![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Barcelona and R600 to be a double release.
- Thread starter visaris
- Start date
Digital Viper-X-
[H]F Junkie
- Joined
- Dec 9, 2000
- Messages
- 15,123
Well it was backed up by some tehcnical info from visaris, and while you find it hashed out and tired, it makes alot of sense, AMD can drop a new cpu into their existing platform to take the lead, but will intel adding more cache allow them to catch up? or can they even use the same platform to catch up?
Dont get me wrong though, On the desktop Intel's technology is more then capable ind is the only real choice if you were buying new today. The technical issues it has wont affect you as a user even one little bit...
So dont let my bias effect your buying decisions.... If you decide to buy new, then Intel is the way to go. The technical issues wont affect you as a user in the slightest way.
Digital Viper-X-
[H]F Junkie
- Joined
- Dec 9, 2000
- Messages
- 15,123
don't worry, even though I dont know the underlying issues with archectures, I usually go with what fits my needs and financial status, I've been from amd to intel to amd to intel, I don't reallly have a bias, usually best bang for the buck gets my $! I personaly can't wait for AMD to release K10 it should be interesting and drive down $ of current core 2 + X2 chips, I'm already shocked to see X2 3800s @ 139 CAD,
imagine a low price, low power dual core htpc or "value" system,
While this is all certainly true, it's also the case that the vast majority of data tranfers in a system are either into or out of main memory subjecting the AMD to much the same issues you attribute to Intel's. Namely, a unidirectional bus with stiff penalties for switching between read and write. Now, if we're just talking bout CPU <-> System RAM communications, AMD has effectively eliminated one of the hops, thereby reducing first read / write latency signifigantly, and hiding those penalties in that case.
And honestly,the underlying problem is one that Core 2 is better equipped to deal with at this point in time, not only due to a larger shared cache to reduce the frequency of memory accesses, but also technologies that AMD will be implementing or expanding upon in Barcelona, like memory disambiguation to better predict indirect loads and out of order loads to lessen the amount of time wasted switching between read and write. Intel is reducing the overall number of read / writes to main memory from the CPU and reducing the number of switches between reading and writing, while AMD has focused on reducing latency of the those operations as much as possible.
Would Core 2 be faster with an HTT style I/O link and integrated memory controller to reduce latency where possible? Obviously, I doubt anyone would argue otherwise. But the flip side of that is AMD could stand to benefit from Intel's approach as well of reducing the overall number of times those penalties are incurred.
For some things HTT is a leap forward in design for the desktop and small server computers, but for most of our uses, its benefit is reducing the latency on that intermediate hop between I/O devices and main memory. I wouldn't trade ccHTT links for connections in a NUMA computer for much anything else, PCIe and 10GB ethernet hold promise as backplane solutions as well, but ccHTT does an excellent job of being both cost effective and fast in 2, 4, and 8 CPU designs. But at the same time the practical impact of HTT on the desktop is pretty minimal (as the fact that you can drop your HTT link speed down to 2x or so before having any real performance impact shows).
The placement of the northbridge of optimize CPU <-> Memory transfers is the real practical advantage AMD has.
I have this problem a lot too. I'll post something that is a generally accepted fact (such as HT > FSB), and the responses are "yeah?!? prove it you n00b!!!", (often followed by someone else posting: "pwned"), when a simple 5 min google search would have been sufficient. To be honest, it is hard to let it bother me any more. Most of the time I just laugh it off. How could I get mad at someone who is too lazy to educate themselves with a simple 5 min (or less) google search? I'm filled more with pity than anything else.
-----
Anyways, I'll try to sum up some arguments I've been trying to make here (and I suspect dubby has as well), and I do not think it is unreasonable to skip the links. The numbers/facts presented here are available all over the web for anyone willing to spend 5 min looking for them and are not currently under debate.
- Intel's chips, even at 1333MHz (which is, to my knowledge, not yet available on the desktop) and because of the FSB, do not have as much bandwidth available to them as comparable AMD chips. At 1333MHz, Intel's chips have a single 10.6GB/s link to them. That's it. This isn't even enough bandwidth to match the memory speed at DDR2 800 (12.8GB/s). This link is unidirectional, and time must be wasted to "flip" the bus around the other way. AMD, on the other hand, provides up to 20.8 GB/s for each chip, and this number is significantly increased in the server space due to the addition of more HT links. AMD's HT "bus" does not need to be "flipped", is bidirectional, and offers latencies that are much, much lower than those of Intel's FSB.
I can hear it now from all the people who don't want to learn anything: "Why does all that junk matter?!? All I care about is my FPS!!" If all you care about is your FPS, go check out the benches of the latest game: Core2 is probably on top, and you can go ahead and step out of the discussion now; this discussion isn't really for you. This discussion is for people who don't have any agenda and are simply pondering the details of the technology, frankly, because it interests them.
- Core2 (the actual chip) has a design such that these bus issues largely disappear. Intel uses a very large cache that is great at hiding memory and bus latency, and Intel also uses some really great prefetchers to keep the cache filled with relevant data. When it comes down to the bottom-line performance, Intel takes the win because a really good core architecture is able to more than make up for the deficiencies in Intel's platform.
