What makes the Galaxy Nexus so special

[nleksan]

BTW, just to test my theory I tried something with my gaming rig:
I've got a heavily-OC'd and liquid-cooled AMD/ATI setup (Asus Crosshair V, Phenom II X6 1100T @ 4.7ghz, 32GB DDR3-1600 RAM, 2x Radeon HD6970's OC'd, 2x240gb SSD's, 2x120GB SSD's, 1x240GB PCIe x4 SSD, RAID10 2TB 7200rpm HDD's, external-enclosure water cooling w 3 main 360mm radiators with 3push/3pull fans each, 2L refrigerated reservoir, 3 main pumps, 3 inline pumps, 3 inline 120mm radiators, 1/2" tubing, pure-copper heatsinks, etc - resulting in 3 separate H2O lines and temps that STAY at ambient unless pushed hard).

Just to see what difference the RAM makes on its own, I took out 3 of the 4 8GB sticks and ran PCMark. Then, I took out the last 8GB stick and tossed in 2x4GB DDR3-1600 RAM sticks (both are 4/8GB sticks are G.Skill Ripjaws X series, timings were set identically) and ran the same exact PCMark test with the ONLY difference being dual-channel versus single-channel.

423point difference overall, in favor of the dual-channel. There are likely many more variables at play, seeing as how I have a processor running 1.5ghz higher than normal, both GPU's running almost 200mhz faster than normal (with shader and memory speeds increased by ~400mhz and ~1ghz respectively), while NB/SB on the MoBo are liquid-cooled and overclocked, as is the HyperTransport-Link.

Clearly, dual-channel versus single-channel will have noticeable performance effects, particularly combined with a dual-core processor, assuming that the memory controller is optimized to take advantage of dual-channel DDR2, which is 4x effective memory bandwidth.

[Big Ry]

BTW, just to test my theory I tried something with my gaming rig:
I've got a heavily-OC'd and liquid-cooled AMD/ATI setup (Asus Crosshair V, Phenom II X6 1100T @ 4.7ghz, 32GB DDR3-1600 RAM, 2x Radeon HD6970's OC'd, 2x240gb SSD's, 2x120GB SSD's, 1x240GB PCIe x4 SSD, RAID10 2TB 7200rpm HDD's, external-enclosure water cooling w 3 main 360mm radiators with 3push/3pull fans each, 2L refrigerated reservoir, 3 main pumps, 3 inline pumps, 3 inline 120mm radiators, 1/2" tubing, pure-copper heatsinks, etc - resulting in 3 separate H2O lines and temps that STAY at ambient unless pushed hard).

Just to see what difference the RAM makes on its own, I took out 3 of the 4 8GB sticks and ran PCMark. Then, I took out the last 8GB stick and tossed in 2x4GB DDR3-1600 RAM sticks (both are 4/8GB sticks are G.Skill Ripjaws X series, timings were set identically) and ran the same exact PCMark test with the ONLY difference being dual-channel versus single-channel.

423point difference overall, in favor of the dual-channel. There are likely many more variables at play, seeing as how I have a processor running 1.5ghz higher than normal, both GPU's running almost 200mhz faster than normal (with shader and memory speeds increased by ~400mhz and ~1ghz respectively), while NB/SB on the MoBo are liquid-cooled and overclocked, as is the HyperTransport-Link.

Clearly, dual-channel versus single-channel will have noticeable performance effects, particularly combined with a dual-core processor, assuming that the memory controller is optimized to take advantage of dual-channel DDR2, which is 4x effective memory bandwidth.

^^^ I like what this dudes saying

Sent from my DROID2 using Tapatalk

[tgyberg]

BTW, just to test my theory I tried something with my gaming rig:
I've got a heavily-OC'd and liquid-cooled AMD/ATI setup (Asus Crosshair V, Phenom II X6 1100T @ 4.7ghz, 32GB DDR3-1600 RAM, 2x Radeon HD6970's OC'd, 2x240gb SSD's, 2x120GB SSD's, 1x240GB PCIe x4 SSD, RAID10 2TB 7200rpm HDD's, external-enclosure water cooling w 3 main 360mm radiators with 3push/3pull fans each, 2L refrigerated reservoir, 3 main pumps, 3 inline pumps, 3 inline 120mm radiators, 1/2" tubing, pure-copper heatsinks, etc - resulting in 3 separate H2O lines and temps that STAY at ambient unless pushed hard).

