Winstone test then it will also run Linux quicker.

Who cares about Linux tests? I want MSDOS tests.

Re:Direct Links to Benchmark Pages

[Chris Pimlott]

There's no reason to bitch. Unlike most sites, AnandTech lets you see the whole thing on one page - the trick is to click on the "Print this article" link. It also gets rid of the sidebars and such.

For this article, the 1 page link is http://www.anandtech.com/printart icle.html?i=1319.

Athlon+DDR vs Motorola G4?

[swb]

Any comparisons there? Are there benchmarks that can compare these two CPUs?

oh no! (on the other side of the moon off topic)

[twitter]

not another grammar nazi!

Re:Better technology, but...

[Dr. Merkw�rdigliebe]

I think Infineon (Siemens spin-off) and Hyundai can be considered to be in the "Rambus can take their patents and put them you-know-where" camp as well

bend over, its your turn

[supermonkey]

goot, i hope those dumbfucks (intel, rambus) take it in the ass

yip yip zwoopity do!

Re:Almost there...

[White Shadow]

If big business endorses it, it will succeed.

While this may be true, whether big business endorses it or not is highly dependent on what people want. That is, if enough people are told about AMD and how it's an alternative to Intel, and want to see it in systems, then big businesses will endorse it to stay competitive. While big businesses do have lots of control, don't underestimate the power of a large group of consumers. A good read would be this article about FreeBSD Advocacy. Same idea, different product.

Almost there...

[BrK]

If I had a nickle for every time some new product announcment "spelled trouble" for Intel, I'd be retired now. I certainly hope that as new things emerge Intel's death grip on the PeeCee market will erode, but we can only hope. Sometimes I think that the g33k culture gets too excited over things like this, not realizing that Joe Consumer generally doesn't know squat about what's in his box, and is more concerned with what software is pre-loaded, or what kind of speakers come with his new PeeCee. Not to mention that sales to businesses dwarf end-user sales (and the DIY PeeCee builders are less than noise in the equation), we have to ask "Does that make $en$e to the corporate buyers?". If big business endorses it, it will succeed. If Dell/Compaq/Micron/etc don't package this hardware properly, then the Intel/Rambus camp may continue to succeed in spite of its inferior performance.

Re:Almost there...

[Adolf Grammar]

Ahh, my feeble-minded friend. How you have proved your ignorance. Let me inform you that "nickle" is in fact an appropriate variant spelling of "nickel". Either is appropriate.

I suggest in the future that you look the word up in a dictionary, to prevent any further embarassment.

Glad to have cleared things up for you,

Adolf

Re:Almost there...

[nitehawk214]

> If Dell/Compaq/Micron/etc don't package this hardware properly, then the Intel/Rambus camp may continue to succeed in spite of its inferior performance.

You have your grouping wrong, Dell belongs with Intel/Rambus not Compaq/Micron. Dell has not (and does not plan to) ship any AMD based systems. Which is really a shame becaue Dell has made some great systems but will not acknowledge that AMD is here to stay.

Re:Almost there...

[sensate_mass]

This ignorant boob could not find a match for 'nickle' in either the American Heritage Dictionary or the Cambridge International Dictionary of English, although the 1913 edition of Webster's Dictionary did define your "appropriate variant spelling" as Nic"kle (?), n. (Zoöl.) The European woodpecker, or yaffle; -- called also nicker pecker.

I, however, was referring to the American five-cent piece, named for one of the metals it contains, which is spelled 'n-i-c-k-e-l.' Perhaps I missed something. Perhaps pompous people are particular in posts pertaining to their peckers. Many apologies.

Re:Almost there...

[Adolf Grammar]

Please check http://www.webster.com

I believe that is where you will find the information.

Main Entry: nickel
Variant(s): also nickle /'ni-k&l/
Function: noun
Etymology: probably from Swedish nickel, from German Kupfernickel niccolite, probably from Kupfer copper + Nickel goblin; from the deceptive copper color of niccolite
Date: 1755
1 : a silver-white hard malleable ductile metallic element capable of a high polish and resistant to corrosion that is used chiefly in alloys and as a catalyst -- see ELEMENT table
2 a (1) : the U.S. 5-cent piece regularly containing 25 percent nickel and 75 percent copper (2) : the Canadian 5-cent piece b : five cents
3 : a pass defense in football that employs five defensive backs

OWNED

Thanks for playing,

Adolf

Come a long way

[(trb001)]

Man, I can remember when everyone looked at me kinda funny for putting a K6-2/350 chip in my computer. AMD wasn't doing much, their processors weren't as fast as Intel's, blah, blah blah. I think I need to make a few phone calls :P

Question...whatever happened to 3dNow? Did it just never catch on or what? I remember thinking it could have been a great technology, on processor 3d modeling stuff. Sounded like something gaming companies would have jumped on. Is it even included in the new Athlons?

