AMD Barcelona in Action

Page 3 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Also MU...

Given most processor/OS scalability that I have seen as of late...

Things are getting pretty efficient so with either platform (speculation on my part) will probably scale at or about ~1.8:1.

This is just info gathered during scalability tests across multiple CPUs and OSes. (important to note that I did NOT test these specific processors)..

But I must say that scaling from a single to a second has yielded on many many an occasion nearly a 1.8:1 on numerous threaded applications (NOT GAMES).
 
Also MU...

Given most processor/OS scalability that I have seen as of late...

Things are getting pretty efficient so with either platform (speculation on my part) will probably scale at or about ~1.8:1.

This is just info gathered during scalability tests across multiple CPUs and OSes. (important to note that I did NOT test these specific processors)..

But I must say that scaling from a single to a second has yielded on many many an occasion nearly a 1.8:1 on numerous threaded applications (NOT GAMES).
Click to expand...
Yeah...but from single to dual is ~1.8....then dual-quad is 1.8x that IIRC...so 1.8 x 1.8= 3.24 x a single core? At least that's how i understood it. :?
 
I think for encoding a single video stream there is enough sequential dependence between threads that this factor of 1.8 would probably go down as you increase the number of cores. However, if you were encoding 8 separate video streams you could get probably closer to a factor of 2.

Also the encoding bitrate or the particular video sequence used and other hardware could have some impact on speed. A good software implementation would need to be optimized for a particular CPU architecture, not only because of the presence of different instructions like SSE4, but also to account for the number of clock cycles each particular instruction takes on a given micro architecture.
 
Sorry Tanker,

All bets are off going to the third/fourth core/processor.

The problem mostly being the OS overhead for managing this X number of cores.

The 1.8:1 I mentioned is going from single core/proc to dual core/proc.

Above that can have differing numbers.

DEC Alpha processors (old now) will scale very very well with OSF1. Even out to the third and fourth proc (SMP) they still scale at nearly 1.8:1.

I do not think Windows is that efficient. Linux maybe.

Scalability is in direct relationship to the Hardware, Software and OS you are running. Many of the tests I have run are server related.
 
I couldn't help but notice that the R600 card in the DP system has only 1 out of the 2 PCI Express power connectors plugged in, yet it seems to be working fine. More suspiciously, the only cord plugged into the R600 in the photo is a 6-pin PCI Express power connector and not the higher-amperage 8-pin connector. Is this a sign of power consumption of the cards or just error on the part of the system builder?
 
Sadly without benchmarks the only thing we really have to go on is the vague (until the NDAs are lifted) postings of people in the know. Below is a comment from Kyle Bennett of hardocp (posted today in the discussion of the demonstration) - would be curious if anyone has spotted anything similar elsewhere.

I feel very confident as to Barcelona's / Agena's ability compared to Core 2 and that is all I can say.
Click to expand...
 
Also MU...

Given most processor/OS scalability that I have seen as of late...

Things are getting pretty efficient so with either platform (speculation on my part) will probably scale at or about ~1.8:1.

This is just info gathered during scalability tests across multiple CPUs and OSes. (important to note that I did NOT test these specific processors)..

But I must say that scaling from a single to a second has yielded on many many an occasion nearly a 1.8:1 on numerous threaded applications (NOT GAMES).
Click to expand...
Yeah...but from single to dual is ~1.8....then dual-quad is 1.8x that IIRC...so 1.8 x 1.8= 3.24 x a single core? At least that's how i understood it. :?

Actually you would ADD not multply since you're adding the total of the two dual cores. That's what AMD did to get 3.6x the floating point of dual Opteron.
 
Other pics:
00498378.jpg

00498379.jpg
 
Also MU...

Given most processor/OS scalability that I have seen as of late...

Things are getting pretty efficient so with either platform (speculation on my part) will probably scale at or about ~1.8:1.

This is just info gathered during scalability tests across multiple CPUs and OSes. (important to note that I did NOT test these specific processors)..

But I must say that scaling from a single to a second has yielded on many many an occasion nearly a 1.8:1 on numerous threaded applications (NOT GAMES).
Click to expand...
Yeah...but from single to dual is ~1.8....then dual-quad is 1.8x that IIRC...so 1.8 x 1.8= 3.24 x a single core? At least that's how i understood it. :?

