Linked by Thom Holwerda on Mon 12th Dec 2005 15:37 UTC
AMD "AMD's drive to 64-bit processors surprised everyone with its speed, even as detractors commented that there would be little or no performance gain on the desktop without a 64-bit OS and 64-bit applications. Whatever the doubts within the industry, Intel lost little time in offering its own version of AMD64, in the form of the EM64T extensions. Traditionally perceived as the under-dog in the cutthroat world of microprocessors, AMD managed to take the design initiative at exactly that moment Intel was fixated on power consumption and the move to dual cores. DigiTimes recently had an opportunity to discuss AMD's approach to microprocessor design with Dr. Raghuram Tupuri, Design Engineering, AMD."
Order by: Score:
Time to compile!
by Anonymous on Mon 12th Dec 2005 16:14 UTC
Anonymous
Member since:
---

> no performance gain on the desktop without a 64-bit OS and 64-bit applications

Like in Linux (Gentoo comes to my mind)... It's time to compile your code and get the gain!

Reply Score: 1

Impressed
by DittoBox on Mon 12th Dec 2005 16:22 UTC
DittoBox
Member since:
2005-07-08

Frankly I'm not really impressed with the processor speed as much as I'm with my AMD Athlon 64 (3200+, venice) full load operation temperature of a whopping 32C.

That's a huge improvement over my old 2400+ T-bred Athlon XP that underload ran at 70C.

Neither chips are overclocked or underclocked.

As for AMD, I've been a fan for years now. They've been the underdog, they've produced (IMHO) faster, cheaper and overall better chips than intel. I'm not sure how they'll fair for the customer if they overtake intel entirely, but we'll see.

Reply Score: 4

RE: Impressed
by pojo on Mon 12th Dec 2005 16:43 UTC in reply to "Impressed"
pojo Member since:
2005-10-05

The best thing is, I can now be impressed with both the speed, and the temperature. I have a 3500+ and I love the Cool n' Quiet technology that they included with that. When I'm just browsing the web, the processor sits at 996Mhz, but when I start to compile or play games, it ramps up to over 2Ghz. When it's going slower the processor fan barely spins at all..it's very quiet.

Reply Score: 2

RE[2]: Impressed
by DittoBox on Mon 12th Dec 2005 16:57 UTC in reply to "RE: Impressed"
DittoBox Member since:
2005-07-08

I'll have to give that a try. CPU freq. scaling has always seemed to work well on laptops, I guess it makes sense for the desktop.

My only problem is my CPU fan isn't loud at all, it's the 120mm fan I've got on the blowhole at the top of my case that makes so much noise (I built it for the old processor, I really don't need it that much with this new 64). I really need to get a fan speed controller.

I'll give the cool n quiet a try, sounds cool.

Reply Score: 1

AMD is good
by ZaNkY on Mon 12th Dec 2005 16:50 UTC
ZaNkY
Member since:
2005-10-18

I can't wait for the next gen CPUs from AMD. Who knows what they'll come up with next? AMD is constantly innovating. While Intel is desperately trying to extend the life of the Pentium, AMD has gone into 64-bit tech, dual and QUAD core (to be released). Intel has 64-bit and dual core, but only after AMD came out with them and nearly copying AMD's instructions sets and the like. Kind of sad.


My CPU runs at about 35C stock, 45C load, and It's an FX-57. How awesome is that? I can't wait till AMD decreases the DIE to 45 nm ;)

The only good thing I can say about Intel ATM is that they announced 45 nm die size, first, I THINK. I may be wrong, correct me if I am please.

--ZaNkY

Reply Score: 1

Venice
by Tom K on Mon 12th Dec 2005 16:53 UTC
Tom K
Member since:
2005-07-06

Venice is a sweet core. It overclocks very well, all without putting out too much extra heat. I've got my 3200+ Venice running at 2.4 GHz (400 MHz increase) with ease. The only thing holding me back is my 4*512 MB of PC3200 generic-Samsung RAM. I've seen these chips go up to 2.8 GHz on air, while only running ~+5C hotter.

Reply Score: 3

I am...
by suryad on Mon 12th Dec 2005 17:03 UTC
suryad
Member since:
2005-07-09

In the process os switching computers all over to AMD processors. Especially the 64 bit ones. I can only wonder as Zanky already mentioned what the next gen procs are gonna be like. I am waiting for socket M2 as that would give a bit more stability I think in the long run in terms of leeway for performance increases and upgradability. I am probably gonna hold out to ge a dual quad core setup for my DCC rig.