--trb

Re:Come a long way

[Shadow99_1]

Uh it isn't "improved MMX" either. 3dnow was the first real SIMD functions to show that they were worthwhile. Intel looked at 3dnow when they made 'SSE'. AMD extended 3dnow in the athlons & k6+ chips.

AMD's problem with gettign 3dnow adopted has always been that no compiler automatically adds 3dnow where it helps, where as Intel has several compilers that autoconfigure SSE optimizations. & porgrammers being lazy bastards (not all of you, though any real lazy bastards will flame me after this) don't hand code optimizations, so Intel 'won' this war of SIMD instructions on 'ease of optimization'... That's why no one sees 3dnow much anymore...

Re:Come a long way

[-brazil-]

Um... 3dNow is not "on processor 3d modeling stuff". It's merely AMD's slightly improved flavor of MMX, basically a bit of SIMD computing. Of course, 3D software can benefit from it, but it's far away from being a substitute for the dedicated rendering hardware found on the newest graphics cards.

Better technology, but...

[Hanzie]

The Tech is better, but it seems that our court system is going to put Rambus on the top, considering all the lawsuits that the Rambus consortium is winning.

According to Ars Technica, only Micron remains.

This means that we're going to pay extra for our RAM habits. I believe that the CEO of Intel earlier stated that Rambus forcing out DRAM was a scenario that Intel was intentionally pushing for.

Re:Better technology, but...

[Anonymous Coward]

Rambus hasn't won any real lawsuits yet. All the companies that have 'caved' have done so under very limited terms (and the understanding that it was a lot cheaper than fighting it out in court).
The latest settlement, for example, is only in effect until the start of 2001. It'll cost them a lot less to pay royalties for 4 months than to fight this in court. Earlier settlements were under similar limitations or made by companies with strong times to Rambus. They're not really significant, except in a PR kind of way.

Here's why ....

[taniwha]

Athlon has a 200MHz system bus (really 100 but they move data on both edges of the clock). On the other hand the PC100 memory systems only make ata at 1/2 this rate - this means that great bus is starved for performance and stuff must wait - even with PC133 it has to wait for the memory - to make matters worse K7 has cache lines that are twice as long - it has to wait twice as long as an Intel part - now with a 2x memory system and bus the latency for k7 is the same as that for a Pentium (latency is king for reads) and twice as much data is transfered to boot

In other words K7's been crippled by its memory subsystem from day 1 -it's only now you will start to see it running at full speed ....

Re:Here's why ....

[Ralph Wiggam]

That would be "The Athlon has a 200MHz system bus."

-B

Re:Here's why ....

[Black Parrot]

> to make matters worse K7 has cache lines that are twice as long - it has to wait twice as long as an Intel part

Does the Athlon fetch the line elements in order, and wait until they all arive before allowing the processor to proceed? I know that some computers use a more sophisticated algorithm for filling cache lines, and thereby minimize the latency that would otherwise arise from the effect you describe. However, I do no what the Athlon does. So I just wanted to ask: are you making an assumption, or do you know for a fact that the Athlon acts that way?

--

Re:Here's why ....

[Paul Komarek]

I'm not sure about the Athlon in particular, but I know that the Alpha '264 can keep truckin' with several outstanding cache line fills. This may be relevant because of the architectural connections (people and design) between the Athlon and the Alpha.

-Paul Komarek

Rambus better watch out?

[mholve]

When was it ever a "threat?" ;>

P-III vs Athlon = Pointless

[um... Lucas]

The P-III is at the end of it's life... it seems rather pointless to compare it's performance versus the Athlon any longer. In most cases, for this generation, the Athlon either provided superior performance, or at least equal performance, while costing a fraction of what an Intel chip would have costed...

With the P-IV on it's way and having a TON of architectural changes to itself and the rest of the subsystems, any comparison of the P-III vs Athlon should be disregarded come the Pentium 4.

