Where did INtels leads go in Multimedia???

[Duvie]

I was wondering how much has AMD really come in these apps....NOw I have always speculated the quick fix design of the PD and PD EE was merely a band-aid and some of its inefficiences like still using the 800fsb and a shared memory bus were going to hamper it and lead to less increases in the dual core versus same single core....

I for one would have liked to see the so-called "reviewers" look into this. Merely by running a review with a Prescott 3.2e 1mb cache (540) and a FX51 or Clawhammer 3500+ for the 2.2ghz 1mb of L2 cache....

Then we could look at increases over single core and see who was more efficient...

Instead we wil have to lool at previous reviews prior to X2 and see what those chips above related to one another then.....

I reviewed Anandtech, techReport, Xbit, and Tomshardware to ry to get some numbers...now it was harder with Toms as they felt the need to put in their chart most of their A64 chips on Via boards and have not upgraded to NF3 and NF4 (superior platforms)


Divx (5.2 encoding)

Intel P4 3.2e w/ HT previously had ------9% lead-----------now they are 4-7% behind

The PD EE version did worse then the non EE one showing the inefficiencies of HT and handling more then 2 threads. May be helped later on with software development I assume. Some apps do very well with 4 threads like rendering in 3dsmax, POVray SMPOV 3.6, Cinebench, etc....Other do not like most of the audio and video encoding...May be a reason why dual - dual core opteron setups may not be the best setup for this type of workstation activity at current...


Xvid (1.0-1.3 codecs)

Intel P4 3.2e w/ HT previously had ------4-10% lead-----------now they are 10-12% behind

The 4-10% were differnces in reviews and apps used. The asad thing was the EE did worse in all of these apps again showing the inefficiencies with HT trying to handle 4 threads versus 2 threads.


Windows Media Encoder 9 ( I believe it was the HD)

Intel P4 3.2e w/ HT previously had ------5%% lead-----------now they are 5-13% behind

The differences here were AT showed the PD EE do worse but TechReport had the EE do significantly etter then the non EE but still not better then the X2 (they ran it as what it truly was an OPteron 175)

Could ther have been other factors then just the trasformation to dual core??? yes...It is all possible SSE3 had a very small help in a few of those apps, but I reiterate very small if you look at winchester versus venice reviews....Also A64's have come aways from the FX51....

NOw in the end the leads have vanished as I had predicted (many also did) for weeks leading up to this event. These leads by AMD are very conservative now because they were tested on server class mainboards using slower ECC ram, slower timings, 2t, and overall slower chipsets meant for stability not high performance. Those leads will only get better....


So if we sum up the state of multimedia where it appears many of the apps are somewhat to very much adapted for SMT...AMD when they release will be faster then anything Intel has to offer and will be the king of pretty much all the benches. No more "if you encode this" and "if you game that" threads. I dont see any reason to promote an Intel system unless the person already had DDR2 and it may be a bunch of money saved to not swap out as many parts. Even then it will be tough cause to stay cheaper you would have to go with the even slower 2.8ghz model. Remember this is the 4400+ against their flagship dual core model annouced and AMD still has the 4800+ to be released at launch. The leads will only get bigger.

AMD will be the total solution since a 2.2-2.4ghz AMD even X2 is still a heavy performer in all gaming versus comparable priced INtel non dual core chips....

 

[Lithan]

Personally I think that venice's SSE3 is crap and most of the gains came from AMD actually getting SSE3 working right.

 

[Duvie]

Originally posted by: Lithan
Personally I think that venice's SSE3 is crap and most of the gains came from AMD actually getting SSE3 working right.

Please provide links cause it is hard to look at INtels and tell whether SSE3 helps there as well...So many things going on between the Prescott and Northwoods to see what is what...

In the end I think you are wrong. Do some research on the SSE3. PLus we are talking about 15% swings now (probably more later) and no SSE I have seen to date has ever accouted for that...SSE2 never did that much...Why would assume SSE3 would be the end all saviour for multimedia??? Ad why would you think they couldn't get Venice to work??? The did not use like 3 of the 14 instruction sets of SSE3 with the Venice.