This overall win by Intel doesn't change the fact the Intel has a poor platform by comparison. Imagine where your favorite Core2 chip would be able to go if it could get a better, faster supply of data via an IMC and HT links? I mentioned a while ago to another poster that nearly all traffic goes through the bus/HT at one point or another, and thus, GBe, PCIe, SATA, and all other devices frequently take a small penalty on the Intel platform because of the slower bus. While I will not speak on the qualitative impact this would have to an actual user, I will back Dubby up here; he is technically correct.
This small penalty on I/O with Intel chips will probably not be too significant on the desktop space, and it may not effect your purchasing decision. However, the penalty is there. We don't need any more n00bish posts on this issue. While this may only be relevant in the field of academic discussion, I'll remind you that academic discussion (while far more detailed and complicated than this simple post) is what designs your chips (and it's fun anyways).
Just as Intel needs to do some hard work to correct their platform issues (such as no Core2 servers with more than 2 sockets
-----
Yes, I must be: "Wow. This has got to be the worst case of FBism ever." because I have an open mind and actually take the time to think about the technology. Or perhaps it was because I made the claim that an Intel product is not perfect in every way? Bah, poor spectrumbx. It always amazes me that the debater that makes the most personal attacks and provides the least thought and discussion usually "wins" the argument. Lol, what's worse, such a person, or one who supports him? Lol, "pwned".
Anyways, in conclusion, I'm very interested to see how Intel's CSI will turn out, as I think their platform is holding them back. At the same time, I'm very interested to see AMD's new Barcelona core, as I think AMD's core is holding back the platform.
Further, I don't see why it is so hard for some people to admit that both companies, chips, and platforms have both strengths and weaknesses. Only by acknowledging them can we begin to ponder them, and we only arrive at a (semblance of) understanding after the pondering.
I don't think Duby has made this argument. I have no doubt that HT > FSB when you scale into large number of CPUs. The fact remains, that on the desktop, the FSB that Intel uses is good enough. And I didn't even have to Google it. I gather this information from technical articles and benchmarks provided from sites that I frequent. These articles appear on sites that I'm sure we all frequent. The real problem becomes when people make outrageous claims but don't provide any evidence. In this case, evidence is not telling people, "search google."
When I make claims that someone doesn't believe or that they contradict, I post evidence to support myself. Now, I already know that Core had better prefetch logic as I'm sure most here know as well. However, when duby made a claim to the contrary I posted a link to evidence that supports my position. His reply was simply to tell me to search google. There's no need, I've already provided evidence that supports my position and I already know the answer based on previous research that I have done. I thought it was common knowledge that Intel overcomes the advantage that AMD has with the IMC by using superior prefetch logic that helps to mask the latency issues involved with using a seperate northbridge as the memory controller. I saw this covered in entirety when the Core 2 was released.
This post is not meant to rail on you, but I do find it amusing that you pick out a duby post and use it is a springboard to tell him that you and him should feel pity for us. He made that post to contradict a post where I provided real evidence to support my position. This is analagous to someone telling you the sky is yellow, you showing them a scientific article that explains why the sky is blue, and them telling you that you don't know what you are talking about and should go to a library. It's a ridiculous rebuttal and if anyone should feel pity, it's those trying to refute the ridiculous claims that duby is making, "AMD has more functional PCIe," "AMD has better prefetch logic," "AMD has a better ethernet controller with lower latency," "AMD has a better SATA controller," etc. Now, who sounds rational and who sounds ridiculous?
Duby did not make the claim that AMD has more bandwidth with lower latency from the CPU to the ethernet controller or PCIe slots or anything else. He said that AMD implements a "more functional" PCIe controller. This is simply not the case. He also claims that AMD has a faster SATA controller. Anandtech's test (now you don't have to Google it) shows that the G965 performs identically to the SB600 in everything except burst rate. When factoring in burst rate, the SB600 wins by 1 MB/s while using 4% more CPU. Both controllers had a burst rate over 100 MB/s. That's a difference of less than 1%, within margin of error. I would say that the disk contoller of both AMD and Intel chipsets perform identically in most situations, especially those seen on the desktop. This is supported by an objective test that can be found online.
You can also check out here:
http://www.anandtech.com/mb/showdoc.aspx?i=2767&p=9
http://www.anandtech.com/mb/showdoc.aspx?i=2860&p=24
Yep, all the motherboards using the Marvell ethernet controller perform the best and there is much less than 1% difference between AMD boards and Intel boards (once again, well within margin of error). And if I'm reading it right, the Intel boards do it while using less CPU overhead. Not bad for old tech. Oh, also on those two pages you'll see that the USB controller on the Intel platform gets better bandwidth than the USB controller on the AMD platform. ICH8 FTW!
BTW, in ATI's (now AMD's) desktop chipset, the northbridge and southbridge communicate with each other using A-Link Xpress II, a PCI Express implementation. In fact, in AMD's own chipset only uses hypertransport for communication between the CPU and northbridge.
http://www.bit-tech.net/hardware/2006/03/01/ati_rd580_asus_a8r32-mvp/1.html
Gotta love the RD580. A-Link Xpress II is also used in their Intel chipsets.