Just to see what difference the RAM makes on its own, I took out 3 of the 4 8GB sticks and ran PCMark. Then, I took out the last 8GB stick and tossed in 2x4GB DDR3-1600 RAM sticks (both are 4/8GB sticks are G.Skill Ripjaws X series, timings were set identically) and ran the same exact PCMark test with the ONLY difference being dual-channel versus single-channel.

423point difference overall, in favor of the dual-channel. There are likely many more variables at play, seeing as how I have a processor running 1.5ghz higher than normal, both GPU's running almost 200mhz faster than normal (with shader and memory speeds increased by ~400mhz and ~1ghz respectively), while NB/SB on the MoBo are liquid-cooled and overclocked, as is the HyperTransport-Link.

Clearly, dual-channel versus single-channel will have noticeable performance effects, particularly combined with a dual-core processor, assuming that the memory controller is optimized to take advantage of dual-channel DDR2, which is 4x effective memory bandwidth.

Nerdgasm! Man I bet that PC smokes!!

[mightybrick]

Man I bet that PC smokes!!

No, it's cooled.

[tgyberg]

No, it's cooled.

Bazinga!

[Liderc]

Asus Crosshair V, Phenom II X6 1100T @ 4.7ghz

Your first mistake.

[ThaSurgeonGeneral]

BTW, just to test my theory I tried something with my gaming rig:
I've got a heavily-OC'd and liquid-cooled AMD/ATI setup (Asus Crosshair V, Phenom II X6 1100T @ 4.7ghz, 32GB DDR3-1600 RAM, 2x Radeon HD6970's OC'd, 2x240gb SSD's, 2x120GB SSD's, 1x240GB PCIe x4 SSD, RAID10 2TB 7200rpm HDD's, external-enclosure water cooling w 3 main 360mm radiators with 3push/3pull fans each, 2L refrigerated reservoir, 3 main pumps, 3 inline pumps, 3 inline 120mm radiators, 1/2" tubing, pure-copper heatsinks, etc - resulting in 3 separate H2O lines and temps that STAY at ambient unless pushed hard).

Just to see what difference the RAM makes on its own, I took out 3 of the 4 8GB sticks and ran PCMark. Then, I took out the last 8GB stick and tossed in 2x4GB DDR3-1600 RAM sticks (both are 4/8GB sticks are G.Skill Ripjaws X series, timings were set identically) and ran the same exact PCMark test with the ONLY difference being dual-channel versus single-channel.

423point difference overall, in favor of the dual-channel. There are likely many more variables at play, seeing as how I have a processor running 1.5ghz higher than normal, both GPU's running almost 200mhz faster than normal (with shader and memory speeds increased by ~400mhz and ~1ghz respectively), while NB/SB on the MoBo are liquid-cooled and overclocked, as is the HyperTransport-Link.

Clearly, dual-channel versus single-channel will have noticeable performance effects, particularly combined with a dual-core processor, assuming that the memory controller is optimized to take advantage of dual-channel DDR2, which is 4x effective memory bandwidth.

Ok, this is going to be lengthy but please read on...

Funny, I own a similarly built computer myself for video and studio work. HT likes dual channel (128bits wide) for memory and gaming etc but, video and audio editing is actually done better via single channel or 64bits wide. I would guess that TI Omap and Snapdragon CPUs are much the same.

To clarify, dual channel was thought of as double the bandwidth or 128bits vs 64bits but thats actually not true. Sometime ago, someone realized that 64bit and 128bit was actually "all in" or "individual" or Ganged and Unganged.

Meaning, 64bits is the standard memory lane. This has been so for quite some time. What this means is at 64bits wide, your dual core+ CPU can access each individual memory chip and have them effectively performing a couple tasks per chip where at 128bits or dual channel, the memory is put into a pool and the CPU is allowed perform tasks with the shared memory pool rather then the individual memory chips. For some tasks, ie gaming, this is beneficial but for technical tasks, 64bits is not only adequate but often superior.

