AMD and Intel's Next Generation CPU's Revealed

AMD and Intel's Next Generation CPU's Revealed
The hot topic lately is the next generation of Processors from AMD and Intel. The big question is "Will AMD be able to keep their foothold in the market?". The answer looks promising. With both AMD and Intel releasing new chips in the coming months, we at Apu's Hardware have put together a comparison between them.

Processor
Size (micron)
Core
L2 Cache
Voltages
Bus
Speeds
AMD
Mustang
.13
Mustang
1-4MB On Die
133Mhz DDR
1.4Ghz+
AMD
Palomino
.18-.13
Mustang
512K On Die
LOW
133Mhz DDR
AMD
Morgan
.13
Mustang
64-128K On Die
100Mhz DDR
900Mhz+
AMD
Sledgehammer
.13
Sledgehammer
2Ghz+
Intel
Itanium
.13
Itanium
733Mhz+
Intel
Foster
.13
Foster
Quad Pumped 100Mhz
Intel
Pentium 4
.13
Pentium 4
256K
Quad Pumped 100Mhz
1.4Ghz+


The missing blanks in this table are due to our lack of information, most likely because that information hasn't been released to the public, or we couldn't find it =).

As you can see, AMD is planning on hitting it big with their Mustang core. They are using the same core to power four different chips. AMD has a 12-stage pipeline, unlike the 10-stage with the current Athlon and Duron processors. AMD also put in more 3DNow! instructions for increased speed. AMD is shooting big in ALL markets this time. They are going after server, performance, midrange, and low-end markets. Another market they plan to get into is the mobile market with their Palomino chip. The current Athlons and Durons have too high of power consumption to be used in a mobile device, so AMD had to resort to their K6 line of chips to accomplish this task, which are undoubtedly slow.

Palomino Vs. P4

The Palomino is made to compete with Intel's Pentium 4 chip for the mainstream consumer market. The Pentium 4 is really a major jump for Intel. The P4 is the first mainstream chip Intel has made that doesn't use the P6 architecture. Instead it uses the newly developed the "NetBurst Architecture". Obviously Intel is betting on the NetBurst Architecture to outperform the Palomino and it's 266Mhz DDR bus. Intel has one-upped AMD and replaced their 10-stage pipeline with a 20-stage pipeline in the P4. A bigger pipeline isn't always better, though, since the processor isn't able to accomplish as much in a clock cycle with a bigger pipeline, but it also allows for faster clock speeds. We will have to see if this jump helps or hurts Intel. My guess is that Intel is going for paper specs (MHz for MHz) against AMD. While Intel might have a 1.8Ghz chip out when AMD has a 1.4Ghz chip, they might very well perform the same.

The Pentium 4 also has a Double-Pumped ALU (Arithmetic Logic Units). Basically the ALU calculates integers, and this will improve performance in that area. The P4's ALU running at twice clock speed makes up for a lot of the performance lost in going to a 20-stage pipeline. Whether it makes up for all the performance lost we won't know until the chip is officially released for benchmarking. AMD is also betting on the cache size of their Palomino chip. Intel opted for 8K of L1 cache to save money, and 256K of L2 cache, while Palomino has 512K of L2 cache. The difference, though is that Palomino retains the Athlons horrid 64-bit pathway, while the P4 will retain the P3's 256-bit pathway to the cache.

AMD has big plans

Intel has been hurting ever since the Athlon was released on its 200Mhz DDR bus, and they decided it wouldn't happen again. The P4 uses a quad-pumped 100Mhz bus, which effectively runs at 400Mhz compared to the Palomino's 266Mhz bus. The P4 will be released on the Tehama chipset, AKA the i850. The i850 won't allow dual processors, but AMD's 770 chipset will. The i850 uses dual channel RDRAM like the i840. Some see this as good, but I see it as bad. I'm overly zealous about RDRAM and I just don't think its a viable option for the current computer market due to its high latency, heat, and most of all PRICE! I think this move will ultimately hurt Intel considerably. AMD's 770 chipset on the other hand will use DDR SDRAM. I can see this as a much more viable alternative to fix the memory bottleneck in PC's today. DDR offers double the performance of current PC133 SDRAM for very close to the same price.