Reply Score: 1

It's Price/Performance baby!
by Javier O. Augusto on Mon 12th Dec 2005 17:09 UTC
Javier O. Augusto
Member since:
2005-08-10

I've just bought a couple of things for my wife's new PC. A Sempron64 2600+ palermo core, Foxconn mobo based on the SiS761GX chipset (you can see a review right after this article, near the bottom) and a Samsung 512Mb DDR400.
All I can say is Woooaoo, the best bang-for-the-buck I could ever imagined. I runs fast, cool, quiet.. I got impressed! It's a very nice Desktop PC (for what my wife eventually does) and we are both happy with it!

Reply Score: 1

RE: It's Price/Performance baby!
by joelito_pr on Mon 12th Dec 2005 19:23 UTC in reply to "It's Price/Performance baby!"
joelito_pr Member since:
2005-07-07

Yeah, I just built one of those for my brother-in-law but with a via chipset. Heck I think that with a dedicated video card and 1gb RAM could probably run Vista

Reply Score: 1

RE: It's Price/Performance baby!
by Anonymous on Mon 12th Dec 2005 19:39 UTC in reply to "It's Price/Performance baby!"
Anonymous Member since:
---

I definitely agree. I'm converting old PIII boxes to Sempron64 ones.

I got a pair of Sempron64 2800+ setups for myself, and I overclock them to 2.4Ghz (300Mhz FSB) with stock heatsink! That's buying a 1.6Ghz budget CPU, and getting 2.4Ghz out of it! (Using a cheap nForce3 250GB mobo)

I increased the core voltage slightly (about 0.1v) to get it rock stable (as in it doesn't fail Prime95 tests)

Runs fine on Linux, Windows and FreeBSD.
(Haven't tried AMD64 mode yet).

I would NEVER touch a P4-based solution.
(That's my rule since they were released).

Reply Score: 0

amd 64 support
by Anonymous on Mon 12th Dec 2005 18:14 UTC
Anonymous
Member since:
---

yeah I said something like this on the ubuntu thread. basicly as awesome as Amd's 64bit stuff is, no one who develops cares. so until people compile and fully support 64 bit, its useless for any one who has to actually use their computer. I need to get an apple soo bad, I'm so sick and tired of writing everything from scratch. Of wasting weeks months getting ridiculous things like the dvd working. this shit should just work out of the box pleaseeeeeee.

Reply Score: 0

RE: amd 64 support
by miscz on Mon 12th Dec 2005 19:19 UTC in reply to "amd 64 support"
miscz Member since:
2005-07-17

DVD won't work on OS X out of the box. You'll have to pay for codecs and DVD support separately - at least it's what I've read in some review recently. It's cheaper and easier on Ubuntu after all IMO.

Reply Score: 1

RE[2]: amd 64 support
by DittoBox on Mon 12th Dec 2005 19:39 UTC in reply to "RE: amd 64 support"
DittoBox Member since:
2005-07-08

Though I've not tried it, the OS X Panther box states on the right side under key features: DVD Player 4.

I'm not sure about Tiger but I would assume that it includes version 5 maybe?

If the DVD-R drive in this titanium powerbook (1ghz g4, dvi etc. not bad for free!) I've got sitting on my desk worked, I'd give it a try. But the PRAM battery, the dc-in board, the optical drive ribbon cable and the jkey are fubar or missing.

Reply Score: 1

RE: amd 64 support
by rayiner on Mon 12th Dec 2005 22:52 UTC in reply to "amd 64 support"
rayiner Member since:
2005-07-06

Except all the people not on Windows. Everything on my desktop is 64-bit compiled (except Firefox 1.5, grr), and its a nice 10% - 20% speedup over 32-bit mode. Even on Windows, you've got 64-bit native versions of all the bundled Windows apps (well, IE), amd64 versions of Mathematica and Matlab (the amd64 version of Matlab is ridiculously fast), and a whole bunch of other apps.

Reply Score: 1

Anonymous
Member since:
---

There's a repeat mention of x84 architecture in there! Someone was definitely not paying attention, and I'm guessing it isn't the interviewee!

