Cadence Announces First DDR4 Controller and PHY IP in 28-nm

Status
Not open for further replies.

mforce2

Distinguished
Feb 12, 2006
96
0
18,630
I disagree , it didn't use to be a bottleneck. Now though the CPUs have integrated GPUs and those GPUs really do benefit from fast RAM. The CPU part doesn't really care that much, I agree.
 
G

Guest

Guest
Memory is still a bottleneck for integrated GPUs. More bandwidth and lower latency per channel is always welcome.
 

memadmax

Distinguished
Mar 25, 2011
2,492
0
19,960
If you think that ram isn't a bottleneck then either you don't know what you are talking about or all you do is run graphics intensive apps(aka games) all day...

Main ram is still turtle speed compared to a CPU and Cache ram.
 

idroid

Honorable
Aug 18, 2012
1,525
0
11,960
[citation][nom]overzealot[/nom]Memory is still a bottleneck for integrated GPUs. More bandwidth and lower latency per channel is always welcome.[/citation]

That's exactly why it would be awesome if Nvidia or AMD make a GPU with RAM-BUS or something
 

alidan

Splendid
Aug 5, 2009
5,303
0
25,780
[citation][nom]mforce2[/nom]I disagree , it didn't use to be a bottleneck. Now though the CPUs have integrated GPUs and those GPUs really do benefit from fast RAM. The CPU part doesn't really care that much, I agree.[/citation]

you are not using those embedded chips to really do anything high end, its really just an added bouns, for most if not all users of them that the graphics can be done better.

if you are even a mild gamer, a low low end gpu beats out integrated.

even the best integrated gpu is beaten out by a 50$ gpu, so in that regard i refuse to call it a real bottleneck.
 

EDVINASM

Distinguished
Aug 23, 2011
247
0
18,690
[citation][nom]memadmax[/nom]If you think that ram isn't a bottleneck then either you don't know what you are talking about or all you do is run graphics intensive apps(aka games) all day...Main ram is still turtle speed compared to a CPU and Cache ram.[/citation]

Can you please base your answer on real-world example? Any app I know doesn't benefit significantly from RAM throughout increasing even 30%. I think more than just DDR4 introduction is needed to tackle this. Although I have no RAM bottlenecks and I run 16GB 1600MHz (2x4x200MHz to be more precise) - that's in Adobe Preduction Suite and games.
As for post above - while I do agree faster RAM helps integrated GPU I also know that these GPUs for the moment are nowhere near low-mid range dedicated GPUs so RAM speed isn't the only issue here.
 

bustapr

Distinguished
Jan 23, 2009
1,613
0
19,780
[citation][nom]alidan[/nom]you are not using those embedded chips to really do anything high end, its really just an added bouns, for most if not all users of them that the graphics can be done better. if you are even a mild gamer, a low low end gpu beats out integrated. even the best integrated gpu is beaten out by a 50$ gpu, so in that regard i refuse to call it a real bottleneck.[/citation]
try thinking about mainstream laptops instead of desktops. integrated gpus are usually the best you can get on a laptop(like 99% of laptop users dont have discrete gpus in laptops) at its price. and people who have these laptops try to push its capabilities alot of the time either to play some game or to edit a video. better am would indeed help out in this area. true that for desktops, integrated graphics is beat out by cheap discrete, but its not usually the case for laptops.
 

mamailo

Distinguished
Oct 13, 2011
166
0
18,690
The main reason for current CPUs desings not increasing the core/module count is because ram can not keep the pace.
It already happend en server space some models where canceled because of that.
 

Miharu

Distinguished
Jun 14, 2007
241
0
18,690

Rambus have impressive number but it's not deliver in performance reign.
Basicly Rambus have a faster rate but also a higher latency.
Big job take less time but smaller job take more time.
And most time you do small access.

The only technology who can use somehow good use of Rambus is PS3 cpu (Cell cpu technology).
:pt1cable:

It's also why DDR3-4 it's a prefer choice over Rambus with x86-x64 cpu.
Higher latency like Rambus will give to any GPUs horrible performances issues.


