I just wanted to drop a quick comment on the 4X4 vs Kentsfield questions I’ve been receiving.
I believe Kentsfield is to Conroe as FX-60+ was to FX-57. Kentsfield is a drop in replacement to Conroe, and judging by the performance we’ve seen I suspect that all of our Intel liquid cooled machines will migrate to Intel Quad Core as they become available. Intel Kentsfield is a damn good processor, and considering the power requirements are similar to a Pentium 4 I’d say it’s going to be a major player in the enthusiast space. Unlike the Pentium 4, Kentsfield isn't "sell proof" therefore there is no reason for Intel to start a price war on such a processor: Kentsfield will sell itself.
In order for 4X4 to be successful AMD has to somehow find applications that take full advantage of NUMA compliance and massive multithreading (because AMD has very good multi-threading potential). When I said “AMD may create a new category of enthusiast”, this is true. This is not your average gamer, I believe the person buying 4X4 is going to be someone who does tons of content creation and plays games. 4X4 has the potential of being an independent platform with very unique configurations. I do not for one minute believe that AMD is positioning 4X4 to be a volume play.
Until today no one has tried to create a true enthusiast level motherboard with no ECC and unbuffered memory, overclockability and all the rest under a dual processor configuration (with multiple core upgradability). This is what 4X4 is supposed to be – let’s wait and see what happens.
Last thing, I don’t publish comments from competitors pretending to be anonymous. Don’t waste your time, you know who you are.
26 blogger comments:
Call me uneducated if you will, but multi-core processing has never made sense to me.
Why can't the OS and one CPU devise a way to spread out the data streams into 2, 4, or 8 separate streams to take advantage of 8 CPU's? Even if the load is heavier on the first CPU, a single application could take advantage of having 8 cores working hand in hand at the same time.
I've never understood why a multicore processor only really performs during multitasking. Why can't 2, 4, or 8 cores work on say, World of Warcraft at one time?
For instance, one core handles processing half a screen of data, one core handles the other half of the screen data, one core handles audio and in game calculations, and the last core handles synchronization and network broadcasting (or something like this). The OS would naturally decide what operations needed what sort of processing, etc...
Seems a waste to me that a desktop system would have multiple processors that only really benefit the user when multitasking. Or am I missing something?
I´m not expert but I presume it´s game (or any other program) code. OSs are happy with more cores but they can´t hack the code into pieces. Data is streams, so when CPU need to execute A, B and C, it needs A before it can do B but C can be done on other core. Completely single threaded code needs A before B before C. Since I just upgraded to dual core I´ve been monitoring this. My 4200+ runs 60%/40% CPU load with World of Warcraft so WoW definetaly can multithread some degree.
Besides, multitasking is what we all do. We got OS running, anti-virus, firewall, few internet programs...
Why can't the OS and one CPU devise a way to spread out the data streams into 2, 4, or 8 separate streams to take advantage of 8 CPU's? Even if the load is heavier on the first CPU, a single application could take advantage of having 8 cores working hand in hand at the same time.
I've never understood why a multicore processor only really performs during multitasking. Why can't 2, 4, or 8 cores work on say, World of Warcraft at one time?
The way programs are written are mostly sequential in nature. (Think C/C++) There is only so much ILP (instruction level parallelism) available in that single thread of execution. Compilers as well as the hardware try very hard to extract as much of the ILP as possible and the best they can do is execute instructions so that independent operations occur at the same time. Due to the dataflow nature (one instruction's result feeds another) adding more and more processing resources for that single thread will not help as they will just be idle. Speculation can help but only so much. As far as I know efficient parallelizing compilers are also still a long way away.
For instance, one core handles processing half a screen of data, one core handles the other half of the screen data, one core handles audio and in game calculations, and the last core handles synchronization and network broadcasting (or something like this). The OS would naturally decide what operations needed what sort of processing, etc...
Yes, essentially the paradigm shift must come at the programming level. The programmer needs to think of multiple threads handling multiple data sets (data parallelism) or multiple functions (function parallelism) and arrange for proper synchronization among these threads. The algorithms must be parallel. The hardware sees things at a much finer granularity, it does not have scope over the whole program. And even if it did, there is too much intelligence involved to look at a sequential program and make it parallel using hardware; that is a programmer's job.
Choosing between 4x4 and Kentsfield is tough. I see pro's and con's to both. I like 4x4 in the fact that I start with 2 cores, then upgrade to 4, 6, then 8. That's cool. Is that verified? Also, the AMD quad CPU I think is a real quad core versus two dualies on a die. I figure wait for the real quad core.