Reply Score: 1

64bit, no speed gain
by Anonymous on Mon 12th Dec 2005 20:16 UTC
Anonymous
Member since:
---

"...that there would be little or no performance gain on the desktop without a 64-bit OS and 64-bit applications"

Not again this misconception. 64bit address pointers do not, give any extra performance - period! What makes the AMD64 CPUs faster with software specifically compiled for it are mostly EM64T and the extra registers - not the size of pointers or registers. You would get the same speed in 32bit code if you used a similar instruction set and more registers. Just because the 64bit address space and the extra speed come in the same processor, doesn't mean one causes the other.

Of course, none of those who sells 64bit stuff to us (AMD, Apple, etc) has any interest in clearing up this popular misconception. The more customers think "64bit is faster", the better.

Reply Score: 0

RE: 64bit, no speed gain
by suryad on Mon 12th Dec 2005 20:32 UTC in reply to "64bit, no speed gain"
suryad Member since:
2005-07-09

Agreed but the archtiecture of the K8 processor does have benefits over Intels P4 64 bit or not.

Reply Score: 1

RE[2]: 64bit, no speed gain
by Anonymous on Mon 12th Dec 2005 20:56 UTC in reply to "RE: 64bit, no speed gain"
Anonymous Member since:
---

I'm not going down that road...I don't care much for Intel vs AMD. I just buy whatever's cheap and stays cool.

Reply Score: 0

RE: 64bit, no speed gain
by rayiner on Mon 12th Dec 2005 22:56 UTC in reply to "64bit, no speed gain"
rayiner Member since:
2005-07-06

When people say 64-bit OS and 64-bit applications in reference to amd64, they are referring to applications compiled in 64-bit mode, with the microarchitectural enhancements. Nobody is misconceiving anything --- you're just being anal and not picking up on the OPs tacit assumption.

PS: What makes the AMD64 CPUs faster with software specifically compiled for it are mostly EM64T

What does that statement mean? EM64T is just another name for Inte's version of AMD64. Its not a microarchitectural feature, and doesn't make anything faster.

Reply Score: 1

RE[2]: 64bit, no speed gain
by Anonymous on Tue 13th Dec 2005 08:11 UTC in reply to "RE: 64bit, no speed gain"
Anonymous Member since:
---

Nobody is misconceiving anything

A lot of people are, unfortunately. I see a lot of folks raving about how that CPU can now process twice the bytes than before, which is nonsense. Any decent 32bit CPU already supports double precision floats, long long integers and has a memory bus with at least 64bit.

What does that statement mean? EM64T is just another name for Inte's version of AMD64.

My bad, replace it with whatever the name of the new instruction set is. Too many abbreviations...

Reply Score: 0

RE[3]: 64bit, no speed gain
by rayiner on Tue 13th Dec 2005 08:36 UTC in reply to "RE[2]: 64bit, no speed gain"
rayiner Member since:
2005-07-06

A lot of people are

"Nobody is..." is a figure of speech. I was saying that the OP wasn't misconceiving anything, not unless you went out of your way to interpret his words in that manner.

unfortunately. I see a lot of folks raving about how that CPU can now process twice the bytes than before, which is nonsense.

Your point is well taken, but strictly, this is true. 64-bit CPUs can process twice the bytes as before. It's just that, usually, we're concerned about words, not bytes, and words are half the size on 32-bit CPUs, so the number of words per second is the same. However, in many important cases, we are concerned about bytes. Consider something like infinite precision integers (where a single number can be dozens of bytes). Indeed, multiplication on an arbitrary-length integer using 64-bit operations is actually 4x as fast as when using 32-bit operations (because the multiplication operation is non-linear).

Any decent 32bit CPU already supports double precision floats, long long integers and has a memory bus with at least 64bit.

Long long integers are emulated on 32-bit CPUs, using a sequence of 32-bit operations.

Reply Score: 1

re: 64 bit, no speed gain
by Anonymous on Mon 12th Dec 2005 21:15 UTC
Anonymous
Member since:
---

LOL so true. I hate hearing fanboys say 64 bit=performance improvement. There is no logic in it. It's the extra registers, and they're aren't being used right now. 64 bit is hype. When was the last time anybody used windows x64 as their desktop OS?

The real benefit of the athlon 64 was the integrated memory controller and the hypertransport link (for dual-core). Intel won't have any of this for at least a year or two on the desktop.

Please people, if you're thinking about buying a notebook, don't rule out yonah because it's 32 bits and merom is 64 bits.