The fact that Intel has stated that the P4 will not be SMP and only uniprocessor kills it's chances for the hardcore SMP folks (you know who I'm talking to *cough*Celcho*cough*). Intel is planning to go SMP with the Foster chip. AMD, on the other hand, has a very interesting strategy for SMP. The 770 allows two processors per northbridge. This is very interesting, since you can daisy-chain the northbridges via LDT's Bus Technology. What this means is an unlimited amount of chips in a single machine. The only problem is that each processor requires it's own 64-bit bus, and this would be very hard to implement into the motherboard.

The instructions governing both chips have been greatly updated. Intel has adopted their SSE2 technology, which adds 144 new instructions to the currently used SSE technology. This boosts multimedia performance significantly. If software is written specifically to utilize SSE2 you can see significant improvements in encryption, video, speech, and content creation among others. Chances are that most applications won't be written (or rewritten) to use these new instructions, though, so don't get your hopes up. AMD also has updated their instructions. AMD has updated 3DNow in the same way Intel has updated SSE, and AMD has renamed 3DNow to 3DNow+. The Palomino will support both 3DNow and 3DNow plus, allowing full backwards compatibility. All the details for 3DNow+ have not been released yet, but expect some significant performance gains.

Intel also has big plans

VIA is also in the game. AMD and VIA have become almost like business partners. VIA basically saved AMD from near-death by their own Irongate 751 chipset. The Irongate was a far cry from the best in technology, and VIA came to the rescue with the KX133 and KT133 chipsets. Now almost all Athlon and Duron motherboards are made on these chipsets. AMD has given VIA the green light to create more chipsets for the upcoming AMD chips. VIA has already gotten to work on a DDR SDRAM platform slated for release as the KX266.

Intel on the other hand has shunned VIA. Since VIA P3 motherboard sales have been rising, Intel has gotten increasingly angered at VIA. To this date, VIA has not paid the licensing fees to create a Pentium 4 chipset, so it is unclear if VIA will play any role in the P4. I think this is a bad move by Intel to try and corner the chipset market. With no alternative to RDRAM Intel is really digging a hole for themselves. RAMBUS has also put patents on DDR SDRAM and expects to collect royalties on it. If RAMBUS has their way I think DDR SDRAM could be more expensive than RDRAM. Fortunately, several of the major RAM manufacturers (Micron, Toshiba) have filed lawsuits against RAMBUS for patent infringement. For the sake of the consumer I hope they win.

Here we have the P4 architecture

VIA is also planning to get into the integrated video value market. VIA plans on shrinking their die sizes to .18 or even .15 micron and incorporate the S3 Savage4 video into their value chipsets. The Savage4 is considerably faster than Intel's integrated video chip and it looks like VIA will be able to win over the value market in terms of graphical speeds.

Without all of the information, it is extremely hard to draw conclusions as to which chip will be able to perform. A big issue will be if Intel will be able to produce. As we have seen with the P3, Intel had extremely low yields in their high-end chips. AMD was exactly the opposite. AMD had extremely high yields on their high-end chips, which in turn, made low yields for their slower chips. Price will also play an important factor, as AMD's Athlon and Duron chips are currently much cheaper (Mhz for Mhz) than Intel's. If Intel can increase its yields then expect to see much cheaper chips from Intel.

Sledgehammer Vs Itanium

The battle for the high-end server market has always gone to Intel. AMD always planned to make a server chip out of their Athlon chip, but never got around it. Intel, on the other hand, has always had their Xeon line of server processors. On the brink of the next generation of chips, AMD has decided to play it's cards right from the start and begin development of a server chip. The big deal is that both chips are 64-bit compatible. 64-bit chips are able to address much more memory and do more in a single clock cycle. The differences between how they deal with 64-bit is the most important part.