So on a computer with a dual core chip if you have four memory lanes and you want to be at 64bits, its actually better to have 2 sticks of ram in the maximum amount possible as the CPU will actually "own" one stick per core. In a quad core setup, its much the same at 64bits wide. Four lanes, each with 1 stick of ram and the CPU will "own" each stick of ram per core.

However, as L2/L3 cache increases, this becomes more and more negligible. Not to mention higher and higher HT clock speeds as well as CPU clock speed increases.

And we won't bother getting into DDR3.

How does this apply? Well, I think from what I've seen is that because of Qualcomms Async CPUs, and being dual core, they actually benefit from a ganged memory setup where each core owns essentially half of the available ram. Where TI OMAPs have Sync CPUs (correct me if I'm wrong), they don't work individually, so to speak, and therefore benefit from an Unganged setup.

Which is also why I believe Qualcomm has not gone DDR2 DUAL channel as of yet. I think they realize, because of the current speed of their processors, and the competitors New chips, it makes sense for them to strive to be more.technically efficient until the release of their next gen hardware.

Qualcomms next gen quad core CPU goes dual channel but, their L2 cache also increases to 1mb so, again, it starts to become negligible at this point.

If you look at chip bench scores, the technically efficient approach still seems to ring true given a previous gen tech vs a current gen tech as well as Async vs Sync and ganged and unganged memory setups.

However, this is my hypothesis and could be untrue but I'm guessing, its most likely on point...

My Rezound Rocks the Red n Black... Get over it... Now to get this thing rooted

[ThaSurgeonGeneral]

Think of it as AMD (Qualcomm) vs Intel (TI/Samsung)... It's actually very similar...

My Rezound Rocks the Red n Black... Get over it... Now to get this thing rooted

[nleksan]

^^^ I like what this dudes saying

Sent from my DROID2 using Tapatalk

Nerdgasm! Man I bet that PC smokes!!

Haha, thank you both! Big Ry, I simply try to be as "logical" or (can't think of the right word) while remaining unbiased, or as unbiased as possible, when comparing things like this. It's a result of my education being so heavily-focused on science (chemical engineering and psychopharmacology), but it has trickled into my "normal" life as well
I do have to say, though, at this point we are all pretty much simply "armchair benchmarking", as none of us has two completely-virgin phones (one Rezound/RAZR and one Nexus) in our hands with everything 100% "factory", to set side-by-side and have them both run through the same exact set of benchmarks at the same exact time (to rule out variables such as ambient temperature, humidity, elevation, tidal pattern, planetary alignment, and all of the other variables people will no doubt bring up ).
THEN, and ONLY THEN, will we as a community be able to say which is the "better phone" out of the box. Actually, no, we'd only be able to say which is the FASTER phone out-of-the-box, not which is "better", as "good/better/best" are highly subjective.


Yeah, tgyberg, it is pretty quick... although I am hitting fantastic benchmark scores and I am able to get consistently-high FPS in almost all games at highest-settings (2560x1600 resolution), the new SandyBridge-E processors are making my Phenom II six-core look like a toy :'(
I've been building up the water-cooling setup for ~2 years, so I've had more than enough time to "fiddle-with", "tinker", and "tweak" it to my satisfaction, and it is pretty much exactly how I want it; best part being that it is completely transferable to a new rig, the only components that would require replacement would be the CPU/Chipset/Video-Card/Possibly-RAM heatsinks, although the RAM will likely come with me to my next rig, as will the SSD's and HDD's. But, until AMD releases a Bulldozer or Piledriver CPU that is truly "next generation", I'll simply continue extracting as much performance out of the Phenom II as I can!

*WARNING: UBER-NERD CONTENT BELOW! READ AT YOUR OWN RISK!!!
Currently, I'm using a Danger Den external enclosure for my main H2O components, which allows me to run three separate "circuits" at once, instead of using the same liquid to try and cool: two overclocked GPU's, an overclocked CPU, overclocked RAM, overclocked chipset, and a case full of storage drives! I just recently moved some things around in the case and re-routed some tubing to try and lower my temps another 2-3*C, so currently I've got it set up as follows...
*(FYI: all radiators have both push and pull fans, all push-fans use shrouds to eliminate no-flow zones)*

*Circuit 1: 5/8" Tubing from D.D. Housing to dual-layer pure-copper CPU heatsink, 1/2" tubing from CPU to pure-copper Chipset heatsink on both NB/SB, then through an inline Swiftech dual-circuit pump and another 3" to the inline 120mm radiator (with 2x120mm fans 1push/1pull; mounted on 1 of 3 top case exhaust slots), from there it splits into a "Y" tube going from 1x1/2" to 2x1/4" and runs first to both OCZ Vertex 3 Max IOPS 240GB SSD's and then runs through two short 1" brass connectors to both Intel SSD's, with the two exits merging (via 2x1/4" to 1x1/2" Y-Tube) into a single 1/2" tubing which is then routed back to the D.D. external enclosure after passing-through the inline dual-circuit pump once again.
*Circuit 2: 1/2" Tubing from D.D. Housing to Radeon 6970(#1, fully-enclosed in pure-copper max-surface-area heatsink), entering from top and exiting from bottom, it then goes through a side-mounted 120mm inline radiator (push/pull fans) and through a Swiftech dual-circuit inline pump, from which it heads to 1/2 of the system RAM and cools it via pure-copper heatsinks (sleeve-style) that are mounted on the RAM, and after it leaves both DIMMs, it heads back to the D.D. case
*Circuit 3: 1/2" Tubing - Identical to Circuit 2 except that it cools Radeon6970(#2), goes through its own 120mm inline radiator, shares the Swiftech pump, and cools the remaining two RAM DIMMs before heading back

(Circuits 2/3 are so much smaller because the GPUs' heatsinks take up a LOT of surface area, and dissipate A LOT of heat; the CPU circuit, while generating a lot of heat, is able to dissipate it more quickly due to the higher pressure system.)

Inside the (customized) Danger Den enclosure, there is the following:
- Three Primary water pumps:
- 1x Innovatek Eheim HPPS Ultra Water Pump with Infinitely-Variable Speed Adjustment (for Circuit 1)
- 2x Swiftech MCP655 Silent Water Pumps with Variable Speed Adjustment (Circuits 2/3)
- 3x modular 360mm triple-core/copper-core Radiators with 36 fins-per-inch (407x146x68mm) - approximately 1250W heat dissipation per radiator
- Each Radiator has 3x120mm Push Fans with a 45mm triple-fan shroud and 3x120mm Pull Fans with a 25mm shroud (Fans are 120x38mm, 4000rpm max and 184cfm at that speed! very loud, but sound insulated)
- ThermalTake 2L Refrigerated Reservoir with Triple-Layer Insulation; I added another layer of exterior Insulation (I have the liquid exiting the reservoir at 19.7*F with my current liquid-mix: 1 bottle Redline Water-Wetter, 3x 700mL bottle Koolance Ultra-Performance Non-Conductive Coolant, and about 500mL distilled H2O - freezing point is now at like -45*F)
- Insulation: Gold Foil (left over from my BMW; used to wrap certain parts in both the enclosure and in the PC case, such as refrigerator coil area), R8 1/8"-thick Polyboard lining interior of case, R12 Heat Wrap around Reservoir, and various other bits and pieces of various insulating materials (i.e. Pumps are a combo of silicone, aluminum reflective barrier, and wrap).
- All hosing inside the housing and between Housing and PC Case is shrouded by 5/8"-or-1/2"-interior-diameter 4-layer Silicone Hosing (1" outer-diameter, left-over from upgrading my BMW's cooling system); not very flexible, but very insulating... to bend the silicone in a circle without kinking it, the diameter is about 2 1/2 feet.
- All hosing inside the PC Case has stainless-steel braiding over it to keep it in place
- All connections (between tubing and barbs) are secured with two stainless-steel clamps (one clockwise and one counterclockwise)
- The entire case has been as sound-insulated as possible, seeing as it's only a few feet away from the PC Case itself (Dyna-Mat Pro, Expanding Foam, etc); the case has also had extra ventilation holes drilled in where necessary, and there is metal piping to direct cold air to both the radiators and the refrigerator coils, with the coils having their own "exhaust duct" leading to the top of the case; the radiators have deflectors that direct exhaust air upwards and away from everything else

Inside PC Case:
- 1/2"ID Clear Tubing (100% UV Resistant)
- Stainless Steel Braided Wrapping
- 99.98% Pure-Copper Heatsinks/Water-Blocks with Nickel e-Plated Coating (on air-exposed areas) for CPU/RAM-DIMMs/Chipset(NB/SB/MOSFET)/SSDs/Video-Cards
- 3x Swiftech Dual-Circuit Mini Ultra-Silent Water Pumps (inline)
- 3x 120mm Single-/Dual-Circuit dual-density Radiators with Copper Cores/Fins (INLINE)
- 6x 120x38mm 3500rpm 138cfm Fans (3x push and 3x pull)
- Case Fans to keep airflow solid for the rest of the components - 2x front-bottom intake, 3x top exhaust, 2x rear exhaust (all are 180-240mm in size, relatively-low RPM)
- All case wires and everything are completely tucked away; not a single wire is visible; the tubing is kept as minimal and to-the-sides (of the case) as possible for airflow

Everything is controlled by: one 5.25"bay controller (5" color LED-LCD display, case/radiator-fan controls, pump controls; Shows up to 32 different things, including: current flow speed, pump speed, temperatures, relative efficiency (enter temp vs exit temp), fan RPM, etc)
There are 12 temperature sensors that I installed, plus 5 integrated into the case/MoBo, so I can see everything from the temperature of the liquid leaving the reservoir, coming into the reservoir, before and after each radiator, etc. Quite nice. I also can see the CPU, GPU, RAM, and SSD/HDD temps.
All heatsinks/water-blocks are attached via Arctic Silver 5 Thermal Compound, which was done after sanding/polishing the surfaces to a mirror-finish (increases surface area by decreasing air bubbles).

Keep in mind, this has taken over 2 years and ~$750+ to put together, and I've had a friend do a lot of the work (he is much more savvy with this stuff than I) but it's a constant work in progress. It started out as a simple CPU-cooler, and now it's big enough it needs its own case! lol
I'm currently thinking of adding a fourth circuit, to take stress off of the CPU circuit, and I have an extra 360mm radiator, 120mm radiator, and about 20 different fans, but I don't have the pumps or the sensors, so we'll see... Depends on how much harder I push the CPU.


Blah, sorry for the long post. Once I get going, it's hard to stop me :O

[nleksan]

Think of it as AMD (Qualcomm) vs Intel (TI/Samsung)... It's actually very similar...

My Rezound Rocks the Red n Black... Get over it... Now to get this thing rooted

That's a good point (including your previous, much longer post), as I hadn't been thinking of it as quite so similar to X86 CPU's as that, but I suppose the architecture is going to be more or less the same at some base level.

I think what is important, regarding dual-channel vs single-channel RAM in these dual-core phones is knowing if the 1GB is a single 1GB DIMM, 2x 512MB DIMMs, 4x 256MB DIMMs, etc... From my understanding, the OMAP4460 should theoretically function most efficiently if the memory is set up as two 512MB DIMMs, while the Snapdragon (despite being dual-core) will function most efficiently if it is only 1 1GB DIMM. This is due to the integrated memory controllers and their inherent limitations, which is why the fastest RAM they're putting into phones runs at ~500mhz or so, although likely with relatively low CAS timings, at least as I understand it.

This has me wondering...

If these manufacturers have yet to "master" dual-core CPU's and dual-channel memory in these phones, especially when integrated together...what kind of mess are we gonna see when the quad-core processors come out in the not-so-distant future? Take the Tegra3, with it's 4+1 core design and its 12-core GPU, limited to a dual-channel memory controller. How are 5 CPU's and 12 GPU's (17 cores, essentially) going to work with 1-2GB of dual-channel memory?
I think the BIG LEAP we're going to see is not the quad-core CPU's, the multi-multi-core GPU's, or whatever... It will be when they are able to integrate a combination of a DDR3-type RAM running as many channels as there are cores, as well as dedicated GDDR5-type memory for the GPU('s).

It's almost as if they're all trying to one-up another so badly, they're putting stuff out on the market before the last product they put out is even optimized! Imagine that!




For the record, I am an AMD/ATI guy... so your comparison seems backwards to me