I don't even see how 64 bits would help a laptop. If you need the extra performance from the extra registers, then you shouldn't be buying a laptop...

Reply Score: 0

RE: re: 64 bit, no speed gain
by Anonymous on Mon 12th Dec 2005 21:37 UTC in reply to "re: 64 bit, no speed gain"
Anonymous Member since:
---

>When was the last time anybody used windows x64 as
>their desktop OS?
I use Windows XP X64 on my laptop.

Reply Score: 0

RE: re: 64 bit, no speed gain
by rayiner on Mon 12th Dec 2005 22:59 UTC in reply to "re: 64 bit, no speed gain"
rayiner Member since:
2005-07-06

LOL so true. I hate hearing fanboys say 64 bit=performance improvement. There is no logic in it. It's the extra registers, and they're aren't being used right now. 64 bit is hype. When was the last time anybody used windows x64 as their desktop OS?

1) Everybody who is not on Windows is using these extra registers. Even a lot of people who are on Windows (the x64 version) are using the extra registers. Think the workstation/multimedia market.

2) Why is the desktop market the only one that matters? Lot's of people use 64-bit OSs on amd64 in the workstation, supercomputing, and server markets. These are AMD's most profitable businesses, remember.

Reply Score: 1

RE: re: 64 bit, no speed gain
by Anonymous on Mon 12th Dec 2005 23:33 UTC in reply to "re: 64 bit, no speed gain"
Anonymous Member since:
---

"LOL so true. I hate hearing fanboys say 64 bit=performance improvement. There is no logic in it. It's the extra registers, and they're aren't being used right now. 64 bit is hype. When was the last time anybody used windows x64 as their desktop OS?"

I know one person on Windows x64. Plenty of Linux and *BSD distros have AMD64 compiled binaries. Also, even if the extra registers in AMD64 are unused right now, they almost certainly will be used in the future, as more 64bit apps become available. Anyways, just replace 64 bit in your paragraph with SSE3.

"I don't even see how 64 bits would help a laptop. If you need the extra performance from the extra registers, then you shouldn't be buying a laptop..."

If you can get an almost free speed boost from a 64bit CPU, why the heck not get a 64bit CPU?

Reply Score: 0

RE: re: 64 bit, no speed gain
by Anonymous on Tue 13th Dec 2005 01:28 UTC in reply to "re: 64 bit, no speed gain"
Anonymous Member since:
---

If you have a bus 64bit-wide, instead of the regular 32bit-wide, you can transmit the double of the information in a bus clock cycle. Now, add the fact that the architecture has extra registers. Now, add the fact that it has an integrated memory controller (that's what's hypertransport is). What is that? You want to use more than 4GiB of memory? No you don't, you're limited to 32bit pointers.

So I guess, yeah, 64bit is a lot of hype.

Reply Score: 0

RE[2]: re: 64 bit, no speed gain
by CPUGuy on Tue 13th Dec 2005 02:12 UTC in reply to "RE: re: 64 bit, no speed gain"
CPUGuy Member since:
2005-07-06

Maybe I missread but...

Hypertransport is the method in which the processor is linked to the memory controller, PCI bus, etc... via hypertransport.

Think of it as getting a fiber connection instead of DSL.

Reply Score: 1

RE[2]: re: 64 bit, no speed gain
by Anonymous on Tue 13th Dec 2005 08:14 UTC in reply to "RE: re: 64 bit, no speed gain"
Anonymous Member since:
---

If you have a bus 64bit-wide, instead of the regular 32bit-wide, you can transmit the double of the information in a bus clock cycle.

You do know that regular Pentium 4s and Athlon XPs have a 64 bit data bus already?

Reply Score: 0

RE: re: 64 bit, no speed gain
by Anonymous on Tue 13th Dec 2005 02:04 UTC in reply to "re: 64 bit, no speed gain"
Anonymous Member since:
---

LOL so true. I hate hearing fanboys say 64 bit=performance improvement. There is no logic in it.

Well, that's not ENTIRELY true. There ARE performance gains from 64 vs. 32 bits, as you now have the ability to move 64 bits in one opcode instead of two. When shoving around large amounts of memory from one point to another it IS faster. Video and texture processing, large string manipulation, basically anything that moves around large amounts of memory WILL see a performance boost. MOVSB (8 bit) < MOVSW (16 bit) < MOVSD (32 bit) < MOVSQ (64 bit), as with the smaller operations it takes more loops to move the same amount of memory. it's that simple.

