Worth losing EM64T?

[Dillusion]

Right now i have a 630 3.0, but i have a chance to trade a shuttle SB95P i have laying around in my closet for a 570J 3.8 CPU. The only thing stopping me is the fact that the 570 does not support Windows x64. Is the x64 windows that imminent that its not worth losing the ability to run it? Or should i not worry about it and stick with 32-bit for awhile...

 

[robberbaron]

64bit is great in benches but at this point is a bitch to use and keep running stabily. I wouldn't give a damn about it, especially since Longhorn will debut in both 32 and 64 bit.

 

[FadedSpark]

go for it, at this point xp x64 is in the baby stages and there aint much that normal windows cant do, PLUS its a 3.8... thats an 800MHZ difference and if you really want to go back to 64bit its not like you want have something to turn too lol, unless your selling your 630 lol

 

[ashmedai]

Longhorn will debut

It will? Seems like forever that MS has been saying that.

But 64-bit, really not incredibly useful for most people (yet?).

 

[Peach]

you can always go back and use your emt64 cpu.

 

[Talonz]

you can always go back and use your emt64 cpu.

QFT

 

[Dillusion]

you can always go back and use your emt64 cpu.

Na, i always sell my old hardware, need the money

Are you sure longhorn will be available in 32-bit? Linkage?

 

[robberbaron]

Na, i always sell my old hardware, need the money

Are you sure longhorn will be available in 32-bit? Linkage?

http://www.windowsitpro.com/windows...rticleID/46175/windowspaulthurrott_46175.html

Microsoft builds 32-bit and 64-bit builds of Longhorn every single day now. Neither takes precedence over the other. The two versions are co-developed and will be launched simultaneously. There will only be one version: During install, you can choose between the 32-bit and 64-bit version if you have a 64-bit PC. Nice.

 

[ashmedai]

Now if only more things had 64 bit drivers.

 

[perplex]

it's not worth thinking about 64bit for another 2 years yet (at the minimum). this is because as people have stated longhorn will be released with 32bit - this shows that even upto early 2007, people will be making 32bit software mostly. also like so many have mentioned, software makers know 32bit is a much bigger market for a long time yet. because they want the software to be able to run on the majority of computers.

multicore is going to be slow too :/

 

[Obi_Kwiet]

Besides Intel's 64 bit capabilities leave something to be desired.

 

[Peach]

Yeah, afaik its not pure 64 bit on the chip, its kind of emulated i believe

 

[jon67]

Go for the 570J man, it wipes the floor with the 630 regardless of 32/64 bit... it's actually the fastest CPU around for some single-thread applications

 

[robberbaron]

Yeah, afaik its not pure 64 bit on the chip, its kind of emulated i believe

Meh, it still gets the job done. Just not as well as the Hammer.

 

[Donnie27]

Besides Intel's 64 bit capabilities leave something to be desired.

When folks make statements like that, it means that it's either a lack of understanding or FUD! 64bit is a mixed bag and each architecture takes turns running it faster or show no difference at all, just as they do with 32bit software. Too many folks are are victims of marketing.

GamePC 64bit tests

At that time these two were in the same price range. 561 should go on sale at the end of this month.

Donnie27

 

[Scali]

Too many folks are are victims of marketing.

That's not marketing. That's ignorant !!!!!!s spreading lies on internet forums etc.
Neither Intel nor AMD ever mentioned that the competition would have an inferior 64-bit implementation or anything (they'd never get away with it anyway. AMD was even reprimanded for claiming that their NX bit would protect against virii, they had to pull the ad off the air). It's all FUD spread by !!!!!!s.

 

[Intel_Hydralisk]

Na, i always sell my old hardware, need the money

Are you sure longhorn will be available in 32-bit? Linkage?

But you have a shuttle laying around in your closet...

 

[visaris]

Besides Intel's 64 bit capabilities leave something to be desired.

This is entirely true. AMD has an IOMMU and Intel doesn't. This means that on an Intel CPU, any time you have a device that wants to do DMA I/O to an address above the 4GB boundry, the OS has to use a bounce buffer! The bounce buffer is slow, and in a way, it defeats the whole point of DMA in the first place, which is to save the processor from I/O.

Perhaps people who do not agree with Obi_Kwiet meant: "AMD and Intel 64-bit is about the same with respect to games on a system with less than 4GB of RAM."
That may be correct. But to say that Intel did as good of a job with 64-bit as AMD would be stupid, IMHO.

EDIT:
Perhaps this version won't be so inflammatory.

 

[Scali]

But to say that Intel did just as good of a job with 64-bit as AMD is just plain stupid.