"Kentsfield is a drop in replacement to Conroe," Are you sure, Intel hasn't has a drop in replacement for, well, forever!!
How do you think K8L will compare?
Why is 4x4 more restricted by threading then Kentsfield? Toms tests show almost no improvement/clock for Kentsfield except on a few synthetic tests.
4x4 4x4 WE WANT 4X4
4x4 would be my preference. I can definantly see the use for 2 processors, each with a dedicated bank of memory.
With AMD hardware virtualization, I would install a server, and dedicate one dual core proc and bank of ram to it, and use the other for gaming, and other stuff.
4x4 is my choice, especially with the opportunity to go to 8x4 (quad core chips)
Why can't the OS and one CPU devise a way to spread out the data streams into 2, 4, or 8 separate streams to take advantage of 8 CPU's? Even if the load is heavier on the first CPU, a single application could take advantage of having 8 cores working hand in hand at the same time.
This is actually the concept behind the Cell Processor in the PS3- one unit distributes work to 8 other units (and some of those are specialized for different tasks as well), but they're all on the same chip. The only problem is that games have to be written for this kind of processing, which is explained pretty well by everyone above me. Oh yeah and Cell yields are in the toilet.
Seems like if you had one dedicated CPU to compute how to distribute data to other CPU's it would kick butt. Leaving it up to the programmer is only a recipe for disaster, since they only care about how THEIR app performs. It needs to be in the hands of the OS.
Seems like the advent of multi-core processing will only benefit hand core multitaskers significantly, and the average Joe in a very little way unless the software is behind it. Does Vista lend extra support to multiple core processors in the desktop arena?
Seems like if you had one dedicated CPU to compute how to distribute data to other CPU's it would kick butt. Leaving it up to the programmer is only a recipe for disaster, since they only care about how THEIR app performs. It needs to be in the hands of the OS.
Are you talking about using the OS to multi-thread a single-threaded application? And how, good sir, may that be done? The OS does not "interpret" any information within the single thread of execution, it merely schedules it, provides OS services through syscalls, etc. The OS is not responsible for understanding the programming logic involved in the thread of execution and making it multi-threaded.
Programmers have had it relatively easy until now, just write straight line code that executes sequentially and trust the hardware guys to come up with the next processor that offers 2x speedup. Now that the focus seems to be more on multi-core rather than monster cores, multi-threading the software is definitely the way to go. Of course I am not ruling out hardware and compiler improvements that may help the programmer more easily multi-thread the program (e.g. transactional memory???). But the programmer must think parallel rather than serial which has traditionally been the case.
So in summarization..Kentsfield will sell itself, while 4x4 is for another category of computing, that depends on ?NUMA? and multithreading(I've only seen Kentsfield sell itself in threaded apps)
4x4 has potential to be 8 cores. If Intel feels that 4x4 is a threat, I believe they can ramp dual socket enthusiast solutions before AMD ramps K8L:)
What are your views on Kentsfield vs 4x4 performance? Also, how will you position these CPUs in your lineup?
I'm curious Rahul, why do you still sell AMD based computers? Back when Dell announced it's Quad SLI computer with a Pentium D you called it "a pig with lipstick on". Wouldn't running either Crossfire or SLI on an AMD based computer now be like "a pig with lipstick on" when compared with a Core 2 based PC?
Pigs with lipstick are hot:)
Intel taking the crown once doesn't mean a total reversal on VooDooPC's AMD support.
nice article. keep up the good work.
I'm curious Rahul, why do you still sell AMD based computers? Back when Dell announced it's Quad SLI computer with a Pentium D you called it "a pig with lipstick on". Wouldn't running either Crossfire or SLI on an AMD based computer now be like "a pig with lipstick on" when compared with a Core 2 based PC?
I would guess that what someone once pointed out on this blog is true. Rahul is ever so slightly (or more) biased towards AMD (for whatever reasons). But he will never come out openly, my guess is because he is in the business of selling computers from both AMD and Intel. Of course AMD has gained a lot of credibility in the past 2-3 years, but its hard not to miss all the constant optimism Rahul portrays about AMD. And then he will throw a little Intel praise around so as to not look biased.
Personally I think his blog is usually pretty ok, which is why I come here, but its obvious that he uses it (and smartly and obviously so) to further his own business. I do feel bad about all the non-techie people who come here and listen to the talk like its gospel though, I wish people would do their own research more.
P.S. Rahul, I am not a competitor, and I am sure you can figure that out.
"why do we still use AMD?"