I'm not AMD bashing - i've got an Athlon 700 system enroute to me as I type this, but P-3 vs Athlon seems to be beating a dead horse at this point in the game.

Re:P-III vs Athlon = Pointless

[Adolf Grammar]

The P-III is at the end of it's life...

My silly little friend -- this should be "The P-III is at the end of its life..."

Hope that helps,

Adolf

Re:P-III vs Athlon = Pointless

[TheCaptain]

Well...if your shopping and deciding between an Athlon or a PIII, the P-IV has very little to do with the equation. The depending on how they are priced, alot of folks might want to use a PIII for a bit yet...and until I have a P-IV in my hands (or even CAN have one for that matter) it's still worth looking at.

Re:P-III vs Athlon = Pointless

[Danh]

the geek appeal of supporting the underdog... Face it, we all love the lame dogs...

We support the underdogs because we know the value of a sound competition. Market domination by a single corporation leads to less quality, higher prices, less choice... So supporting the underdog gives a benefit to the whole comunity.
And underdog does not mean lame dogs: there are plenty of examples where they outperform the market holder (AMD vs Intel, Linux vs Windoze, Opera vs Netscape and IE) although it is ways easier for the monopolist to produce their products. In these cases I see a double profit, a collective and an private one, since underdogs are generally forced to give us better price/service ratios.

Re:P-III vs Athlon = Pointless

[deefer]

Yeah, but IIRC, P4's need new cases and cooling fans to work... I can just rip out my mobo, and slide a lovely Athlon 1GHz on an Asus mobo into it's place... Nice.

AMD originally made their name within the nerderati because they were cheap and vastly overclockable. It became another "geek" badge... I wonder if, now that AMD are getting more mainstream (and deservedly so, IMHO), the takeup by hardcore nerds is dropping off - poss9ibly in favour of Alpha's? And here's a story tip for Katz - the geek appeal of supporting the underdog... Face it, we all love the lame dogs...

Strong data typing is for those with weak minds.

Re:P-III vs Athlon = Pointless

[platos_beard]

"a TON or architectural changes" ???

I tought it was the HEAT SINK that was a TON.

Re:P-III vs Athlon = Pointless

[deKernel]

Normally I would agree with you, but the PIII is going to be around for quite some time. In addition, I don't think you will be seeing the PIV around for quite some time regardless of what Intel says (something about not being able to actually make a product in use quantity!!).

So my 2sense is that this is quite important because anybody upgrading their system should take note of this if performance is an issue.

Re:P-III vs Athlon = Pointless

[um... Lucas]

New instructions (we'll see if they get any use)
400 MHz system bus (twice as fast as the Athlon)
Rambus memory
20 stage pipeline vs 10...

There are just too many changes being made to be able take a P-III's performance versus a P4's.

And yeah, and it sure does ship with a whopper of a heatsink! :)

Re:P-III vs Athlon = Pointless

[Steeltoe]

But comparisons between older systems ARE beneficial. I believe anyone buying the latest K-7, P-4 and 1.whatnot Ghz CPUs get what they deserve.

The "bleeding edge" has a name for a good reason :-)

- Steeltoe

Re:the Slashdot effect?

[John_Booty]

It appears that way!!! Geez, you'd think that a site like Anand's would be able to handle it. MY guess is that word of his DDR tests are spreading very quickly across the net, since we've all been waiting for a good set of DDR benchmarks.

And why would this be so?

[ackthpt]

Looks like this could spell big trouble for Intel & Rambus

I read yesterday that NEC is now caving to Rambus on IP. I haven't read anywhere that Intel isn't doing a DDR chipset. (840 IIRC) While I'm glad to see benches at last, I would dismiss neither Intel or Rambus, as both seem to be in a good position to take advantage of DDR. Please note as well that DDR motherboards and RAM won't be widely available for a couple months, yet. Plenty of time for Intel (who does sometimes hold the cards closer to their chest) to make hay, too.

Now if Infineon and Micron could just dent Rambus' IP claim we might have something to really celebrate.

Vote Naked 2000

Re:And why would this be so?

[JanKotz]

The i840 is a dual channel Rambus setup -- it's marginally the fastest PC chipset in the world. You are partially right, though -- IIRC, Intel's deal with Rambus *does* allow them to make a server/high end workstation chipset using DDR, but they are apparently not allowed to use anything but Rambus in their midrange chipset products for another year.