Until you can prove SSE3 made this much of a difference I think it doesn't make much logical sense...

 

[Lonyo]

Well, Intel will still be good for lower end dual core systems assuming the motherboards aren't too expensive (given that DDR2 actually doesn't have much of a price premium over DDR from what I have seen, nowhere near enough to make up the difference). Total cost of ownership for an entry level PD will be less than TCO for any X2.

If you are looking at the higher end, there's now no reason to go Intel at all. But a PD 2.8GHz vs what, a 3500+ or 3800+ (not sure of prices) is a no brainer, the PD would be much better for encoding at that system price, while AMD would still be better for gaming.

 

[imported_michaelpatrick33]

You are now seeing why Intel is releasing their dualcore at such low prices. They pretty much lose everywhere in performance now. If the X2 4400 (two FX-51's basically with SSE3) was priced at 300-350 when it comes out Intel would be in trouble but since they are prices at 500+ Intel has the advantage of foundry and revenue to undercut AMD.

What people are forgetting when they talk about the expense of the X2 4400 at $550 is that it will destroy a 2800 nonhyperthreaded 2800 dualcore Intel. Obliterate it
AMD X2 4400+ is 2 FX-51's with SSE 3 and tweaks.
Intel 2800dualcore is 2 2800 non hyperthreaded Prescott.
Are you kidding me?
An FX-51 single core will destroy a 2800 prescott single core non-hyperthreaded in everything. Single threaded or multi-threaded. So, two FX-51's in an architecture designed for dualcore will also obliterate any dualcore Intel nonhyperthreaded solution. So Intel strikes back with price and foundry. AMD goes after server, workstation market.

It is funny that AMD has become the performance leader with higher prices and Intel has become the performance follower with lower prices.

 

[Duvie]

Originally posted by: Lonyo
Well, Intel will still be good for lower end dual core systems assuming the motherboards aren't too expensive (given that DDR2 actually doesn't have much of a price premium over DDR from what I have seen, nowhere near enough to make up the difference). Total cost of ownership for an entry level PD will be less than TCO for any X2.

If you are looking at the higher end, there's now no reason to go Intel at all. But a PD 2.8GHz vs what, a 3500+ or 3800+ (not sure of prices) is a no brainer, the PD would be much better for encoding at that system price, while AMD would still be better for gaming.

except for as much a beating INtel got now will only be increased by using a part giving up 15% in more clock speed...Doesn't seem really smart....


1) New Intel system will require New mobo, new DDR2, new cpu...does anyone know for sure if this is BTX and may require new case or new power supply???

Nonetheless...

2.8ghz (820 PD) = 241.00
MObo 955 = 150-200
DDR2 (1gb of decent ram) = 120-170 if you want 533or 667 (At seems to think 667 is a waste)

Total is 511-611.......

2) Now for a person getting a whole new A64 system

Athlon X2 4200+ = 534
Mobo NF3U-NF4 = 90-130
DDR1 (1gb of decent ram) = 100 (gskills cas 2-3-3-6)

Total is 724-764

3) Now for a person upgrading an sckt 939 system

Athlon X2 4200+ = 534

Total 534


4) Now for a person upgrading to Intel dual core from a dual P4 system using DDR2 already

PD 820 = 241
Mobo 955 = 150-200

Total = 391-441


1 versus 2 = a wash...Why??? cause the added cost of 25-41% is about close to the performance advantage the 4200+ will have over the PD 2.8...Multimedia wiould be now25+% using the scores from the srever board....

1 versus 3 = no brainer AMD users have a viable reaso to upgrade for huge leads....

2 versus 4 = Intel owners have a better route for sure...like I stated.....



This is all based on the threory MM would be the reaosn for the PC....If a person does pretty much anything else a 2.8ghz prescott really stinks in a lot of things.....


I also firmly believe we will hear about some lower cost model like a 2.0ghz one around launch time with an availability after the others....