Oh, according to AMD, the X2 has 8 GB/s bandwidth to the system.
http://www.amd.com/us-en/Processors/ProductInformation/0,,30_118_9485_13041%5E13076,00.html
The 20 GB/s number you get is only when you combine HT bandwidth + memory bandwidth. Unfortunately, AMD has the same memory problems that Intel does, a parallel unidirectional connection. The memory connection will still have to "flip" in an AMD system like it does in an Intel system. The only serial (bidirectional) connection that AMD chips have is the HT connection which, according to your own numbers, provides less bandwidth than Intel's FSB connection. Granted, Intel has to share this connection with the memory while the AMD chip doesn't. However, the HT number is also misleading. The 8 GB/s is the combined aggregate bandwidth. Meaning, that's the maximum bandwidth when you combine bandwidth going into and out of the chip on the unidirectional links. If you look at a link in one direction you get a maximum of 4 GB/s. However, in Intel's implementation, while information is coming into the chip, you could have a maximum incoming bandwidth equal to the entire bandwidth available 10.6 GB/s. You have to examine everything.
And I agree with you on some points. Hypertransport is simply the better option when moving into the future and into more CPUs. Intel also knows this which is why they are creating their own serial, p2p bus. My contention is simply that hypertransport doesn't do much to help AMD's platform in the desktop market. And because it is paired with the slower AMD CPU, the Intel desktop solution is simply better. The FSB is good enough for now when talking about connections needed by desktop CPUs.
It would be really interesting if the entire system went serial. Although, that would mean that we'd have to go back to RAMBUS-type memory and I think people have bad memories of that. However, I believe it is inevitable. Hard drives are now on serial connections, peripherals are on serial connections, expansion cards are moving quickly to serial connections (if only they made pci-express sound cards), ethernet controllers are on PCIe, north/south interconnects are using hypertransport (nVidia) or PCIe (AMD), AMD already has a serial bus and Intel is going that way. I welcome my serial overlords.
I think alot of this bickering and argument has to do with what I call the Ford vs Chevy syndrome.
Person A went shopping and bought a Ford. Person B went shopping and bought a Chevy.
Person A meets Person B, and they both discover that they bought a new vehicle. Person A doesn't want to have been bettered by Person B, and vice versa. Person A will never accept that Person B's product is better than his, because that meant that he made the wrong decision, and his feeble mind can't handle that kind of disappointment. So, Person A attacks Person B's product, calling it "crap", "shit", "stupid", and basically telling him that he got "wtfpwned". Person B makes similar claims, because with a similar feeble mind, he can't accept the fact that he may have made an incorrect purchase decision.
Which product is superior is highly irrelevant in this case. So it is in the Intel vs AMD crap that gets regurgitated here every day for the last....oh since the forum started.
I'll say this. Intel and AMD both suck. Nothing plays games better than my UltraSPARC running Solaris with no video card and only a serial interface. WTFPWNED NOOBLERS!
Person A went shopping and bought a Ford. Person B went shopping and bought a Chevy.
Person A meets Person B, and they both discover that they bought a new vehicle. Person A doesn't want to have been bettered by Person B, and vice versa. Person A will never accept that Person B's product is better than his, because that meant that he made the wrong decision, and his feeble mind can't handle that kind of disappointment. So, Person A attacks Person B's product, calling it "crap", "shit", "stupid", and basically telling him that he got "wtfpwned". Person B makes similar claims, because with a similar feeble mind, he can't accept the fact that he may have made an incorrect purchase decision.
Which product is superior is highly irrelevant in this case. So it is in the Intel vs AMD crap that gets regurgitated here every day for the last....oh since the forum started.
I'll say this. Intel and AMD both suck. Nothing plays games better than my UltraSPARC running Solaris with no video card and only a serial interface. WTFPWNED NOOBLERS!
AMD has to catch up by improving their core. The need a better execution engine. This is why they are behind Intel. Neither Intel or AMD chips are all that starved for bandwidth on the desktop. Simply put, AMD migrated to the serial bus before they really needed to (at least on the desktop). This does give them an advantage in terms of experience. However, Intel already has the upgrade to CSI (and an IMC) on their roadmap.
However, performance at this point is just speculation. We think that Barcelona will be faster than Core 2 but mostly because of improvements made to its execution engine. Intel might simply add more cache (which doesn't always help, BTW), or they may improve their own core to overcome the delta. Can they use the same bus? Don't know, but it won't be an issue for much longer anyway.
BTW, Barcelona introduces HT3, the next revision of hypertransport. However, and I may be wrong on this, you'll have to upgrade your motherboard to take full advantage.
That was my understanding as well. It should still work in the sAM2 boards, but you'll need to upgrade to sAM2+ in order to use HT3.
As far as your overly long rant, again that is your opinion.Of course you can have whatever opinion you want so I'll just leave it at that.
To settle this issue once and for all, your not going to change my mind on what I think a platform is and is not. Intels definition is wrong. In my opinion AMD has a better desktop platform. In my opinion AMD has a better architecture. My opinions are based on facts that I've observed over the years, which can be easily found using a quick google search.
