800Mhz (Bus) Processor + 667Mhz RAM = 667Mhz Total Processing Speed?

I'm wondering if a computer with a 2.0GHz/800Mhz FSB processor will run just as fast as a computer with a 2.0GHz/667Mhz FSB processor because the RAM's speed is 667Mhz in both.

Is it true?

 

Nope.

RAM will be the bottle neck but if RAM is run in Dual channel the increase in bandwidth should be close to what the CPU can use.

 

Even 667mhz ram without dual channel wont bottleneck the FSB.

667mhz DDR2 has a higher bandwidth then a 800mhz FSB. Plus there is more being send on the FSB then just Ram read and writes.

 

Thanks! You guys are awesome.

 

I don't see how, a 800FSB CPU has ~6.4gigs of bandwidth in theory. 667DDR2 has 5.4gigs.

 

So the intel systems of today are only about 8 years behind the DEC alpha? They're closing the gap! Too bad hp decided to buy Compaq and kill the alpha line.

* The DP264 (and DS20) from 1998/1999 had 5.2 GB/s of memory bandwidth in dual-channel mode.

 

Well 8 years ago that was considered a super computer. Technology has come a long way. A desktop was what? A P2-300 64megs and 6-10gig HDD? That was the screamer for a desktop. My iPod and cell phone has more power than that computer now.

 

No, that was not considered a super-computer. The DP264 and DS20 were desktop computers with one hard drive and a CDROM. (see attached photo). It was a large desktop case, like some of the enthusiast tower cases today, but a desktop none the less. Also, it could take 4GB of memory (and actually use all 4 )

The super-computer models were the GS60/140, replaced by the GS80/160/320 line. Those were large refrigerator-sized machines. Internal code-name for those was 'Wildfire'. They had memory bandwidth exceeding 60 GB/s.

Between the DS and GS line was the ES. ES40/45 came in two configurations, desk-side for a BIG workstation, or rack-mount as a server. ES45 internal codename was 'Privateer'.

I should know, I used to work for Compaq servicing their Alpha product line.

For many years, the Alpha processor was so far ahead of intel processors, it was just silly. Don't forget that all Alphas, from the very first one ever made in 1992, through the current models, are all 100% 64-bit running a 100% 64-bit OS. I guess they're still ahead of intel in some ways.

 

Here's some specs from the last model of GS class super-computer:

CPU: 2 to 64 Alpha processors; EV7z ( EV7 @ 1.3 GHz) and EV7 1.15 GHz
Cache: 1.75 MB ECC on-chip L2 cache, 7-way set associative
Memory: Up to 512 GB (8 GB/processor); RAID memory optional
I/O Expansion: Up to 512 PCI-X slots, 192 PCI slots, 64 AGP 4X slots
I/O Bandwidth: Up to 222 GB/s I/O bandwidth

(from here: http://h18002.www1.hp.com/alphaserver/gs1280/ )

 

Shows how much I know. What was the price tag on those? Weren't they used in research only?

Back in 98 I was 10 years old and I only knew how to use word and tinker in with Regedit now I'm an admin at a law firm. How times have changed indeed.

In my opinion Intel's weakest link is always in it's memory access. The bus in the P3 trails all the way to the PPro doesn't scale well. The P4's inital design was smart but it came from RAMBUS(problems of it's own) and had HUGE latency. AMD's EV6 bus licensed from Alpha was a huge step forward. Even now with the C2D Intel relies heavily on it's L2 Cache and prefetch to make up for it's huge latency and lack of bandwidth compared to AMD.

 

Well, they could run Tru64 UNIX, OpenVMS, and Linux. So you could use one for whatever you wanted. Browse the web, check email, whatever. id Software even released DooM and also Quake for Linux on the Alpha! But people didn't buy them for these reasons... The DS20 was around $8k for a basic model and as much as $15k for a well endowed version. ES series cost a lot more. GS series cost a WHOLE lot more (like six figure price tag!)

The Alpha was always a lot faster than intel at integer math, but it was WAAAAY faster than intel at floating point math. That made it the ideal platform for 3D graphics (the movie 'Titanic' was rendered on a farm of Alpha's running Linux) and just about any scientific application (modeling and simulations for pharmaceutical industry, weather prediction, and nuclear blast simulations, for example). Serious computing.

Quite true. Not just memory bandwidth, but all I/O bandwidth. The intel chipsets just couldn't deliver much bandwidth to/from the CPU.

intel memory bandwidth was limited to a single puny 64-bit memory path. Even today, the dual-channel DDR is just two 64-bit memory paths. The DS20 from 8 years ago used two 256-bit memory paths in dual-channel mode. So even though it used the same speed of memory (66 Mhz or 83 Mhz) as the PC's of that time, it had a MUCH bigger pipe to access that memory. The alpha also had a big advantage with the memory size. The largest Alpha servers in 1998 could use 128 GB of system RAM! The largest intel servers at that time couldn't even use 4 GB. Even today, 32 GB is the most you'll find in an intel server (some 8-CPU intel servers can take 32 GB) while the GS1280 Alphaserver can take 512 GB of system RAM!

Most desktop PC's had just one PCI bus, with servers having 2 or 3 PCI buses. The largest alpha servers had well over one hundred PCI busses! If you figure that the I/O bandwidth of 64 bit, 66 Mhz PCI bus (The fastest available in 1998) is 533 MB/s. That *3 is 1.6 GB/s of I/O bandwidth for your typical intel server. The largest Alphaserver in 1998 with the same 64bit/66Mhz busses had 128 PCI busses! 533 MB/s * 128 = 68 GB/s of I/O bandwidth.

So the advantage with Alpha was not only that the processors were waaaay faster than the intel's of the time, but that they could keep those processors fed with massive bandwidth. Think about that for a second - in the Santa Rosa laptops that we're discussing here, they have 5.4 GB of memory bandwidth. That's the bandwidth between the system RAM and the CPU. That's fine and well if your entire data set fits into system RAM. Well what happens if the data set you're working on doesn't fit inside the system RAM? What if your data set is 800 GB? (obviously not any current game!) You need some huge I/O bandwidth to get that 800 GB worth of data to/from the CPU for processing. That's where the PCI bandwidth comes into play, and why the Alphas ruled the world of large data processing.

That's why it's such a shame that hp killed off the alpha line in favor of the Itanium. Alpha is superior to Itanium, even today, yet they killed it anyways because Itanium was designed jointly between hp and intel (most people don't know that) and hp had sunk toooo much money into co-developing the Itanium that they *had* to ensure its success. The only way to do that was to buy its superior competitor and then kill it off (they sold the alpha development division and all the alpha patents back to intel). intel then took that and incorporated some Alpha-esque technology into Itanium2, and sells the Itanium chips right back to hp for use in their current line of Tru64 and OpenVMS servers. It's one big 'WTF'?

Back to the subject of Santa Rosa (to try and keep things on topic here ), the CPU has 800 Mhz FSB but the memory bandwidth is 'only' 667 Mhz. Does that mean that the extra 133 Mhz of FSB speed is wasted? No! The CPU talks to the large 4MB of L2 cache at 800 Mhz, and talks to the main memory at 667 Mhz. So whatever code the CPU is executing out of L2 cache has the benefit of the faster FSB speed.