Rambus is going to prove that its power with the P4, but it will be too little value, demonstrated way too late. It will disappear quickly, just like MCA and VESA Local Bus.

I think the biggest lesson learned from this yearlong memor issue is this: don't mess with price/performance ratios.
--

East Germany?

[MarkusH]

But I thought East Germany no longer existed. Yet they are making RAM Chips?

Re:East Germany?

[Anonymous Coward]

But I thought East Germany no longer existed. Yet they are making RAM Chips?

Yes, you are correct, they DID get rid of East Germany. Germany now has no Eastern region, as during the German re-unification a relativistic spatial contraction was performed upon the German landmass, thus compressing Germany into a two-dimensional region. Under EU law, this sort of national compression is legal, and is totally unobservable by the residents of Germany.

Unfortunately, the huge area of Europe which was formerly occupied by the three-dimensional Germany is now a dangerous void. Anyone who falls into the aforementioned void will instantly be sucked into a parallel world of frightening paradoxes such as "What if I killed my own Grandfather?" and "If God made the world, then who made God?".

People who happen to fall into this void are forced into an infinite lifetime of hard labour, where they are forced to make RAM chips from ear wax and their own anal hair. Once a week, a huge space cockroach called Tarquin descends into the void to harvest the RAM chips. The RAM chips are then distributed (by supernatural means) to shops around the world where customers may purchase them. This is how East Germany manufactures RAM.

Thank you.

Direct Links to Benchmark Pages

[John_Booty]

Anand's site seems to be Slashdotted or something. Sometimes the pages come up and sometimes they don't.

Since pulling up any given page on Anand's site right now is sort of a crapshoot, and the benchmarks start on PAGE 8 or so of the review, you might want to jump straight to the benchmarks... here's the link...

http://www.anandtech.com/showdoc.h tml?i=1319&p=8

Just change the "p=8" to "p=9" or something to jump to page 9 instead of 8... gotta love those excessively-paginated articles to boost those ad hit rates.... :-D

Re:Direct Links to Benchmark Pages

[Megahurts]

> Anandtech is running NT. Serves 'em right.

I'd say it doesn't seem to be serving them anything.

---

Re:Direct Links to Benchmark Pages

[Steveftoth]

This site is always BUSY, you'd think that with all the hits they're getting that they'd be able to upgrade to a faster system. Everytime a post gets on slashdot about his cool review (they are cool), I can't see them cause the site is too busY! This is why they should have a broadcasting protocol that works. Or maybe he should just stop running Nt, we'll see.

"at it's introduction"...

[Wakko Warner]

...whenever it occurs, I'm sure the Athlon will be running at comparable speeds.

Maybe, in a few years, when there's actually enough Pentium 4 chips around that it matters, we won't even be using x86 anymore.

- A.P.
--

"One World, one Web, one Program" - Microsoft promotional ad

Re:Corporate PC Purchases Rule

[ackthpt]

Similar to the Blue affect of IBM. Although, looking back over the past 10 years, I'd say IBM has learned that they actually have to be smart and competitive, telling people "Just buy from us, we know what's good for you" doesn't work anymore. With Intel's stumbles of late, the sheen of invulnerablility is showing some tarnish. AMD needs to work on pushing their track record in the faces of those corporate buyers. Same applies to the SDRAM, RDRAM and DDR SDRAM front. If purchasing agents were made aware that all systems up to now have used SDRAM they'll cast a jaundiced eye at anything else, unless Intel is pushing it and they trust the Intel name and aegis.


Vote Naked 2000

Not Pointless

[twitter]

They compared what they could get, and what most of us will be able to afford for a while.

I can tell you the result of the test you want right now. The Pentium four, err, P IV, 5-4, 9 to 5, vapor now in this box is not here. The AMD system in this box is as reported in the last test.

Someone is sure to give you what you demand as soon as that vapor materializes.

Intel's death == good?

[Fervent]

Does anyone fear that Intel's slide over the past year is a little less than good? Granted, I love competition (one of the reasons I bought a Dreamcast, and plan to buy a PS2) but people ringing the death bells get a little more than irksome.

What if we were to say "the Athlon performs good, and the Pentium 4 performs good" instead of "the Athlon kicks Intel's ass"? (Granted, I'm hypocritical: I use an Athlon for my Win2000/Linux box. But enough is enough).

Hmm.. Anand an AMD supporter?