And the bootleneck it's mostly your motherboard... but when you upgrade for DDR4 you also change your motherboard. So you surely fix most case upgrading to DDR4.
 

memadmax

Distinguished
Mar 25, 2011
2,492
0
19,960
[citation][nom]EDVINASM[/nom]Can you please base your answer on real-world example? Any app I know doesn't benefit significantly from RAM throughout increasing even 30%. I think more than just DDR4 introduction is needed to tackle this. Although I have no RAM bottlenecks and I run 16GB 1600MHz (2x4x200MHz to be more precise) - that's in Adobe Preduction Suite and games.As for post above - while I do agree faster RAM helps integrated GPU I also know that these GPUs for the moment are nowhere near low-mid range dedicated GPUs so RAM speed isn't the only issue here.[/citation]

You didn't even bother reading my post did you?

"or all you do is run graphics intensive apps(aka games) all day"

 

atminside

Distinguished
Mar 2, 2011
134
0
18,680
What memory does the Xbox 360 use? I heard it was even faster than DDR3. Is that kind of memory compatible with x86? I knew it's supper expensive though.
 
[citation][nom]Miharu[/nom]Rambus have impressive number but it's not deliver in performance reign.Basicly Rambus have a faster rate but also a higher latency.Big job take less time but smaller job take more time.And most time you do small access.The only technology who can use somehow good use of Rambus is PS3 cpu (Cell cpu technology). It's also why DDR3-4 it's a prefer choice over Rambus with x86-x64 cpu.Higher latency like Rambus will give to any GPUs horrible performances issues.And the bootleneck it's mostly your motherboard... but when you upgrade for DDR4 you also change your motherboard. So you surely fix most case upgrading to DDR4.[/citation]

Someone knows nothing about Rambus' modern memory interfaces.
 
[citation][nom]alidan[/nom]you are not using those embedded chips to really do anything high end, its really just an added bouns, for most if not all users of them that the graphics can be done better. if you are even a mild gamer, a low low end gpu beats out integrated. even the best integrated gpu is beaten out by a 50$ gpu, so in that regard i refuse to call it a real bottleneck.[/citation]

The Radeon 6550D of the Llano A8s can fight with the Radeon 6570 and is only beaten somewhat by the Radeon 6670 when given dual-channel DDR3-1866 RAM (most 1600 kits will have no trouble overclocking to 1866). That's an old AMD IGP. The new Trinity ones will be better and the next APU family after Trinity will probably have a huge performance boost.
 
[citation][nom]ATMinside[/nom]What memory does the Xbox 360 use? I heard it was even faster than DDR3. Is that kind of memory compatible with x86? I knew it's supper expensive though.[/citation]

The 360 uses mere GDDR3. It's not expensive these days and it's not better than DDR3 for CPUs, although it can be better than DDR3 for GPUs.
 
Also, everyone, RAM is a bottle-neck for some applications. Compression, rendering, and some other things can benefit from faster RAM (although not with high efficiency, IE doubling RAM bandwidth and keeping the same latency might net a 30-50% performance increase assuming that the CPU can keep up). However, for some AVX accelerated work and some other things, RAM can increase performance with near 100% efficiency in performance increases per RAM bandwidth increases.
 

A Bad Day

Distinguished
Nov 25, 2011
2,256
0
19,790
[citation][nom]alidan[/nom]you are not using those embedded chips to really do anything high end, its really just an added bouns, for most if not all users of them that the graphics can be done better. if you are even a mild gamer, a low low end gpu beats out integrated. even the best integrated gpu is beaten out by a 50$ gpu, so in that regard i refuse to call it a real bottleneck.[/citation]

A mid-end APU with fast RAM can run around a high-end APU with standard RAM, especially with AA, Anisotropic filtering, and high resolution.
 