 

[Lonyo]

$80 for some cheap DDR (OCZ) 1GB
$105 for cheap DDR2 (Corsair)

Same price difference as in your comparison.
But the motherboards? I haven't yet seen any prices on Intel 945/955 motherboards. You would probably be able to pick up a 945 for closer to $100, meaning a low end package could be ~$450 for a 2.8GHz PD, making it more attractive, especially for encoding on a budget, than the X2 setup which will almost certainly be $250 or around 50% more expensive.

Pentium D will still be useful for limited budget encoders (and I mean very limited), but not for anything else.
Some people can't magically pull extra money out of their pocket, so the added cost of an X2 would be too much for them.

 

[Duvie]

I agree the 945 may be cheaper but I went for decent cas 4 DDR2 I didn't lok for ultra cheap and every time you do that just start lopping off more performance percentage....

IN the end when we compare we dont get perfect percentage of performance increase now for cost...this is no worse then ppl getting 3500+'s over 3000+ winchesters in price performance...

I see budget minded ocers getting more excited about this then a non opcer who is basically paying 50% less for a 30+ percentage (maybe more with the crappier ram) decrease in performance...That is a big hit and as big as some of the top to bottom lines now actively being produced....


I can see your point, but I dont see myself recommending one (DDr2 is possibly a dead format from industry reports I linked the other day so yet again INtel users will be screwed again with a dead format) as much as I believe AMD has wiggle room in reduction of price by the time it starts arriving...I like the spread wont be as high as it is listed currently and any introduction of a lower speed 3800+ and 4000+ models would close if not fill this gap....Super low budget ppl willing to take a 30%+ hit for not wanting to pay 50% more likely dont even need the power of the dual cores anyways....



Lets get back to the point of the new MM leader and why.....

 

[CheesePoofs]

Why isn't it that AMD can't/won't pruduce a 2ghz Athlon 64 X2, or even a 1.8ghz one? They could use lower speed tranistors which equals higher yeilds which equals lower costs. Even at that speeds, I wouldn't be suprised if they could beat most of what intel had to offer, but at the same price as the intel parts.

Did Intel used to be so good at multimedia stuff before they came out with HT? THat seems to me to be the reason they had such a lead-intel could execute 2 threads, athlon 64's could only do one. Now that they can do 2 treads witht the dual core chips, that lead has completely vanished because the A64's main bottleneck has been removed.

 

[Lithan]

Compair Prescott to Northwood. Except SSE3 optimizations, I can't think of a better reason for Prescott to beat NW in media work (I doubt the cache boost helped... esp considering they also boosted cache latency). And Prescott has a longer pipeline than NW, so It's already making up for a SIGNIFICANT performance hit there.

 

[CheesePoofs]

One of the reasons prescott beats northwood in many apps is because prescott has a much better branch predictor than northwood does.

 

[Duvie]

Originally posted by: Lithan
Compair Prescott to Northwood. Except SSE3 optimizations, I can't think of a better reason for Prescott to beat NW in media work (I doubt the cache boost helped... esp considering they also boosted cache latency). And Prescott has a longer pipeline than NW, so It's already making up for a SIGNIFICANT performance hit there.

here is the problem...The spread between the 2.8's, 3.0s and 3.2's are all different...So something else is at work here and it may be some of the cache.....SSE3 does not make the big of a difference I am sure of that.....

 

[Zebo]

Nice work Duvie (X2 makes you excited much?)

And that was not an X2 but some crippled step child in these reviews...

anyway here was my prediction before any of these reviews came out from another thread:

All those MM leads Intel now has, bye bye!

 

[Duvie]

What do you think SSE3 the diff?? I dont think so....

 

[Zebo]

All I can think is SSE3 implementation is different betwen E4 and E3? Because in the Venice (E3) reviews SSE3 did'nt do squat. But I lean twards SSE3 is still not doing anything, but instead, we are seeing what two live real cores can do as opposed to HT, which is hardly a substitution for the real thing. AMD has the superior core, two of them makes gives them a multiplier of 2x superior.