AMD has laid out a clear path to what they want to achieve in the next two release cycles Intel has not. In the mean time AMD has a better buss, as well as some architectural features that will allow them to continue working with the existing architecture for at least the foreseeable future. We all have at least an idea what Barcelona will be. We dont know exactly how well it will perform, but I'm sure most of us have a clue. AMD also has Fusion on the books, with an integrated SIMD coprocessor that should work well with CTM....
As far as Intel goes, the should have CSI out on the desktop round about 2009, with no clear indication on a IMC. They should have a discreet GPU out round about 2010, with no clear indication on a integrated SIMD coprocessor. They claim to have improved release cycles to every two years, but still have no clear indications of what the next release will be despite being only a year away. Intel is still using a shared inclusive L2 cache despite contention, with no clear indication of what they will do to fix, or improve it.
I'm not saying that Intel has bad products, I'm just saying that they have some major problems to address that AMD already has. They got alot of work to do in order to catch up. I think they are on the right track with Conroe, but they still have alot of work to do.
As far as your overly long rant, again that is your opinion.Of course you can have whatever opinion you want so I'll just leave it at that.
To settle this issue once and for all, your not going to change my mind on what I think a platform is and is not. Intels definition is wrong. In my opinion AMD has a better desktop platform. In my opinion AMD has a better architecture. My opinions are based on facts that I've observed over the years, which can be easily found using a quick google search.
AMD has laid out a clear path to what they want to achieve in the next two release cycles Intel has not. In the mean time AMD has a better buss, as well as some architectural features that will allow them to continue working with the existing architecture for at least the foreseeable future. We all have at least an idea what Barcelona will be. We dont know exactly how well it will perform, but I'm sure most of us have a clue. AMD also has Fusion on the books, with an integrated SIMD coprocessor that should work well with CTM....
As far as Intel goes, the should have CSI out on the desktop round about 2009, with no clear indication on a IMC. They should have a discreet GPU out round about 2010, with no clear indication on a integrated SIMD coprocessor. They claim to have improved release cycles to every two years, but still have no clear indications of what the next release will be despite being only a year away. Intel is still using a shared inclusive L2 cache despite contention, with no clear indication of what they will do to fix, or improve it.
I'm not saying that Intel has bad products, I'm just saying that they have some major problems to address that AMD already has. They got alot of work to do in order to catch up. I think they are on the right track with Conroe, but they still have alot of work to do.
My "overly long rant" had plenty of links to evidence found on websites you might find if you actually searched. I stated fact, not opinion. Facts are supported by evidence, like the evidence I gave. As you can see, I've clearly done my research.
This only bothers me because people come here looking for advice and information. The last thing they need is someone clearly spouting false information. You shouldn't do that.
The whole argument is about the definition of the word platform, which Duby has completely wrong, and BigDH01 dominated him time and time again.
Considering socket only, theoretically AMD could "drop a new CPU in AM2" and take the lead. They could drop 65 nm chips into socket 939 if they want to. Intel could continue on 775 indefinitely if they wanted to. Neither company wants to do that because there is collusion going on between the chip manufacturers and the motherboard manufacturers.
Digital Viper-X-
[H]F Junkie
- Joined
- Dec 9, 2000
- Messages
- 15,123
No offense to BigDH01, but i don't consider support part of a hardware platform either =p, maybe part of a BUISNESS SOLUTION platform, Maybe they are argueing about 2 different types of platforms
No offense to BigDH01, but i don't consider support part of a hardware platform either =p, maybe part of a BUISNESS SOLUTION platform, Maybe they are argueing about 2 different types of platforms
Um.... Intel says there are several components to a platform. It lists hardware and then says...
Hardware, such as processors, chipsets, communications, memory, boards, and systems.
Seems pretty straightfoward to me. Or, you can view the specs of their Centrino platform. You'll see that the processor is mentioned specifically.
http://www.intel.com/products/centrino/duo/310199.pdf
It seems pretty clear that Intel considers the CPU to be part of the platform. I think Intel might know better how to define the word than either one of us, I mean, they've been manufacturing platforms for many years now.
Or, you could take Rich Heye's word on it.
“AMD and our infrastructure partners are committed to building innovative technologies based on the next-generation AMD Athlon and AMD Opteron platform,” said Rich Heye, vice president, platform engineering and infrastructure, Computation Products Group, AMD. “AMD is gaining critical mass with broad partner support of our eighth-generation platforms, and will help lead the industry toward an easy-to-adopt 64-bit computing standard.”
http://www.amd.com/us-en/0,,3715_13530_1260_6291%5E6310~23293,00.html
He doesn't say AMD's HT platform, or AMD's SATA controller platform, he says AMD's Opteron and Athlon platforms. Opterons and Athlons are what??? CPUs. I really don't feel like getting into this again... you can google it and see for yourself. I think everyone but duby considers the CPU to be part of the computing platform.
While you are correct that the CPU is part of the overall platform, I don't see any problem with further splitting this into two conceptual parts: the processor (CPU) and the supporting infrastructure (platform). While you may not like his use of terminology, I do not have a problem with his assertion that the CPU is not part of the platform, because of the way he decides to define the context.