Actually you would ADD not multply since you're adding the total of the two dual cores. That's what AMD did to get 3.6x the floating point of dual Opteron.
...and i'm glad that Baron isn't a math teacher..
let's just set an example alright?

IF single core performance = 10, dual core performance = 1.8 x 10 = 18.

dual core performance = 18, quad core performance = 1.8 x 18 = 32.4.

SO quad / single = 32.4/10 = 3.24.
its by multiplication, not by addition.

now I don't know how to calculate dual-quad cores. assuming that we follow the same rule,
1.8 x 32.4 = 58.32.

so theoretically we should see 5.832x performance in dual-quad cores, compared to single core, assuming that octa core is 1.8x the performance of quad core.

now i really feel that Baron is a middle school kid. i learned this back in middle school, if not elementary school.
 
It's a freakin' WIND TUNNEL! Can you imagine having that on your desktop and trying to keep your papers from blowing across your desk? That thing must generate as much heat as 8 Prescott 660s which is the heat source which will be used for the French nuclear fusion reactor.

I've got the perfect way to determine the relevant speeds of the various processors. Launch Photoshop CS3. Take a 600 resolution 2' x 3' poster in PSD and convert it to JPG at max res. Your stopwatch will tell the tale. And at least it will be a benchmark that will mean something in the real world!
 
Hiya Capt. I don't really hate cats. Hate is such a strong word and I'm a really laid back kinda guy. I just dislike them very much. And besides, if they wanna go stand on my pretty beemer I showed you the other day they sure as hell are askin for it. BTW, I sold it.

Can you immagine the sound of that wind tunnel along with the papers flying across your table all the time?

PS. Nice to have you back on the forumz.
 
The C2Q 2.67 took 135 seconds to encode 24 seconds of 1080p video in H.264. Assuming perfect scaling, then an 8-core Core 2 setup would need 68 seconds, which is 2.8x real time. It has been mentioned that 720p encoding is about 3 times as fast as 1080p encoding. So a pair of 2.67 quad-cores could theoretically encode 720p in real-time as well. Of course this will not necessarily happen as there are scaling issues to deal with.

So the bottom line is that the CPUs demonstrated are at least as fast as the C2Qs if the video was 720p, with the scaling ability of the chips being the difference. If the video was 1080p, then the 10h *smokes* what Intel has right now by a factor of 3 or more in video encoding. However, The Barcelona has SSE4a and it is very possible that they used a tweaked SSE4a-capable encoder like Intel used for the 45 nm demo. The Yorkfield/Wolfdales showed a 100% boost in performance with SSE4a, so the video could have very well have been 1080p and encoded in real-time if SSE4a were used. That still says the 10h is 50% faster than the theoretical 8-core-with-perfect-scaling Core 2 Duo if you throw out the SSE4a enhancements.

Personally, I think that it probably was a normal SSE3 encoder and 720p video. This would put a single Agena in the same league as the QX6700, perhaps a bit better, but not a lot. However, we don't know the clock speed of the Agena FXes used nor do we fully know the top launch speeds. If the chip under those sinks was a 2.9 GHz unit, then AMD only matched the QX6700s, but if it's a 2.0 GHz model and the top goes to 2.9, they will clean the C2Q's clock.
Click to expand...

I'm fairly certain the 1080p encoding demo used the R600 to help with the encoding. That might explain why it was so fast.
 
Also MU...

Given most processor/OS scalability that I have seen as of late...

Things are getting pretty efficient so with either platform (speculation on my part) will probably scale at or about ~1.8:1.

This is just info gathered during scalability tests across multiple CPUs and OSes. (important to note that I did NOT test these specific processors)..

But I must say that scaling from a single to a second has yielded on many many an occasion nearly a 1.8:1 on numerous threaded applications (NOT GAMES).
Click to expand...
Yeah...but from single to dual is ~1.8....then dual-quad is 1.8x that IIRC...so 1.8 x 1.8= 3.24 x a single core? At least that's how i understood it. :?

