G.Skill's Trident Z DDR4 RAM Kits Push CAS Latency Below Competition

[IInuyasha74]

G.Skill announced new Trident Z-branded DDR4 RAM kits coming in June that push the boundaries of DDR4 one step further with lower latency and higher higher capacities. The new RAM kits will operate at 3200 MHz, 3466 MHz or 4266 MHz.

G.Skill's Trident Z DDR4 RAM Kits Push CAS Latency Below Competition : Read more

 

[2Be_or_Not2Be]

For starters they need to make the dims smaller itx boards are just not big enough for all the heat-sinks and capacitors and have 2 huge dims on it. Every chip manufacture is working on making things smaller except desktop dims.

Have you used Corsair's Vengeance LPX low-profile DIMMs? Good memory, and I haven't had any problems w/their height in several mini-ITX systems I've built. A well-designed CPU cooler that accounts for memory nearby also helps.

 

[Nintendork]

We need ram to be in a cube like form instead of this 1990 stick form.

My idea is a single cube (like HBM) with 16GB/32GB in self dual channel.
2 cubes for 32GB/64GB and quad channel.

High end mobos, workstations with 4 "cube socktets".

 

[RedJaron]

I bet they still pull more volts then the claim and use more power then they claim...

If the modules pull more voltage than stated in their XMP, that's the bias of the motherboard, not the module.

... and are still not as fast as a really low latency DDR3

In which way? At lower frequencies, DDR4 does tend to have higher latency ( at least it's more common for DDR4-2133 and DDR402400 modules to have higher CAS than their DDR3 counterparts ) but that's hardly conclusive. I easily found DDR4-2400 CAS 13 modules for the last SBM. And they easily provided 40% more bandwidth than the DDR3-1600 CAS 9 modules I used the quarter before. So what does it matter if the latency is 0.75ns slower than DDR3 if the overall bandwidth and performance are so much higher?

However that's the mainstream stuff. This is high-end. I have one of these TridentZ kits. DDR4-3200 CAS 14. If you want me to do that math for you, that's the same latency as a DDR3-1600 CAS 7 kit ( which is the exact definition of your "low-latency DDR3" ). Except that DDR3 kit can't get 58.5 GB/s bandwidth like I can out of these puppies.

DD4 isn't worth it yet.

Again, based on what? The DDR4 modules I bought were literally a few dollars more than the DDR3 I purchased three months before them. If you're building a new computer now, you won't save any appreciable money sticking with DDR3.

DDR4 in theory was good but actual testing done by toms proves DDR4 is worse then DDR3 in almost every aspect.

And what testing do you refer to? I've done some of that "actual testing done by Tom's" and I have yet to experience DDR4 performing worse.

I'm talking about shorther in length not in height you know similair to a sodim

If you want DDR4 in SO-DIMM format, I can't fault you there. It'd be a nice option. However, perhaps you better fault motherboard manufacturers for not offering that option, rather than blaming RAM makers for not offering a product that isn't even supported in the DIY market.

 

[knowom]

What we need is DDR DIMM's to be designed more like M.2 NVME and closer to the CPU so we can shorter the motherboard traces further for better stability, performance, and efficiency. It would be great if we got away from copper traces as well, but unlikely. Though I don't see why fiber optics couldn't and shouldn't be used in short lengths.

They could angle the DIMM slots as well like the I-Ram did as another option which would allow for better heat sink clearance and closer motherboard placement. It's not as eloquent as M.2 NVME style solution would be, but probably less involved and easier to do so.

 

[knowom]

Those are some well binned 16GB DIMM's for those voltages bravo G.Skill wish they'd been available few months ago.

 

[nycalex]

are these available to purchase yet?

did not see them @ newegg

 

[peppermintpatty5]

We need ram to be in a cube like form instead of this 1990 stick form.

My idea is a single cube (like HBM) with 16GB/32GB in self dual channel.
2 cubes for 32GB/64GB and quad channel.

High end mobos, workstations with 4 "cube socktets".

I know this is a bit old, but shall we all revert to DIP memory chips?

 

[RedJaron]

I don't believe socketed HBM would work. So far as I know, it has to be very close to the processing unit in order to work; not quite on-die, but definitely on-package.

In order to have this, you'd have to buy your RAM as part of the CPU, which means one very expensive part ( and a LOT of SKUs as you'd need each CPU variation offered in a number of RAM configurations ).