Here's the simple answer. AMD is not exactly in the stone age, whereas prior to Conroe Intel WAS in the stone age (on desktop/server/worktation). That being said, AMD is behind in performance, but the cost of their CPU's is favorable to us in many configurations. In some cases a person can save upwards of $500+ for a high end machine that still delivers decent performance.
We are still selling AMD because they offer a decent value in many cases. Although my personal preference is Intel Core 2 Duo - I don't care about the price as much as performance, and most of our sales team feels the same way.
One quick note, AMD has *never* paid me or Voodoo a dime for this blog. In fact AMD doesn't give us any preferential marketing development funds either. Sorry to disappoint all of the people who think I get paid to form an honest opinion..
I agree particulary with that.
If the High End is defined by Conroe 6200 to 6800, it can not be denied, that AMD ones do share a broad range of this performance area. As well as a performance advantage of a 30 % in most cases is a intellectual property.
If it wasn't Intel should not have saled any cpu in the last years. Nor will we say this about AMD in a few years.
You were the first who said clearly "there are now 2, love it or not".
I see the market halfed now, at least in the middle of next year.
My personal experience with my 6th amd cpu (x2 3.8) under _x64 is great.
I had that much resets daily in the Pentium 166 days, as i had in the last 2 years.
If you ever weared Levis, you might do in future too.
regards jk
Ps: There are many out there, stockholders for example, who are pleased with your professional outlook.
Rahul, I honestly think that the only person disappointed by your neutrality is sharikou. :P
If Intel feels that 4x4 is a threat, I believe they can ramp dual socket enthusiast solutions before AMD ramps K8L:)
The Apple Mac Pro currently supports a pair of dual core processors and will support a pair of 4 core processors once Intel releases them.
Unfortunately Intel doesn’t have a desktop chipset that support a pair Core 2 Duo chips, so Apple is using a workstation chipset, which requires FBDIMMS and a Xeon (Woodcrest) processor. For Intel to make a compelling 4 x 4 product, they are going to have to either tweak the workstation chipset to support standard desktop DDR2 memory or create something new.
One thing I don’t get is why is AMD calling it 4 x 4. Should it not be 2 x 2 (two processors x two core) then 2 x 4 which would probably by followed by 2 x 8.
I was leaning to get a Apple workstation with processors running at 3 ghz but I decided not to when I read the review on those machines and the huge latency that is being incurred because of the usage of FB-DIMMS. ALso there is no RAID support in hardware and also there is no SCSI support let alone SCSI RAID.
Rahul,
Given recent reports that 4x4 will be on socket F, do you still believe it will use unbuffered memory?
Congratulations Rahul. Hope this turns out to be just as great news for hardcore gamers as Voodoopc and HP. Maybe we'll see drop in prices for the top of the line Voodoo systems now huh?! ;)
Personal preference for me is the 4x4 system. I was planning on upgrading to a dual processor board anyway for my CAD, Solidworks, and CAM work, so I could do all of that at once while listening to music, or doing multimedia work in the background.
To me, kentsfield is the exact same thing as 4x4, since you have two separate processors, and the tdp and voltage requirements double from those of their respective dual core variants.
Since I would also like a quiet machine, spreading out the thermal requirements across two processors makes more sense to me. Dual memory banks should also mean amazing memory performance.
However, the downside should be that if one application that is single threaded needs 2 gigs (i.e. battlefield 2), the whole system will most likely need 4 gigs of ram (since both procs have to have the same amount of memory). Maybe amd can find a way around this, but for now, that is the one theoretical downside.
Intel is totaly wrong and much too late, Conroe etc. are nice chips compared to the good ol' K8 chips.
But it'll end up to be just like Northwood back in the days. AMD will strike back hard and put intel 3 years behind again. Why? Because the processor it self is not the most important factor anymore. AMD has proven that.
No matter how great Kentsfield is, it will lack the great options AMD's platform will offer in 2007 and beyond. Imagine using your GPU and other custom chips for extra processing power. That would be something wouldn't it? And that's what AMD has in mind, maybe less raw processor power but overal excellent processing power, by using custom chips. ATI already has optimized software ( F@Home ) that can take advantage of the GPU.
So think about it. Intel is late, they have to do something like Torrenza! And sooon!
Intel's roadmap is much better than what they had during Northwood times. They will be almost one process cycle ahead of AMD soon. With alternating process switch and architecture change for next 4 years 45nm/Nehalem/32nm/Gesher, there is no way on earth that AMD is going to catch them on wrong foot. Also with Nehalem Intel will implement CSI and IMC, any advantages that AMD have till then would evaporate.
Post a Comment