Centrino is just a name. Just like Athlon or Opteron. I see no reason why "Athlon" and "Opteron" cannot refer to both a CPU and a platform. Or is it not cool to use the same name more than once?
-----
Any news on the HTX for the desktop front, or is that still off in the distant future?
Digital Viper-X-
[H]F Junkie
- Joined
- Dec 9, 2000
- Messages
- 15,123
I don't consider a cpu part of a platform, neither do I consider any add in cards or sata controllers, I do think that a platform is designed for a cpu or a cpu archetecture for me a hardware platform is the basic design of the chipset, motherboard and its interconnects,
AMDs platform allows for more diversity because of its current cpu archetecture,also it allows for better scaling with 4+ cpus and over all for the future AMD's platform should preform faster when paired with a faster cpu, where as intels is more limited because of the archecture of the cpu that it will connect with(not saying this has any say for current desktop pcs though as desktops don't seem to stress it) but from what I understand AMDs 4 and 8 way opterons scale better because of that. I guess shortest word I see a platform as a possibility, that is based on the CPU archetecture
Keep in mind this is my Opinion and in no way fact =p
the problem is both of you are wrong and right, you are argueing that Intels definition of a platform is right, he is arguing that his is right, the problem is that there are many different types of platforms, and many differnet definitions for each, so in essense the word platform is too general to use when debating what it is, even the dictionary has about 10 different definitions for it, a train platform is not the same as an artilary platform is it?
AMDs platform allows for more diversity because of its current cpu archetecture,also it allows for better scaling with 4+ cpus and over all for the future AMD's platform should preform faster when paired with a faster cpu, where as intels is more limited because of the archecture of the cpu that it will connect with(not saying this has any say for current desktop pcs though as desktops don't seem to stress it) but from what I understand AMDs 4 and 8 way opterons scale better because of that. I guess shortest word I see a platform as a possibility, that is based on the CPU archetecture
Keep in mind this is my Opinion and in no way fact =p
the problem is both of you are wrong and right, you are argueing that Intels definition of a platform is right, he is arguing that his is right, the problem is that there are many different types of platforms, and many differnet definitions for each, so in essense the word platform is too general to use when debating what it is, even the dictionary has about 10 different definitions for it, a train platform is not the same as an artilary platform is it
?While you are correct that the CPU is part of the overall platform, I don't see any problem with further splitting this into two conceptual parts: the processor (CPU) and the supporting infrastructure (platform). While you may not like his use of terminology, I do not have a problem with his assertion that the CPU is not part of the platform, because of the way he decides to define the context.
Centrino is just a name. Just like Athlon or Opteron. I see no reason why "Athlon" and "Opteron" cannot refer to both a CPU and a platform. Or is it not cool to use the same name more than once?
-----
Any news on the HTX for the desktop front, or is that still off in the distant future?
He can define platform however he'd like. That doesn't make his definition correct. It is in direct opposition to the largest semiconductor manufacturer in the world. This would be like me saying that an engine isn't part of a car because a car consists only of the transmission, suspension, frame, body, and battery. Would you say I was correct or incorrect? I would say incorrect as it is in opposition to every car manufacturer. This is the case that we have here.
Athlon and Opteron refers to both a CPU and a platform. The Athlon and Opteron platform are named as such because they include the Athlon and Opteron CPUs. It would be crazy to say that an Opteron platform (as defined by AMD) doesn't actually include the Opteron CPU. I mean, they named the platforms after the CPUs so I think they consider the CPU to be crucial to the platform as a whole.
I think PCIe is pretty much being used instead of HTX. When CSI is released, there will be three different serial protocols in use, HT, PCIe, and CSI.
I don't consider a cpu part of a platform, neither do I consider any add in cards or sata controllers, I do think that a platform is designed for a cpu or a cpu archetecture for me a hardware platform is the basic design of the chipset, motherboard and its interconnects,
AMDs platform allows for more diversity because of its current cpu archetecture,also it allows for better scaling with 4+ cpus and over all for the future AMD's platform should preform faster when paired with a faster cpu, where as intels is more limited because of the archecture of the cpu that it will connect with(not saying this has any say for current desktop pcs though as desktops don't seem to stress it) but from what I understand AMDs 4 and 8 way opterons scale better because of that. I guess shortest word I see a platform as a possibility, that is based on the CPU archetecture
Keep in mind this is my Opinion and in no way fact =p
the problem is both of you are wrong and right, you are argueing that Intels definition of a platform is right, he is arguing that his is right, the problem is that there are many different types of platforms, and many differnet definitions for each, so in essense the word platform is too general to use when debating what it is, even the dictionary has about 10 different definitions for it, a train platform is not the same as an artilary platform is it
How can the AMD "platform" be more flexible because of the CPU when the CPU is not part of the platform? Either you include the CPU or you don't. If you don't include the CPU as part of the platform, then your platform doesn't have direct connect architecture because the logic required to do this is located within the CPU. Nor does your platform have a memory controller.