goodguy713

Distinguished
Oct 23, 2009
1,177
0
19,460
look at it this way IF its not on the same die as the cpu the interface as a whole will be riddled with latenancy issues .. so data storage ram gpu all that needs to be on the same chip ... then the latenancy issues will be minimal .. are we any where near that kind of level baring some of the low power system on chips like arm but even they do not have every thing on die .. once that happens motherboards will basically just be a socket and a bunch of plugins for various cables .. this may be too far off in to the future but that's where i see thing could should go .. there already talking about 3d layering in cpu production i could see these parts looking very different in 20 30 years
 

goodguy713

Distinguished
Oct 23, 2009
1,177
0
19,460
[citation][nom]goodguy713[/nom]look at it this way IF its not on the same die as the cpu the interface as a whole will be riddled with latenancy issues .. so data storage ram gpu all that needs to be on the same chip ... then the latenancy issues will be minimal .. are we any where near that kind of level baring some of the low power system on chips like arm but even they do not have every thing on die .. once that happens motherboards will basically just be a socket and a bunch of plugins for various cables .. this may be too far off in to the future but that's where i see thing could should go .. there already talking about 3d layering in cpu production i could see these parts looking very different in 20 30 years[/citation]

lol meaning no ram slots no gpu slots every thing intergrated with drive storage space.. and id bet it would be in the form of a cube.. lol
 
[citation][nom]goodguy713[/nom]look at it this way IF its not on the same die as the cpu the interface as a whole will be riddled with latenancy issues .. so data storage ram gpu all that needs to be on the same chip ... then the latenancy issues will be minimal .. are we any where near that kind of level baring some of the low power system on chips like arm but even they do not have every thing on die .. once that happens motherboards will basically just be a socket and a bunch of plugins for various cables .. this may be too far off in to the future but that's where i see thing could should go .. there already talking about 3d layering in cpu production i could see these parts looking very different in 20 30 years[/citation]

Some things such as mid-ranged and high-end GPUs, memory, and more are simply far too large to be put on the CPU die any time soon. We may get to such a point where this is no longer true, but not anytime soon.
 

alidan

Splendid
Aug 5, 2009
5,303
0
25,780
[citation][nom]blazorthon[/nom]The Radeon 6550D of the Llano A8s can fight with the Radeon 6570 and is only beaten somewhat by the Radeon 6670 when given dual-channel DDR3-1866 RAM (most 1600 kits will have no trouble overclocking to 1866). That's an old AMD IGP. The new Trinity ones will be better and the next APU family after Trinity will probably have a huge performance boost.[/citation]

im going with this chard here
http://www.tomshardware.com/reviews/gaming-graphics-card-review,3107-7.html
with what im saying about the gpu

[citation][nom]bustapr[/nom]try thinking about mainstream laptops instead of desktops. integrated gpus are usually the best you can get on a laptop(like 99% of laptop users dont have discrete gpus in laptops) at its price. and people who have these laptops try to push its capabilities alot of the time either to play some game or to edit a video. better am would indeed help out in this area. true that for desktops, integrated graphics is beat out by cheap discrete, but its not usually the case for laptops.[/citation]

if you got a cheap laptop and suddenly wanted to do photoshop on it, or you got a cheap laptop and wanted to than play a high end game on it out of nowhere, i would call you stupid. but thats my opinion on that. there are laptops that can do that, but they aren't budget laptops.

now lets go with looking up a non integrated gpu in a laptop that is better than the integrated on that same chart

525 - (referb) its paired with an i7, 8gb of memory, at 679$
6630M - (amd, only one i think was on the list) paired with an i5 and 6gb of ram for for 779

if you actually have a use for a gpu, you get one with a dedicated gpu, not an integrated one, because its not a substantial amount more you are paying, but you will get far better performance.

[citation][nom]A Bad Day[/nom]A mid-end APU with fast RAM can run around a high-end APU with standard RAM, especially with AA, Anisotropic filtering, and high resolution.[/citation]

really... some one using an integrated gpu is going to crank the aa af and resolution up? i would call them a moron for that, try to remember that even the best integrated isn't much better than a

X1800 XL, X1950 GT, HD 4650 (DDR2), HD 6450, 7800 GT, 7900 GS, 8600 GTS, 9500 GT (GDDR3), GT 220 (DDR2)

for the desktop.
 
Status
Not open for further replies.