Actually you would ADD not multply since you're adding the total of the two dual cores. That's what AMD did to get 3.6x the floating point of dual Opteron.
...and i'm glad that Baron isn't a math teacher..
let's just set an example alright?

IF single core performance = 10, dual core performance = 1.8 x 10 = 18.

dual core performance = 18, quad core performance = 1.8 x 18 = 32.4.

SO quad / single = 32.4/10 = 3.24.
its by multiplication, not by addition.

now I don't know how to calculate dual-quad cores. assuming that we follow the same rule,
1.8 x 32.4 = 58.32.

so theoretically we should see 5.832x performance in dual-quad cores, compared to single core, assuming that octa core is 1.8x the performance of quad core.

now i really feel that Baron is a middle school kid. i learned this back in middle school, if not elementary school.Makes sense to me...i don't think it should be additive either. :wink:
 
Also MU...

Given most processor/OS scalability that I have seen as of late...

Things are getting pretty efficient so with either platform (speculation on my part) will probably scale at or about ~1.8:1.

This is just info gathered during scalability tests across multiple CPUs and OSes. (important to note that I did NOT test these specific processors)..

But I must say that scaling from a single to a second has yielded on many many an occasion nearly a 1.8:1 on numerous threaded applications (NOT GAMES).
Click to expand...
Yeah...but from single to dual is ~1.8....then dual-quad is 1.8x that IIRC...so 1.8 x 1.8= 3.24 x a single core? At least that's how i understood it. :?

Actually you would ADD not multply since you're adding the total of the two dual cores. That's what AMD did to get 3.6x the floating point of dual Opteron.
...and i'm glad that Baron isn't a math teacher..
let's just set an example alright?

IF single core performance = 10, dual core performance = 1.8 x 10 = 18.

dual core performance = 18, quad core performance = 1.8 x 18 = 32.4.

SO quad / single = 32.4/10 = 3.24.
its by multiplication, not by addition.

now I don't know how to calculate dual-quad cores. assuming that we follow the same rule,
1.8 x 32.4 = 58.32.

so theoretically we should see 5.832x performance in dual-quad cores, compared to single core, assuming that octa core is 1.8x the performance of quad core.

now i really feel that Baron is a middle school kid. i learned this back in middle school, if not elementary school.Makes sense to me...i don't think it should be additive either. :wink:

Hmmm.... Sounds like the 80 core Intel CPU could really suck in relative comparison to the number of cores.

16 cores : 1.8 * 58.32 = 105 (equivalent to approximately 10 single core)
32 cores : 1.8 * 105 = 189 (equivalent to approximately 19 single core)
64 cores : 1.8 * 189 = 340 (equivalent to approximately 34 single core)
128 cores : 1.8 * 340 = 612 (equivalent to approximately 61 single core)

guesstimation: 80 core CPU is roughly equivalent to 40 single core CPUs.


Question: I'm guessing the 1.8 factor is not a hard set number. What factors could help or hinder this number?
 
I am in California and none of this news is new to me.

AMD has a large market of cell phone chips as well as GFX and CPU IC chip production.

Some company divisions earn more than others....here on THW we tend to only view the CPU division.

Take it from me that AMD's CPU division is about do what it has in the past 4 years as far as "market".

If I did not read yesterdays news wrong then AMD demo'ed a Native Quad Core CPU here in California built on a 45nm process befor Intil has shown any 45nm builds.

So goes the life of us NOW system builders and OC'ers.....4-6 months and we are behind the times...again.
 
As far as I know those quads were Agena fx 65nm. the 45nm parts are considered typhoon series so far.
Click to expand...

So was yesterdays news in error or is it my 72 years of age playing tricks on me....again?

Old age and decades of Pot smoking "might" have some effect on my short term memory...or not.

AMD had TWO demos here in California a day apart from each other and my post noted the 2'nd demo in S.F,California.
 
Hard to say and I am up late (missed Jay Leno tonight!).

If I can recalll when I drag myself to a pot full of coffee in the dawn I will look for May 10th,2007 (or 9th ?) news posts and see.

If you want PM me to remind me to do so.
I do forget to do things so easy these days.