I don't think anyone said that in the first place.
I see that someone said that both have their strenghts and weaknesses. And that is the smartest thing anyone said in this thread, if you ask me.

 

[Donnie27]

That's not marketing. That's ignorant !!!!!!s spreading lies on internet forums etc.
Neither Intel nor AMD ever mentioned that the competition would have an inferior 64-bit implementation or anything (they'd never get away with it anyway. AMD was even reprimanded for claiming that their NX bit would protect against virii, they had to pull the ad off the air). It's all FUD spread by !!!!!!s.

I just said marketing because I didn't want to say something that'd get me banned or warned, I know you're right though. I like it when the market watchdogs keep Intel and AMD in line and keeps the BS count low.

With that said, there have been tons of almost truths or little white lies.

Games People Play We finish up our slide show review with some pretty slick AMD graphs.

Most of what I said had more to do with AMD posters, AMD friendly Webmasters, and etc.. say, not AMD themselves BTW.

Donnie27

 

[Dillusion]

But you have a shuttle laying around in your closet...

Its been waiting to be sold....im in debt from that.

 

[Donnie27]

This is entirely true. AMD has an IOMMU and Intel doesn't. This means that on an Intel CPU, any time you have a device that wants to do DMA I/O to an address above the 4GB boundry, the OS has to use a bounce buffer! The bounce buffer is slow, and in a way, it defeats the whole point of DMA in the first place, which is to save the processor from I/O.

Perhaps people who do not agree with Obi_Kwiet meant: "AMD and Intel 64-bit is about the same with respect to games on a system with less than 4GB of RAM."
That may be correct. But to say that Intel did as good of a job with 64-bit as AMD would be stupid, IMHO.

EDIT:
Perhaps this version won't be so inflammatory.

No, what's stupid is you more than likely didn't read that link and then make crazy statements like that. 2CPU also has 64bit results that don't look much different than GamePC's. The you ought to make sure it's Venice based so the RAM doesn't drop down to PC-2700 speed when all slots are used, no matter how much RAM is used.

The amount of memory talked about is MOOT if the controller is limited to 4 banks@128bit. You'd need some awefully big sticks (At least 2GB singled sided each) to go past 4GB. Since AMD copied Alpha pretty much, that's how and why Alpha went to an integrated controller long before the first Hammer shipped. More controllers was better than one. IBM did this with Summit (multiple North Bridges) to combat that as well.

Donnie27

 

[visaris]

No, what's stupid is you more than likely didn't read that link and then make crazy statements like that. 2CPU also has 64bit results that don't look much different than GamePC's. The you ought to make sure it's Venice based so the RAM doesn't drop down to PC-2700 speed when all slots are used, no matter how much RAM is used.

The amount of memory talked about is MOOT if the controller is limited to 4 banks@128bit. You'd need some awefully big sticks (At least 2GB singled sided each) to go past 4GB. Since AMD copied Alpha pretty much, that's how and why Alpha went to an integrated controller long before the first Hammer shipped. More controllers was better than one. IBM did this with Summit (multiple North Bridges) to combat that as well.

Donnie27

Are you locked into desktop mode or something? The only thing I commented on was this statement:

Besides Intel's 64 bit capabilities leave something to be desired.

Which is true. Intel does lack an IOMMU, and for boxes with more than 4GB of ram this can be an issue (in some cases a rather large one).


I think I even said:

AMD and Intel 64-bit is about the same with respect to games on a system with less than 4GB of RAM.

Which I think agrees with most of your post.

 

[Donnie27]

Are you locked into desktop mode or something? The only thing I commented on was this statement:

Which is true. Intel does lack an IOMMU, and for boxes with more than 4GB of ram this can be an issue (in some cases a rather large one).


I think I even said:

Which I think agrees with most of your post.

As I asked you to please read the link. GamePC's results make that statement wrong as hell. That's what I was disagreeing with in the first place.

Yes and as I said, talking about more than 4GB of RAM is MOOT since the controller can't support what the processor can. I wasn't being argumentive or anything, just pointing that out. If this were a car, it'd be like saying your motor can crank out 650HP but your transmission can only tranfer 200HP to the rear end.

Donnie27

 

[visaris]

I wasn't being argumentive or anything, just pointing that out.

Yeah, sorry. I get a little excited while posting at times. I'm not trying to piss anyone off either.

Yes and as I said, talking about more than 4GB of RAM is MOOT since the controller can't support what the processor can. ... If this were a car, it'd be like saying your motor can crank out 650HP but your transmission can only tranfer 200HP to the rear end.