[Score Whore]

After reading the review and seeing these two paragraphs:

Quake III Arena paints us an interesting picture. While the AMD 760 with its PC1600 DDR SDRAM is able to offer a 14% improvement over the KT133/PC133 platform, it is still defeated by the i815/PC133 platform by a couple of fps. In order to understand this you have to understand the nature of Quake III Arena as a benchmark. A final board/chipset would most likely be able to pull ahead of the i815/PC133 platform however, because of the SSE optimizations in Quake III Arena it's very likely that a Pentium III on a solid platform could give even an Athlon with DDR SDRAM a run for its money.

UnrealTournament provides us with a more 'normal' comparison, as the AMD 760 platform pulls ahead yet again. This time 13% faster than the KT133 based Athlon platform, the AMD 760/DDR setup is only about 3% faster than the i815/PC133 platform however we can expect a final release to be much more competitive.

So a comparision where AMD wins is more 'normal'? And how about those excuses for why the P3 is beating the Athlon? These sorts of tests should really be done without any bias. I mean I'm all for whacking the yutzes on the head for screwing around with RDRAM, but let's get real here... if a platform is faster for a particular benchmark it's faster and there really isn't any point in rationalizing away the fact. Who cares if it uses optimized instructions for a particular intel chip. For hell sakes, I'd rather have 3dnow on my celeron instead of MMX, but that doesn't mean that I will ignore the fact that software which leverages 3dnow is faster than the car wreck that is MMX.

that was funny

[twitter]

One too many gargle blasters for you, eh?

That was the funniest thing I've read all week, why did you AC it?

Re:Corporate PC Purchases Rule

[Shadow99_1]

Well some OEM's have aimed AMD systems at corporate markets. Though only Sys coems to mind right now... Their are at least 3 (though none of the top 10) that do market corporate athlon systems...

no screaming cindy?

[_|()|\|]

there isn't any [SIMD optimizations] the id codebase

The AnandTech article implied that there were. I could have sworn I'd read elsewhere that there were, but maybe it was another AnandTech article. I did find a cameo interview with Carmack at FiringSquad that paraphrases to what you said:

Most of the Katmai optimizations [for Quake 3] are in the OpenGL drivers. We may have some loops in the main code Katmai optimized, but it is a low priority. Because up to 75% of the execution time of the game is in the graphics driver, most of the burden of optimization is theirs.

By the way, you keep comparing 3DNow to MMX. Are you including SSE in MMX?

Re:no screaming cindy?

[Score Whore]

The AnandTech article implied that there were.

I got that from the Anand article too. But I hadn't heard any kind of announcement from id, and they are typically pretty open about things like this. So I mailed them and asked.

By the way, you keep comparing 3DNow to MMX. Are you including SSE in MMX?

Actually no I'm not. I'm comparing my Celeron with what is available from AMD. I've got MMX, which is useful in a very limited number of applications that I typically use. 3DNow! would be a lot more useful to me, but alas when I bought my machine you got a lot more bang for your buck if you OC-ed a 300A to 450...

Re:When do GHz chips get cheap (== under $500)?

[Some Dumbass...]

What the hell are you talking about?

The Ghz Athlon sells for around $500 now. See, for example, www.astak.com, a reliable vendor which is presently selling the 1Ghz Athlon Classic and Thunderbird (both) for $475. Both are OEM of course, but still...

Yep. They're also all foreign companies

[Tridus]

All of the settlements I have heard of so far were foreign companies who have a track record of not really fighting american companies in american courts anyway.

That changes with Micron, who have far more of a level playing field.

They're the ones we should be watching. Micron has the best chance of winning. If they do, that will quickly turn things around and these settlements will stop happening. If they settle, well, then everybody who is left is likely to.

clue phone, it's for you...

[Wakko Warner]

Rambus RDRAM is manufactured in asia by companies like Toshiba and Samsung, just like every other kind of RAM. None of it is made in the US.

Jeez.

- A.P.
--

"One World, one Web, one Program" - Microsoft promotional ad

Re:the Slashdot effect?

[rayvd]

Hmmmm... looks like Anand has some decent hardware--maybe it's the OS? :-)

Re:err...

[Hammer]

Intels own demo pitched a 1.5 GHz P-4 vs a .75 GHz P-!!!
Draw your own conclusions ;->

Re:err...