And yes, AMD's platforms do scale better in part because of technology found IN THE CPU. And it's not difficult to find a definition for computing platform. Google "platform definition." It's there. You don't even have to take Intel's or AMD's word on it. This is despite the fact that Intel and AMD basically get to decide what the word actually means, considering they actually make the darn things.
The HTT controller is not the CPU.... It doesnt process anything.... It is however the platform device that makes the CPU work... Location doesnt matter...
Digital Viper-X-
[H]F Junkie
- Joined
- Dec 9, 2000
- Messages
- 15,123
Maybe you misread what I wrote, pretty sure I said I believe a platform is designed FOR a cpu archececture. So the cpu is not tecnically part of the platform, but the platform is designed around certain archectural designs of the cpu.
But the HT logic is located in the CPU.
http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2939&p=2
You can see 4 HT links there, right in the die. And sandwiched inbetween the 4 cores? The northbridge. Yes, without the CPUs there is no direct connect architecture. There is no platform. There are open traces on a motherboard, that is it.
Not to mention, direct connect architecture also refers to NUMA. Where do you think information is stored as to what data is located in which memory bank? That's right, the CPU. With no CPU, this is no NUMA.
Just face it, AMD's platform DEPENDS on their CPU. It is a crucial part of their design. Without the CPU installed, there is NO direct connect.
I feel like this would be much easier for you to understand if every part of the system were socketed and upgradeable. Would the USB controller be a part of the system if it were socketed and could be removed and replaced? Or would it now just be a "component?"
You know, back in the day, when people said "platform" they were referring to the ISA used by the processor. In those days, the CPU *was* the platform and everything else was just a component. Then x86 came to completely dominate the desktop market and Intel and AMD are forced to compete with each other's computing platforms. They've redefined the word platform and I must agree with their definition. It's their word to define.
I guess you could look at it that way. It is a valid interpretation. But in my view, it is the supporting devices that make up a platform....
For example IBM decided to make a PPC chip with hypertransport controller, it works in AMD's existing platform without any modification.... Apple uses this configuration in their older PPC products..... So the platform is architecture agnostic. It just simply requires the appropriate platform devices...
For example IBM decided to make a PPC chip with hypertransport controller, it works in AMD's existing platform without any modification.... Apple uses this configuration in their older PPC products..... So the platform is architecture agnostic. It just simply requires the appropriate platform devices...
And the northbridge consists of the crossbar switch, and the SRQ controller, and not much else... Connected to that are independant CP[U's, the memory controller, and the HTT controllers....
All of these devices are functionally independant, and only require the crossbar switch for communication.
No, it was one giant run-on sentence but I'm pretty sure I read it correctly.
Digital Viper-X- said:
Right there, first line. Platform is more diverse because of CPU architecture. You do realize that DCA is only possible because of transistors found in the CPU, correct? And you also realize that by not including the CPU in your definition of platform that you are in contradiction with both Intel and AMD? You know, those two companies that make platforms.
In this case, hypertransport is supported in the CPU and "northbridge" of the AMD chipset. Without HT logic in both components, there is no platform. There is an empty motherboard. HT is a point to point protocol, not point to empty socket. Furthermore, all the addressing needed for P2P CPU to CPU communication is done in the CPU. A little hard to do that without one.
No, IBM would have to make a chip that was socket and protocol compatible. That's all. When IBM plugs in the chip, then the platform consists of an IBM processor and AMD chipset. The performance will largely be determined by the speed of the IBM processor.
It's not about performance, it's about platform....
And the fact is IBM has a PPC chip that works on AMD's existing platform.... Platform is architecture agnostic...
That's wrong. The platform is highly dependent on the CPU used. The CPU will determine what applications can be run and directly impacts the performance of the platform. It is also required for any of the other controllers to work. Platform entails all of those devices that allows software to run. You don't even have to take my word on it... you can look it up yourself (I think).
In this case, the CPU would have a huge impact as it's ISA is PPC and not x86. They are not the same platform when you drop in an IBM chip. They are two different platforms that use the same motherboard.
I did not consider my Pentium and K6-2 chips to be a part of the same platform even though you could use both chips in the same motherboard. They even used the same ISA, but the Pentium was clearly faster when you wanted to do anything involving an FPU.
BTW, your last line "Platform is architecture agnostic" is pretty much the opposite of every platform definition that I've seen. Even AMD touts their AMD64 platform. What is AMD64? An ISA. I really wish you would try to look this up. At least make an effort.
Digital Viper-X-
[H]F Junkie
- Joined
- Dec 9, 2000
- Messages
- 15,123
RadiationMan
n00b
- Joined
- Nov 21, 2006
- Messages
- 38
So what you are saying Digital.. Is that, AMD and Intel both have an equal platform then?
you don't like to read the whole post
First paragraph!
Yes, you describe a motherboard. You are describing the PCB and chipset. That's a motherboard. And for the most part, chipsets are not designed for a certain architecture. The chipset includes the southbridge and many of the same southbridges are used in both Intel and AMD systems. In AMD chipsets, the same interconnect is used between the north and south bridges in both Intel and AMD systems (PCIe). The only difference, in fact, is that the AMD north bridge doesn't have a memory controller and communicates with the CPU via HT. Otherwise, they are quite similar.