This is the part I don't agree with. At my work there are boxes with 16GB of ram in them. I just don't see how one can say that talking about boxes with more than 4GB is moot when I work with one every day. I it's perfectly valid to complain about features that effect my work. And yeah, again, I realize that most people don't use a 64-bit os, don't use 64-bit apps, and any differences in AMD vs. Intel 64-bit support doesn't really matter for them. I further understand that the IOMMU issue I mentioned will not effect most users, etc, etc.

The point is that this IOMMU issue does effect me, and this brings me to a rant that's not really about you, but about crap I get for a lot of my posts:
[rant]
I just don't feel I should have to add the words "for me" or "in my work" the front of every post I make. It shoud be assumed that my posts are comming from my point of vew and are talking about my needs. I also don't think I should have to add "IMO" ,or "IMHO", or "just my two cents", etc to the end of my posts. My posts are obviously what I think, and obviously my opinion, and I don't think I should have to lable them as such.
[/rant]

But, back the link you wanted me to check out. Sure intel may do a little bit better with 64-bit with lame, photoshop, or <insert benchmark here>. Those benchmarks really mean nothing to me. To make a car analogy, those show that intel can accelerate to 90MPH (90mph <-> little apps on a < 4GB box) faster than AMD, but they don't talk about how long it takes to reach 200MPH (200mph <-> a database/file server with 16GB ram).

Anyways, my BOIS has an option to turn the IOMMU off and on, so I'll do some tests and let you know what I find.

 

[Donnie27]

Yeah, sorry. I get a little excited while posting at times. I'm not trying to piss anyone off either.

This is the part I don't agree with. At my work there are boxes with 16GB of ram in them. I just don't see how one can say that talking about boxes with more than 4GB is moot when I work with one every day. I it's perfectly valid to complain about features that effect my work. And yeah, again, I realize that most people don't use a 64-bit os, don't use 64-bit apps, and any differences in AMD vs. Intel 64-bit support doesn't really matter for them. I further understand that the IOMMU issue I mentioned will not effect most users, etc, etc.

The point is that this IOMMU issue does effect me, and this brings me to a rant that's not really about you, but about crap I get for a lot of my posts:
[rant]
I just don't feel I should have to add the words "for me" or "in my work" the front of every post I make. It shoud be assumed that my posts are comming from my point of vew and are talking about my needs. I also don't think I should have to add "IMO" ,or "IMHO", or "just my two cents", etc to the end of my posts. My posts are obviously what I think, and obviously my opinion, and I don't think I should have to lable them as such.
[/rant]

But, back the link you wanted me to check out. Sure intel may do a little bit better with 64-bit with lame, photoshop, or <insert benchmark here>. Those benchmarks really mean nothing to me. To make a car analogy, those show that intel can accelerate to 90MPH (90mph <-> little apps on a < 4GB box) faster than AMD, but they don't talk about how long it takes to reach 200MPH (200mph <-> a database/file server with 16GB ram).

Anyways, my BOIS has an option to turn the IOMMU off and on, so I'll do some tests and let you know what I find.

See boxes with 16GB in them too, but they're not AMD. No, I didn't make it an Intel vs. AMD war either. They both have lied so much either company wouldn't know the truth if it jumped up and bit them on the asses. Believe me, I agree with you more than I disagree. I can only go by what I've seen and I'm talking about 4GB per Processor, NOT the total amount in the rig. I've not seen or heard of an AMD rig with more than 8GB for a Dualie. I saw one on a Web site with 16GB but that was a Quad proc system.

If you got a link, pics or etc.., I'd like to see it? You're right that I was mainly talking about the Desktop though. So let's get back to Desktops because that's what the thread was about.

I don't really care *that much about 64bit right now but would still like a 3700+. It's more like, "hey, may as well" when it comes to 64bit. Better gaming performance, less power draw, easier to cool, more mature platforms and DDR1 are reasons that seem more important. No, I don't have to trash Intel because I want to build an AMD based rig. Just as I didn't have to trash AMD to build my Intel one. If I were going Intel I'd wait for the 561 or maybe a 551. These are Prescotts with X86-64, 1MB (no 17% higher latency like the 2MB model) and they're a newer stepping that supposed to run slightly cooler.

Now 2CPU and GamePC, have tested Dual Core and Dual Processor systems with WinXP64 Pro and that line about how Intel's did it, is just wrong, both results show as much. A couple of tests show 64bit helping Intel more than AMD or was that missed?