 

[Zebo]

Why isn't it that AMD can't/won't pruduce a 2ghz Athlon 64 X2, or even a 1.8ghz one?

Of course they CAN. If they can produce 2.4's, 1.8 is a trivial affair. They will in fact, later, patience grasshopper. AMD always relases it's highend A64 parts first to lure the buyers with fat wallets.

 

[akugami]

My assumption was always that due to the P4's hyperthreading that allowed it to do better in tasks such as video and audio encoding. Now with the A64 X2's the AMD chips also have enabled the hyperthreading bits but of course, it's using two CPU's to process the instructions. This accounts for part of the boost. Programs written specifically for SMT would better benefit the faster CPU's (A64's) even though both dual core chips would benefit. This is why we're seeing the larger boost in performance by A64's over the PD EE's.

 

[Duvie]

This is all good, but how do we explain the fact that in the test the actual PD EE model lost out to the one without HT many of the times??? That is the question I have...Is it do to the fact the app is dual theaded but not truly multithreaded???

 

[sharad]

I thought that none (or hardly any) applications currently make use of the SSE3 registers. So SSE3 surely can't be the reason for the performance increase.

 

[Duvie]

Encoding apps (a handful), a patched superpi, is all I have ever heard of.....I dont see any links or data to support that soI dont buy it...

 

[CheesePoofs]

Originally posted by: Duvie
This is all good, but how do we explain the fact that in the test the actual PD EE model lost out to the one without HT many of the times??? That is the question I have...Is it do to the fact the app is dual theaded but not truly multithreaded???

I think its just because of how inefficient the Windows scheduler is. It sees four threads and just starts throwing all the processes it can at the CPU, which results in the CPU not being able to process all of those processes.

What would be interesting would be to see some of the tests where Hyperthreading hurt repeated on Linux.

 

[imported_g33k]

I reviewed Anandtech, techReport, Xbit, and Tomshardware to ry to get some numbers...now it was harder with Toms as they felt the need to put in their chart most of their A64 chips on Via boards and have not upgraded to NF3 and NF4 (superior platforms)

Is it just my observation, or do you think that Tom's hardware always tends to show Intel's procs in better light? I don't know it just seems that their benchmarks tend to show Intel beating AMD procs. Maybe it has something to do with the type of benchies and the hardware. But I'm tending to think that Tom's is somewhat biased towards intel.

 

[StrangerGuy]

BTW Anyone has tried multitasking Linux on a single-core CPU? What is the performance like?

I was under this impression that the Windows scheduler was to blame for poor multitasking performance on single-core CPUs.

 

[Duvie]

Originally posted by: g33k

I reviewed Anandtech, techReport, Xbit, and Tomshardware to ry to get some numbers...now it was harder with Toms as they felt the need to put in their chart most of their A64 chips on Via boards and have not upgraded to NF3 and NF4 (superior platforms)

Is it just my observation, or do you think that Tom's hardware always tends to show Intel's procs in better light? I don't know it just seems that their benchmarks tend to show Intel beating AMD procs. Maybe it has something to do with the type of benchies and the hardware. But I'm tending to think that Tom's is somewhat biased towards intel.

Nope your not the only one...

 

[Keysplayr]

Looks like if you don't mind spending that much cash for just a processor, AMD x2 is the way to go these days. Intel won't become interesting until the Prescott goes the way of the dodo. So don't look to intel for the next year to come out with anything decent compared to AMD.

There is one thing that most people overlook that I find annoying a little bit. When you see a benchmark of a AMD cpu and an Intel CPU, and the AMD graph is ahead of Intel, it is automatically assumed that the Intel chip is slow. Is it slower compared to the AMD? Yes. Is it slow? No. Just my pet peeve.

Don't look to Intel to become interesting until prescott is dead. Or at least the die is shrunk and an integrated memory controller is devised like AMD's solution.

Intel may have reached the point of diminishing returns. Who knows?