BTY....I recall when they built the Folsom Lake and back in the early 70's I worked on the Rancho Seco power plant (shut down now).

Good hunting and fishing up in your part of California.

Z
 
"encoding nearly ran real time"
do anyone have try same task on a C2Q ?
Click to expand...

i think its funny when people say real time and stuff
they can barely get past 30 fps
so is 30 fps real time to them?
lol
please show me benches of what "real time" isYeah, it is. IIRC, video is ~29fps, and tv is around ~24fps. :wink:
 
Hiya Capt. I don't really hate cats. Hate is such a strong word and I'm a really laid back kinda guy. I just dislike them very much. And besides, if they wanna go stand on my pretty beemer I showed you the other day they sure as hell are askin for it. BTW, I sold it.

Can you immagine the sound of that wind tunnel along with the papers flying across your table all the time?

PS. Nice to have you back on the forumz.
Click to expand...

I wouldn't care if my cats sharpened their claws on the Mocassin leather upholstery of my new Rolls Royce Phantom... er... ok, yeah, I'd probably mind then! :lol:

Besides, don't worry about it... I love cats, but I do realize that there are lots of people who don't. Different strokes for different folks!

Perhaps AMD should run their first Wahoo ad with a guy sitting on a desk happily working away on his new 2xQuad on the wing of an A380 in flight. AMD: STAY COOL.

Yeah, sorry about my sudden departure but that's the way it is in the Captain's world. I get a secret message that self destructs and I have to leave at 3 am for all sorts of sordid destinations. I'm here for the foreseeable future, however I am on a ball-bustin' contract so my hanging around will still be somewhat sporadic, depending on what deadlines are hanging over my head.

So ya sold the Beemer, huh? Good on ya! What's your next project? Maybe shoehorning a Viper V10 into a Fiat 500? :lol:

P.S. I have one more thing to say:

WHERE ARE THE F*****G BENCHYS??????????

This is gettin' ridiculous. Shame on you Hector!!!!!!!!!
 
Well look at it this way, you get to travel alot. If I could get an oppertunity like that I'd drop whatever I'm doing now in a heartbeat.

Concerning the cat thing, just like some people have their preferrences in pc hardware, they'll have it in pets too. To each their own I always say.

I just bought a small little car for the time being that's less work than my beemer. Too much work maintaining all the badazz mods on that car. This one has a much smaller engine as well seeing as the fuel prices skyrocketed this side over the last 2 months. It's now just over US$1 per litre making it US$3.85 per gallon. Soon I'll buy me a vespa scooter and pretend to be cool like Jamie Oliver.

Can't agree more about the benchies...
 
Well look at it this way, you get to travel alot. If I could get an oppertunity like that I'd drop whatever I'm doing now in a heartbeat.

Concerning the cat thing, just like some people have their preferrences in pc hardware, they'll have it in pets too. To each their own I always say.

I just bought a small little car for the time being that's less work than my beemer. Too much work maintaining all the badazz mods on that car. This one has a much smaller engine as well seeing as the fuel prices skyrocketed this side over the last 2 months. It's now just over US$1 per litre making it US$3.85 per gallon. Soon I'll buy me a vespa scooter and pretend to be cool like Jamie Oliver.

Can't agree more about the benchies...
Click to expand...

Yeah, travel is great, but after you've travelled as much and as far as I have, (in the words of that Jackson Browne song) "These towns all look the same..."

Here is a good test for ya. Go to the local shelter. Pick up a 6-8 week old kitten. Sit in a quiet place with the kitten and just watch it intently for a while. You can hold it or stroke it if you want and then see what happens. If anyone can do that and still dislike felines, then there is no hope! 😀

I'm paying about the same prices here for fuel, but that is absolutely a bargooooon as compared to a few weeks ago when I filled up at a petrol station in London. GBP1.09/litre for regular unleaded. That's US$2.17/litre or over $8.25 a US gallon! I think that the day that we will all have to drive scooters, and electric ones at that, may not be too far in the future.

Let's all say it together: BENCHYZZZZZZZZZZZZZZZZZZZZZZZZZ!!!!!
 
You must log in or register to reply here.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom