• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

AMD "Istanbul" Performance Demo

Wow it takes that long for it to render a picture of a motorcycle. I am just not that impressed :)
 
Dman014 said:
Wow it takes that long for it to render a picture of a motorcycle. I am just not that impressed :)
Click to expand...


try to do that with your CPU and see how long it would take ...

then you will realize how crappy your CPU is compare to what they have over there :p
 
Not too shabby for a quick bench. However, I'll wait to evaluate until Kyle and the rest of the crew take a real [H]ard look at that. Not saying that to pimp the site, I'm saying that because I trust them to cut the crap and bring facts, not spin and definitely not BS.

Now, a question: I caught a glimpse of the rendered resolution in the bench they ran: 800x600. Is that a standard/default size, a CPU-bound size (i.e., no VGA stress) or is that a size that tests the entire system?
 
That's really nice power numbers for a 6 core cpu, if its in the same power envelope as they claim(75W).
 
For a 3D guy like me, 30% is a HUGE gain in performance. As for why it takes so long to render a motorcycle, its all about ray tracing. Thats a massive load on a system and always brings a render down.
 
knob said:
Not too shabby for a quick bench. However, I'll wait to evaluate until Kyle and the rest of the crew take a real [H]ard look at that. Not saying that to pimp the site, I'm saying that because I trust them to cut the crap and bring facts, not spin and definitely not BS.

Now, a question: I caught a glimpse of the rendered resolution in the bench they ran: 800x600. Is that a standard/default size, a CPU-bound size (i.e., no VGA stress) or is that a size that tests the entire system?
Click to expand...

I'm fairly sure that render test was completely CPU bound.
 
Is it really a great demo to show? You have 50% more cores and your only getting 30% better performance......not only that but the new cores are meant to be further optimised than shangai.

I know that you dont always get the same boost from every extra core you get, but still, it doesnt seem that great, especially in an application (raytracing) which normally scales almost perfectly with more cores.
 
Toytown said:
Is it really a great demo to show? You have 50% more cores and your only getting 30% better performance......not only that but the new cores are meant to be further optimised than shangai.

I know that you dont always get the same boost from every extra core you get, but still, it doesnt seem that great, especially in an application (raytracing) which normally scales almost perfectly with more cores.
Click to expand...

QFT

Just look at some 2vs4 core raytracing benches, its almost twice as fast. A 30% performance increase with 2 more cores is not impressive at all. (unless clockspeed is significantly slower)
 
Sorry, I didn't get a rise out of this.

I was hoping for 40-50% improvements since the architecture has(apparently) been upgraded but... all I'm seeing here is 6 x AMD cores putting out what 4 intel Cores has been putting out for sometime now(Core i7).

What's even worst is when intel decides to up the anti on there own packaging(8 cores)... Which will make this demonstration even less interesting. :rolleyes:
 
Esstee said:
Sorry, I didn't get a rise out of this.

I was hoping for 40-50% improvements since the architecture has(apparently) been upgraded but... all I'm seeing here is 6 x AMD cores putting out what 4 intel Cores has been putting out for sometime now(Core i7).

What's even worst is when intel decides to up the anti on there own packaging(8 cores)... Which will make this demonstration even less interesting. :rolleyes:
Click to expand...

I'm sorry, when can one purchase such famed 8-core CPU? :)
 
Stoly said:
QFT

Just look at some 2vs4 core raytracing benches, its almost twice as fast. A 30% performance increase with 2 more cores is not impressive at all. (unless clockspeed is significantly slower)
Click to expand...
I don't know much about rendering, but last I checked 4 was double the amount of 2, so ~ 50% would seem right, but this isn't doubling. Maybe I'm crazy, but it seems that those 2 extra cores make up 1/3 of the total instead of 1/2 of the total in the case of going from 2 to 4 cores. My math sucks these days, but if all was perfectly scaling, if doubling gained ~ 50%, shouldn't 1/3 gain ~ 33%? If I've got it wrong I'm open to some education.
 
Astronomica said:
If he benched those two chips together this forum would erupt into flames.



Yah i guess he should of benched it against all the other 6 core cpu's? :rolleyes:
Click to expand...

lol erupt? how about explode in a ball of flames.. it would turn into a ban stick fest for the admins..

also it was 2px4 and 2px6.. so it was actually 8 cores vs 12 cores.. not 4 vs 6..


and for the other guy that asked.. 800x600 is the default resolution for cinbench.. and the score they got with the istanbul makes me want to cry.. lol.. best i ever got with my 940 is 13580 CB.. even though it is running 8 cores more then mine.. still nice..
 
Esstee said:
Sorry, I didn't get a rise out of this.

I was hoping for 40-50% improvements since the architecture has(apparently) been upgraded but... all I'm seeing here is 6 x AMD cores putting out what 4 intel Cores has been putting out for sometime now(Core i7).

What's even worst is when intel decides to up the anti on there own packaging(8 cores)... Which will make this demonstration even less interesting. :rolleyes:
Click to expand...


its a server processor.. they arent even running close to what the consumer level i7 does.. and uses half the power.. you cant compare server hardware to consumer hardware..
 
mynamehere said:
I don't know much about rendering, but last I checked 4 was double the amount of 2, so ~ 50% would seem right, but this isn't doubling. Maybe I'm crazy, but it seems that those 2 extra cores make up 1/3 of the total instead of 1/2 of the total in the case of going from 2 to 4 cores. My math sucks these days, but if all was perfectly scaling, if doubling gained ~ 50%, shouldn't 1/3 gain ~ 33%? If I've got it wrong I'm open to some education.
Click to expand...
By adding two more to the original four you would expect a 50% performance increase compared to a four core cpu because two cores is half or 50% of the original four so you would expect an increase performance by 50% compared to a four core. As a whole, 2 of the 6 cores is 33% like you said, but this is comparing the performance to a four core, not a 6. I think that's right...
 
Astronomica said:
If he benched those two chips together this forum would erupt into flames.
Click to expand...
It's already been extrapolated using AMD's 30% figure, and 2S Nehalem (8 cores) would still lead 2S Istanbul (12 cores) by 30-40% in the published workstation and server benchmarks. It's not really unexpected since in many benchmarks 2S Nehalem = 4S Shanghai.
 
silent hero said:
By adding two more to the original four you would expect a 50% performance increase compared to a four core cpu because two cores is half or 50% of the original four so you would expect an increase performance by 50% compared to a four core. As a whole, 2 of the 6 cores is 33% like you said, but this is comparing the performance to a four core, not a 6. I think that's right...
Click to expand...

You got it. 4 x 1.5=6. Therefore the difference between the original is half again as much.
 
For those of you keep using the 1.5x figure, it doesn't work like that. If you have two computers as opposed to one, you'll get twice the processing power, but that's just because you're running something twice in parallel instead of once. For this example, it's not perfect scaling since the cores are going to be sharing something, be it on-chip circuitry, memory bus, cache, whatever. I didn't look at the tech specs of this so I don't know cores per die, die per chip, or anything, but you're never going to see perfect scaling.

Here's just one way to think of it from a non-hardware perspective. If you have one core then the operating system and all processes are going to be running on that single core. By adding another core, in addition to the sharing of hardware resources, you're adding the burden of multi-core scheduling. The OS has to determine what core is executing what on top of normal scheduling, plus there has to be communication between each processor core, so the performance will degrade somewhat. Each core is probably doing equal work, but there's just some overhead that prevents performance gains being equal to the number of chips/cores.
 
I think the bigger problem with the 1.5x scaling is the assumption that the clock speed will be the same for Istanbul. Istanbul is likely clocked several hundred MHz slower than the top bin Shanghai. The "30%" performance increase strongly hints at that.
 
(edit) The guy in the video says the processors are running at the same speed. I guess they decided to sandbag the 2S quad core server.
 
silent hero said:
By adding two more to the original four you would expect a 50% performance increase compared to a four core cpu because two cores is half or 50% of the original four so you would expect an increase performance by 50% compared to a four core. As a whole, 2 of the 6 cores is 33% like you said, but this is comparing the performance to a four core, not a 6. I think that's right...
Click to expand...

That's American education for you... just kidding...

You're confusing two different measurements.

Quantity:
6 is 1.5 more than 4 so the number of cores increased by 50%.
difference is 2. 2 / 4 = .5 or 50%
BUT... This does NOT mean the performance also increased by 50%.

Performance or Time lapse using a simple example:
Given that 1 person can unload a truck at a rate of 1 box per second.
How long will it take 4 people to unload 120 boxes? Answer: 120 / 4 = 30 seconds.
How long will it take 6 people to unload 120 boxes? Answer: 120 / 6 = 20 seconds.

20 seconds is not 50% faster. 50% faster would be 15 seconds.
20 seconds is 33.3% faster. The difference (10sec) divided by original time (30sec).
 
I don't know I'm not impressed. Watching the video and hearing about how Istanbul performs better with the same constraints as Shanghai is much like watching the Special Olympics. It's nice that there's a performance increase, but renderfarms and HPC clusters are all about top high end performance and not necessarily how efficient a chip is.
 
You must log in or register to reply here.
Back
Top