[tak amalak]

Remember that the P4 only has 8K of L1 data cache. It's 4 way set associative versus the 2 way of the Athlon but thats got to hurt it in realworld.
--

Re:Athlon SMP is delayed to next year

[-|Oblom|-]

It's supposed to be possible to link a few 760m chipstes together with LDT bus. So by linking 2 760m chips you will have 4 way system. By linking 4 chips you will have 8 way system, and so on...

Re:When do GHz chips get cheap (== under $500)?

[Steveftoth]

The 1.12Ghz chip from intel was a disaster. It showed to the public that ghz speed are not ready for prime time (with Intel chips). The only x86 ghz chip is the Athlon (that works). Personally, I don't see the need for a ghz chip yet. Even java swing runs great on my 650mhz.

Hmmm...

[CodeRx]

Anandtech seems to having problems handling the load.

Requested File is not currently available in cluster 'anandcluster'. Please try again later.

Netcraft sez:

www.anandtech.com is running Microsoft-IIS/5.0 on Windows 2000

OBOnTopic: It's nice to see DDR SDRAM starting to make it's way into the market. The sooner rambus goes away, the better.

Included in DirectX

[Big Nothing]

If i'm not gravely mistaken, the 3DNow instruction set is implemented in DirectExx, so basically any game run under Windows (using Direct X) is exployting this feature (if you have an AMD processor, that is...). I don't need a .sig

Intel is changing markets anyways

[pyros]

They are moving towards network chips and away from desktop chips. Meanwhile, AMD is moving towards desktop chips and away from network chips. It's like corporate wife-swapping.

err...

[Wakko Warner]

...you *do* know that, clock for clock, the P4 is *significantly* slower than the P3?

- A.P.
--

"One World, one Web, one Program" - Microsoft promotional ad

Re:err...

[um... Lucas]

Clock for clock disregarding everything else, yes, the PIII is faster, but that's just looking at the pipelines. It's also got the 4x faster system bus which should make up for that, and then maybe add some... Really, it's hard to jump and make an assumption about the performance of a non-shipping chip, simply because it's not yet shipping. That's all i'm saying. The Athlon could very well beat out the P4 in similar benchmarks, but we just don't have any P4's out there with which to know for sure. That's all i'm saying, here.

Corporate PC Purchases Rule

[4of12]

I agree that Joe Sixpack will probably just get a Duron for Christmas since it's advertised to go more Megahurts (see the Register)than Intel and will be cheaper and more available.

Personally, it's really nice to see some competition out there in hardware land, but the corporate buyers at my workplace don't worry as much about

• MHz (a forgeable performance metric)
• my tasks/unit time (a better metric)
• the initial price

so much as the demonstrated reliability and trouble-free operation over the past 3 years. What those buyers are worried more about is how much it will cost to maintain that box over the next 3 years, i.e., how many visits required by a support tech.

That basically means, in x86 Windoze compatible land, that you get Dells with Intel chips, expensive RDRAM notwithstanding, since they have historically been rated as the most reliable.

I love the Athlon and would like to get a NICE DDR Thunderbird system so it wouldn't be so starved for memory. By NICE, I mean with redundantly extravagant cooling fans with quiet ball bearings and a large Ultra160 disk with high MTBF and low noise/heat.

But Guess What?!? All the OEMs see AMD and think,

"Hmmm...what kind of cheap 5hit components can I throw together and make a profit selling?"

The only alternative is to build your own Athlon system. But I think it shows where there are gaps in marketplace based on building a solid AMD based system. Until someone puts together solid AMD-based systems for 3 years running, or gets Dell to abandon its Intel-only policy, there won't be the penetration and competition in the corporate marketplace to give those buyers a real choice and reasonable prices.

All the performance leading technology that the K7 or K8 can muster won't change this reality.

Athlon SMP is delayed to next year

[idot]

There will be also a SMP version of the 760 DDR chipset, however sources say that the 2 way system will be ready only next year, and the status of the 4 way system is still unknown.

Re:When do GHz chips get cheap (== under $500)?

[ackthpt]

More to the (OT) point, AMD is more responsible for Intel's dropping prices. They may regard AMD as an mere imitator, but they certainly are reacting to AMD, which speaks volumes.

DDR will be a good thing when we have a number of chipsets and motherboards to choose from. Fortunately for us, TaiwanInc. will have a number of these available by the end of the year, so we don't have to wait for Intel to "invent" it.

Vote Naked 2000