AMD wanted a protocol that was P2P and serial that it could use for communication between CPUs (!) and the CPU I/O. They, for the most part, created HT. HT only works in this capacity if a CPU is installed. It's not like the CPU sits on the motherboard just receiving information from the chipset. It interacts. Without either the chipset or the CPU, the system is neither a platform nor a functional anything. Have you ever watched your CPU usage when accessing your hard drive, transferring a large file over the network, accessing huge blocks of RAM? You'll see the CPU usage go up because all of those controllers require the CPU to load and retire commands for them to work.
The platform you describe was designed with and for a particular CPU. That's because the CPU is part of the platform. If the two were totally independent, they would be designed separately, which they are clearly not.
Digital Viper-X-
[H]F Junkie
- Joined
- Dec 9, 2000
- Messages
- 15,123
where do you get that from my post
? and bigDH, you can interp it as you would like. But my opinion stays the same, the CPU influences the design of the platform, when has it ever happened that and old motherboard can support a next gen cpu? Never, it always ended up being "backwards" compatible.
I think this is the first time it has happened,with AM2 being out before K10 and being able to run K10 and K8 cpus. thats why its versatile, now you have 2 different Generations of AMD cpus, the K8 and the K10, the K8 has been out for a while, aswell as the K10 not being out yet, but these year old motherboards will be able to run a K10 based cpu,
thats why the platform is more "versatile", thats also why I don't believe a cpu is part of the platform.
Usually we see this
new cpu core / generation = new motherboard and chipset, basically a new "platform" is required for this cpu to run
with AMDs K10 at least they can design the CPU for the platform, as they idealy already have a good platform to work with, that is scalable, and has enough performance not to hold the cpu back.
I guess that makes this point a bit worthless
?where do you get that from my post
and bigDH, you can interp it as you would like. But my opinion stays the same, the CPU influences the design of the platform, when has it ever happened that and old motherboard can support a next gen cpu? Never, it always ended up being "backwards" compatible.
I think this is the first time it has happened,with AM2 being out before K10 and being able to run K10 and K8 cpus. thats why its versatile, now you have 2 different Generations of AMD cpus, the K8 and the K10, the K8 has been out for a while, aswell as the K10 not being out yet, but these year old motherboards will be able to run a K10 based cpu,
thats why the platform is more "versatile", thats also why I don't believe a cpu is part of the platform.
Usually we see this
new cpu core / generation = new motherboard and chipset, basically a new "platform" is required for this cpu to run
with AMDs K10 at least they can design the CPU for the platform, as they idealy already have a good platform to work with, that is scalable, and has enough performance not to hold the cpu back.
I guess that makes this point a bit worthless
Actually, the FSB has been utilized by Intel for quite awhile now. In fact, there are boards out there that can run both P4s and Core 2s. So I guess Intel's going to beat Barcelona to that punch. Here is the CPU list for my motherboard...
http://www.gigabyte.com.tw/Support/Motherboard/CPUSupport_Model.aspx?ClassValue=Motherboard&ProductID=2424&ProductName=GA-965P-DS3
Yep, my motherboard supports Celerons through Core 2 Quad. It can do this because the FSB didn't change between P4 and Core 2. And don't talk to me about the longevity of AMD "platforms." How soon you forget that AMD killed desktop socket 940, 754, and 939 all of which used HT.
And sure, you will (at least in theory) be able to plug in Barcelona into an AM2 motherboard but you won't get HT3 which is included in the Barcelona. To get HT3, you'll need a new motherboard.
You can keep this opinion as long as you want but it doesn't change the fact that it is wrong. I've seen people who still don't believe the holocaust ever took place, it doesn't make their opinion right.
BTW, the Hypertransport Technology Consortium owns hypertransport. Manufacturers that want to use HT have to get a royalty free license from HTC. So let me ask you this... If I buy a motherboard made by Abit (although manufacturing is outsourced), that includes a chipset made by nVidia then what platform am I running? nVidia licenses hypertransport tech from the HTC and make the chipset. The actual PCB and interconnects were made by Abit. Am I running an Abit platform, an nVidia platform, and HTC platform? Which is it? If the CPU has no impact on platform, then what platform would I be running in this case?
I think you keep getting tripped up on I/O interface and chipset. Like I said, you wouldn't have this opinion if chipsets were socketed and could be replaced.
All in all, it doesn't really matter. You're basically saying that AMD and Intel are not as wise as you when it comes to defining a platform. An arrogant stance IMO.
Digital Viper-X-
[H]F Junkie
- Joined
- Dec 9, 2000
- Messages
- 15,123
ok I take it back lets start over
first of all
You tell me what platform you are running?
so what platforms are these setups?
Intel Celeron D on a 975
Intel Core 2 Quad on a 975
Pentium D on a 975
core 2 on a 945
core 2 on a Via chipset
and Core 2 on an NVidia chipset
first of all
You tell me what platform you are running
?
so what platforms are these setups?