Donnie27

 

[visaris]

See boxes with 16GB in them too, but they're not AMD. No, I didn't make it an Intel vs. AMD war either. They both have lied so much either company wouldn't know the truth if it jumped up and bit them on the asses. Believe me, I agree with you more than I disagree. I can only go by what I've seen and I'm talking about 4GB per Processor, NOT the total amount in the rig. I've not seen or heard of an AMD rig with more than 8GB for a Dualie. I saw one on a Web site with 16GB but that was a Quad proc system.

If you got a link, pics or etc.., I'd like to see it? You're right that I was mainly talking about the Desktop though. So let's get back to Desktops because that's what the thread was about.

I don't really care *that much about 64bit right now but would still like a 3700+. It's more like, "hey, may as well" when it comes to 64bit. Better gaming performance, less power draw, easier to cool, more mature platforms and DDR1 are reasons that seem more important. No, I don't have to trash Intel because I want to build an AMD based rig. Just as I didn't have to trash AMD to build my Intel one. If I were going Intel I'd wait for the 561 or maybe a 551. These are Prescotts with X86-64, 1MB (no 17% higher latency like the 2MB model) and they're a newer stepping that supposed to run slightly cooler.

Now 2CPU and GamePC, have tested Dual Core and Dual Processor systems with WinXP64 Pro and that line about how Intel's did it, is just wrong, both results show as much. A couple of tests show 64bit helping Intel more than AMD or was that missed?

Donnie27

Well, it sounds like we agree for the most part. I'm just really dissapointed that Intel did not provide an IOMMU with their 64-bit extentions. Perhaps they will in newer chips though.

 

[Scali]

So you rant about the IOMMU (got any benchmarks on how that affects anything? I can't imagine what kind of performance advantage we're talking about here)...
I could rant about the poor performance of SSE/SSE2/SSE3 on AMD systems, which is something that Intel has implemented far better (even before going 64-bit, but with 64-bit it's become even more important, since x87/MMX/3DNow! are no longer supported by Windows).

Bottom line is still: both have their strong and weak points.

 

[robberbaron]

I could rant about the poor performance of SSE/SSE2/SSE3 on AMD systems, which is something that Intel has implemented far better

I'm pretty sure SSE depends on high clocks regardless of IPC in order to do well. Since AMD lacks the clock cycles of a Pentium 4, the SSE won't perform that great. The K8 still manages to kick some serious floating point ass in 64bit mode, however.

 

[visaris]

So you rant about the IOMMU (got any benchmarks on how that affects anything? I can't imagine what kind of performance advantage we're talking about here)...

From a theoretical standpoint one can see that the impact is huge:
For normal DMA, an I/O device (HDD, GBE, etc) will write a block directly into userspace memory and flag an interrupt when it is finished. This means that the I/O device hardware can give data to the application that needs it directly. Sounds great doesn't it? The trouble comes in with 64-bit. While the CPU may have a 64-bit address space, the PCI standard has been left unchanged. This means that PCI devices cannot write directly to memory above the 4GB boundry. AMD solves this problem by adding an IOMMU. The IOMMU is used to map a portion of the pci "hole" to 64-bit addresses. This allows AMD, in hardware, to allow DMA to addresses higher than 4GB.

To summarize:
On an Intel system, when an application wants data from a device, and wants to put that data into memory above the 4GB boundry, the data must necessarily be coppied from a buffer below the 4GB boundry to the place the application wants it, above the 4GB boundry. This is really slow.

Let me put it this way. 64-bit is about the future. One of the main reasons for switching to 64-bit systems is to be able to run with more than 4GB of ram. It just doesn't make sense to me, from the forward direction i'm facing, to say that Intel's 64-bit implementation is any good, because one of the main reasons for going 64-bit in the first place is crippled by their lack of an IOMMU and the ability to get data from a device to where it is wanted without having to use a buffer under the 4GB boundry as a stepping stone.

Bah, I don't know how to say this cleanly. Perhaps RedHat can say it better:

“Software IOTLB — Intel EM64T does not support an IOMMU in hardware while AMD64 processors do. This means that physical addresses above 4GB (32 bits) cannot reliably be the source or destination of DMA operations. Therefore, the Red Hat Enterprise Linux 3 Update 2 kernel "bounces" all DMA operations to or from physical addresses above 4GB to buffers that the kernel pre-allocated below 4GB at boot time. This is likely to result in lower performance for IO-intensive workloads for Intel EM64T as compared to AMD64 processors.”

Benchmarks. I could give you numbers from our systems at work but that's about all that I could really provide. Benchmarks of servers with multiple raid arrays, and multiple gigabit ethernet connections aren't exactly as common as people running sandra on their home boxes. When you do find good, I/O focused, benchmarks of the systems you want to compare you find that these systems are configured too differently and aren't exactly tested as AMD 64-bit with I/O vs. Intel's 64-bit with I/O. However, I think anyone who spent some time to be familiar with the IOMMU issue would see that Intel dropped the ball a little here.

I could rant about the poor performance of SSE/SSE2/SSE3 on AMD systems, which is something that Intel has implemented far better (even before going 64-bit, but with 64-bit it's become even more important, since x87/MMX/3DNow! are no longer supported by Windows).

And you would be right to do so. From most things that I've seen, AMD's SSE performance is not up to par with Intel, so this is an easy thing to agree with.

Bottom line is still: both have their strong and weak points.

I can agree with that. But still, from my point of view, where 64-bit is the future of x86 for years to come, I think that Intel's lack of an IOMMU is a serious weak point in their 64-bit offering.

 

[Donnie27]

And you would be right to do so. From most things that I've seen, AMD's SSE performance is not up to par with Intel, so this is an easy thing to agree with.

I can agree with that. But still, from my point of view, where 64-bit is the future of x86 for years to come, I think that Intel's lack of an IOMMU is a serious weak point in their 64-bit offering.

I shorten down to this part for a reason. Sure Intel SIMDs isn't up to par on AMD processors. Yet, these SIMDs do improve performance on AMD processors, just not as much as they do on Intel processors. Tech Report, Ace's, 2CPU, Hexus and others have shown just how much SIMDs do work and they're not just limited to Streaming as most folks talke about.

Intel has always depended on software. As one programer put it; "When they (Intel) get SIMD support, they rock. When they don't get it (SIMD support), they get rocked".

Here's a blast from the past that many didn't see!

Willy 1.5GHz

Anna at XBits had a very similar review but when Anton moved in, he promptly removed it. It feature 2000+ vs. 2GHz and 2.2GHz, same results LOL! Yes it's Intel's fault for depending on SIMDs, but IA32-X86 is software based anyway, right?

Donnie27

 

[LstBrunnenG]

Problem is that the chipsets that run these processors only support 4 GB of RAM anyway. Shall we start smashing NVidia, VIA, and Intel's chipset department?

 

[Epicenter]

Don't sweat it. EM64T on the P4 is a pretty kludgy implementation, and doesn't usually result in a significant performance gain. It actually causes DROPS In performance on a lot of tasks. Intel mainly just implemented it for compatibility's sake so there wouldn't end up being a Microsoft OS an AMD CPU would run that an Intel one wouldn't (the consumers would start to wonder what was wrong with the Pentiums if they couldn't run the latest incremental update to XP.)

 

[Scali]

I'm pretty sure SSE depends on high clocks regardless of IPC in order to do well. Since AMD lacks the clock cycles of a Pentium 4, the SSE won't perform that great. The K8 still manages to kick some serious floating point ass in 64bit mode, however.

I disagree. If AMD can get high performance with most things, regardless of clockspeed... then why not here? From what I heard, the SSE/SSE2/SSE3 implementation itself is rather weak (working on 64-bit processing units). We'll see where Intel's next-gen will take us. Then we'll see if you REALLY need high clockspeeds, or if Intel will manage to maintain its lead even when the clockspeed goes down. My guess is the latter.

 

[Scali]

From a theoretical standpoint one can see that the impact is huge:

Well, from a theoretical standpoint, the whole x86/PC architecture is pretty inefficient.
In practice it's no big deal.

I can agree with that. But still, from my point of view, where 64-bit is the future of x86 for years to come, I think that Intel's lack of an IOMMU is a serious weak point in their 64-bit offering.

Ofcourse, after all, it was added as an afterthought to this chip. Intel has yet to release its first 64-bit processor that was designed from the ground up with 64-bit in mind. The EM64T on P4 is basically nothing more than 'damage-control'.
And it does fine too. Intel offers 64-bit systems already, and before there will be any noticeable gains from the Athlon64, Intel will be ready with its new architecture. So I'm not too worried about any potential performance problems (as people pointed out, it's physically impossible to have more than 4 gb in common 64-bit Intel desktops anyway).

 

[Epicenter]

I'm pretty sure SSE depends on high clocks regardless of IPC in order to do well. Since AMD lacks the clock cycles of a Pentium 4, the SSE won't perform that great. The K8 still manages to kick some serious floating point ass in 64bit mode, however.

I'm fairly certain it's dependent on rapid re-execution of cached instructions, which are something you see a lot in loop-heavy code executing on deeply-pipelined, heavily-cacheladen processors like the P4s.