But there's also some 'waste' on the bus, as for example if you are only working on information one byte at a time or try to move around sections of memory that are not an exact multiple of eight bytes.

Of course, that the A64 preprocessor will actually turn "MOV ECX,9; REP MOVSD" into "MOV ECX,4; REP MOVSQ; MOV [esi],[edi];" so the 64 bit bus actually gets used for most of that transfer doesn't hurt... which is probably a hefty part of why the Barton 2500+ and the A64 3000+ both run at 1.82mhz and why the fastest A64 chip they make is a mere 2.4ghz (rated at 4800... what is this a PPC?)

Even if it didn't, the extra 32 datalines are used as read-ahead for the next operation... essentially doubling your FSB speed (AGAIN). Processors now run at many times the base bus speed, meaning that if you are doing a memory operation that takes two bus ticks but only one internal tick, given that the CPU can do two (or more) internal clocks per bus tick, reducing that memory read to one bus tick nets a noticeable performance gain as well.

Actually hits on one of the things you'll hear discussed from time to time (this is much more true on the PPC and AMD chips) that you reach a point where increasing the CPU's clock multiplier stops having any effect on the actual execution speed. If all your opcodes only took one CPU tick, and reading the code and data required for said opcode only took one tick, there would be no need to run the CPU faster than the bus - clock multipliers would have no effect... to the point that you could (in theory) have a 800mhz cpu that runs code faster than a 3.8 ghz Prescott. In fact, there are a LOT of opcodes that only take one clock cycle, and as such actually run no faster on uber-multiplied CPU's as they would if you were running 1:1 (increment of a single register for example) unless of course you are repeatedly calling the same section of code and the the higher speed cache inside the processor prevents you from needing to go to the memory bus. This is actually WHY cache sizes are climbing is the cache is faster than BUS ram, as it actually lets the multipliers have some effect.

As to just 'adding' the opcodes to a 32 bit chip, that would actually be slower; You'd have the same situation as the 386SX where you are trying to do memory operations bigger than your data bus, resulting in needing to perform twice as many memory operations - hardly an improvement. More registers and opcodes for them WOULD offer a speed boost, but not on existing code - something a wider data bus, preprocessing and a good caching model CAN do.

Before making wild statements about it having no impact, you might want to study up on assembly - you might learn something.

Reply Score: 3

What's a "Geode" CPU?
by JohnMG on Tue 13th Dec 2005 02:03 UTC
JohnMG
Member since:
2005-07-06

I was just poking around the Tyan site looking at their system boards and came across one that works with an AMD Geode proc:
http://www.tyan.com/products/html/tomcatk7m.html

What's this Geode CPU? It looks like it works with the same CPU mobo socket as the Athlon (socket A). The motherboard is a so-called "FlexATX"... Anyone have any experience with this Geode processor?

Reply Score: 1

Anonymous
Member since:
---

Is Windows XP X64 faster on day to day tasks? How about 3D games?

Will Vista come out as 64 or 32 bit?

Reply Score: 0

hypertransport
by Anonymous on Tue 13th Dec 2005 04:00 UTC
Anonymous
Member since:
---

.. is just an I/O technology to connect between cpus and from cpu to I/O devices. It's analogous to PCI-Express/Infiniband/etc. So every AMD64 chipset would contain a hypertransport-to-whatever bridge. For example, Nforce4 uses a HT-to-PCIexpress bridge internally for its PCI-E root complex (top level PCI-E device under the cpu).

However hypertransport has it's own slot form factor: HTX, for native, non-bridged I/O. But the only companies I'm aware of currently that offer HTX boards and devices are Iwill and Pathscale.

Reply Score: 0

re: RE: re: 64 bit, no speed gain
by Anonymous on Tue 13th Dec 2005 04:23 UTC
Anonymous
Member since:
---

I should've explicitly qualified that statement as applying to desktop user though I felt it was understand given that I referred to notebooks which aren't known for their great performance or >4GB. For web-browsing, word, and even gaming for now there is no benefit. Far cry 64 bit does give extra visuals for no performance hit, but one game isn't worth buying a 64 bit processor. Even then, the visuals aren't worth it.

I'm just saying that 64 bit is more often used as a marketing word. It's one of those things where if you're asking if you need it, then you don't.

Reply Score: 0