Intel Celeron D on a 975
Intel Core 2 Quad on a 975
Pentium D on a 975
core 2 on a 945
core 2 on a Via chipset
and Core 2 on an NVidia chipset
Nope, the same chipsets run both the P4 and Core 2. It's the VRMs on many motherboards that wouldn't work.
Wait, backwards compatibility is good when AMD does it but bad when Intel does it? Barcelona is backwards compatible with the motherboard you have now and it's good but if my motherboard is backwards compatible with P4 then it's bad? Duby stated that AMD has a better platform and architecture on the desktop. Both are incorrect. Platform includes the CPU and AMD has a poor competitor right now. The Athlon architecture is also inferior to Intel's.
Why would I be running an nVidia platform? The interconnect between AMD CPUs and the chipset is always hypertransport. Shouldn't it be the HTC's platform then? What about AMD chipsets that run Intel chips. The chipset is NOT the platform. You can go on and on but you'll still be contradicting the definition of the people that define the word. Hence, yours is incorrect.
Your definition of a platform is an OS... Which is true, but that is a software platform, and not a hardware platform....
And to be quite honest a OS is more an agency then it is a platform. An ISA does not = a platform... If that was the case it would be called a platform.... Instead it is called an Instruction set Architecture....
Architecture does not equal platform....
Nope, my definition of a hardware platform is the hardware required to run an OS. You're not running an OS without a CPU. You're a linux fan right? These linux guys agree with me:
http://www.bellevuelinux.org/platform.html
No I think the point he is trying to make is that Intels platform runs two seperate architectures, therefore how could it possibly be architecture dependant? I didnt read anywhere he said it was bad... Just that the facts contradict what you believe.
No the platform is the hardware required to make a components function together..... You need a HTT controller to make the CPU work... It is a platform device. You need a SATA controller to make the harddrive work.... It is platform device....
Your definition of a platform is a completely built system, OS, and support, and marketing, and all.... Which is nonsense.. Maybe you should look up what an agency is...
Or maybe your definition is the ISA? Maybe your definition is the complete system? If so you'ds be wrong again... We already have terminology vailable that better suits the need... Platform isnt it...
440BX Slot 1 supported Slot PIIs, Slot PIIIs as well as Socket 370 PIIIs with a slocket converter.
Super socket 7 boards often supported both AMD K6-* and Via CPUs, and in the earlier days of Socket 7 isn't wasn't uncommon to see boards support K6, Via and Pentium MMX chips.
Any number of 945/955 based boards support Core 2 Duo and Core 2 Quad despite being released for Pentium D.
I'm sure there are other examples but I won't get into semantics over what is a 'next gen chip' vs what is a core refresh. (I'd put Barcelona squarely in the latter, but whatever).
I think you really point out the greatest downside to AMD's design, every change to the CPU necessitates a change to entire system. Single to dual channel ram, DDR1 to DDR2, Really the only reason Barcelona et al are backwards compatible is because AMD basically made all the improvements in the system interface for the new chips optional. AM2 doesn't support Agena and Kuma so much as the new chips were designed to work with existing boards.
ok I take it back lets start over
first of all
You tell me what platform you are running
so what platforms are these setups?
Intel Celeron D on a 975
Intel Core 2 Quad on a 975
Pentium D on a 975
core 2 on a 945
core 2 on a Via chipset
and Core 2 on an NVidia chipset
You should definitely click on that link that you posted. You can go and click on their .pdf and read about the nVidia business platform. Of course, they include the CPU in their platform.
NVIDIA Business Platform
2007
NVIDIA Quadro NVS 210S GPU
NVIDIA nForce 430 MCP
NVIDIA MediaShield Storage
NVIDIA Gigabit Ethernet
AMD CSIP Processor (Socket AM2)
ATX form-factor motherboard
1 16x PCI Express® slot
2 1x PCI Express slots
4 PCI slots
4 USB ports (front), 4 USB ports (back)
PS2/Parallel ports
High-Definition Audio
DVI and VGA display
TV Out
Hmm... I see the processor mentioned. You claimed that the CPU wasn't part of the platform. nVidia feels otherwise.
My platform is mixed. It includes an nVidia Video Card, Western Digital Hard Drive, Intel Chipset, Intel CPU, Marvell ethernet controller, etc
You?
Digital Viper-X-
[H]F Junkie
- Joined
- Dec 9, 2000
- Messages
- 15,123
re-read my post
I give up on the platform thing, theres no way to argue it because everyone will have a different definition
I never said or didn't mean to say backwards compability is BAD or GOOD when either company does it, I tried to say AMDs is FUTURE compatible because of the basic archecture of their CPUs , where as eventually as desktops get into quad and more cores they will need a new platform and cpu redesign
about a better archectecture, which I still believe AMD has, simply because it's modular, but I'm a programmer so thats possibly why I think that way ;P
as to your most recent post last for me on platforms!
I Would consider my hardware platform an Intel 965p platform, everything else is added on later, what you posted there is a BUISNESS PLATFORM, platform being something that everything else i based from, you can add a WD hd, LINUX, OS2, NOVELL w/e the hell you want on it, but it will still be an Nvidia BUISNESS platform, just like you can add what ever you want to a chipset / motherboard but it will still be Chipset X platform