The Sledgehammer's specifications were released by AMD two and a half weeks ago. AMD has announced the chip will be 64-bit compatible. It goes about this in a very interesting way. Over ten years ago, Intel moved from 16-bit to 32-bit with the x86-32. This chip was fully backwards compatible with the current 16-bit applications and had no performance loss. AMD has done the same thing Intel did ten years ago today. AMD announced the x86-64 specifications for their Sledgehammer chip. Basically the Sledgehammer retains all current 32-bit compatibility while being able to perform 64-bit functions with no performance loss. This allows the industry to ease into the 64-bit craze instead of having to convert every single application to 64-bit.

The operating modes table for the Sledgehammer

Intel also released the specifications for their Itanium processor. The Itanium, like the Sledgehammer, will be 64-bit compatible. The Itanium (or IA) will shred the x86 architecture that has been cast around processors for the last twenty years. It creates an entirely new architecture, entitled IA-64. This, unlike the Sledgehammer, will not have any backwards compatibility with 32-bit applications. If big businesses decide to switch over to the IA-64 architecture, the Sledgehammer will be left behind. More than the chip itself, this is what will hurt the Itanium in my opinion.

SMP is also very important in the server market. The Itanium, will, of course be SMP, along with the Sledgehammer. What's interesting is again how they accomplish this. AMD has already stated that they will be implementing multiple cores on the same die. This, in effect, is like two processors in one. I'm not sure how they will design this since they haven't released the specifications for it yet, but we do know it will be coming. We don't know much about the Itanium's SMP ability, other than that it is there. The Sledgehammer on the other hand we know will have multiple cores on the same die, and motherboard manufacturers will be able to daisy-chain the northbridge to allow for an unlimited amount of processors.

When you look at it from a business point of view, it starts to make sense. You have millions of dollars of computer equipment, and you want to upgrade it. Do you A) Spend twice what you paid originally to upgrade to the COSTLY Itanium chip or B) Upgrade what you need with the moderately priced Sledgehammer. Keep in mind, you would have to retrain all programmers to program for the Itanium, and have them reprogram every piece of software to use with the Itanium. Considering Intel wants you to use the extremely costly RDRAM with the Itanium it is a never ending hole of money waiting to be lost in the maze. Already we are seeing problems with the Itanium. Intel told us that the Itanium would debut at 733Mhz, instead of it's intended 800Mhz. This obviously indicates fabrication problems with the Itanium. It's not good when the first batch of chips you make has fabrication problems.

Here are the Sledgehammer x86-64 specs, the gray shows what is supported in 64-bit mode which, needless to say, is quite a bit more than it's 32-bit counterpart

If I sound biased toward AMD, I'm really trying not to be. The Itanium is really a new generation in super computing, but do we need it? I don't think we are ready for the profound leaps that Intel wants us to take with RAMBUS, and Itanium's IA-64. Remember, though, 64 bits is a far greater jump over 32 than just 2x. All computer integers are calculated by 2^x. In 32-bit's case this is 2^32, or 4,294,967,296 (about 4GB of addressable RAM). In 64-bit's case this is 2^64 or 18,446,744,073,709,551,616 (about 18 exabytes, or 4 times all words ever spoken by human beings). We won't need more than 64-bit for a LONG time, or ever.

Overall we can see that the race for the fastest CPU will continue far into the coming years. Both Intel and AMD have an excellent lineup which will bring us faster and more intense applications and games. It's funny how fast things can change in this industry, though. Two years ago Intel was considered the creme a la creme in the processor industry, until AMD came along and took a considerable share of the market. Who will make it big? The P4 or the Palomino? The Itanium or the Sledgehammer? We won't know for quite some time, but until then we can only speculate.
UYCnu 2012-2014 ©