ddr4
Images are sometimes not shown due to bandwidth/network limitations. Refreshing the page usually helps.
You are currently reading a thread in /g/ - Technology
You are currently reading a thread in /g/ - Technology
as there is no ddr4 thread here:
is it worth waiting for ddr4 to be the standard if i want to build a new pc?
also other new hardware-things that are worth waiting for until 2015/16?
Please tell me what you think RAM does that is worth basing an entire build around. I'm genuinely curious as to why DDR4 would be the lynch pin for you building a system.
ddr is dead hail HDM overlord
>>46624894
HBM isn't going to be used as system memory. AMD's future platforms will use DDR4, HBM will be placed on package and utilized as VRAM and flat address space for HSA compliant things.
>>46624809
You need to install ddr into your anus.
>>46624855
i dont know man, i just need a good reason to buy a new pc....
>>46624809
No, it's not. Build your PC and shut the fuck up.
>>46624958
you are not very nice my friend :(
>>46624809
ddr4 is only worth it for iGPUs since it reduces bandwidth contraint
>>46624983
the thing is i only buy a new pc every 5 years or so. do you think there will be problems with ddr3 in 2 or 3 years?
>>46625018
No. Don't fucking worry. If you keep waiting for new tech to arrive you are never going to upgrade your computer.
>>46624936
What makes you think the HBM standard won't be used for Main Memory?
>>46624809
it depends what you have now. if you're using sandy bridge, ivy bridge, or haswell, there's no point in upgrading to haswell-e (the only current desktop platform to use ddr4). 2016 will have desktop skylake and AMD zen, both will use ddr4.
for video cards there's almost always a new high end card every 6 months or so that beats the last competing one in one way or another
ssd's are slowly getting cheaper per GB and most high end ones have been bottlenecked by SATA III for a couple years.
>>46625047
well, i have a phenom II 955 and a 550 TI, so everything will be a big jump. i have a ssd though that i will reuse for the next pc.
>>46625047
thanks for the information btw.
>>46625091
i had a phenom II 965 and upgraded to an i5 4690 and i don't notice anything.
i think you won't notice a big jump either.
>>46624809
well you should wait for Skylake-k in August and I think all Skylake systems outside of ultrabooks will be DDR4. This fall will be a good time to build, Intel 3D nand, Skylake, DDR4, Freesync IPS 4k and 1440p monitors, AMD 300 series
>>46624809
http://www.anandtech.com/show/8959/ddr4-haswell-e-scaling-review-2133-to-3200-with-gskill-corsair-adata-and-crucial/8
Across the board DDR4 is not an improvement over DDR3.
Some DDR3 kits on the market are CL10/CL11 at 2400mhz. DDR4 kits at a similar speed are CL14/CL15.
The difference between 2400mhz DDR4 and 3200mhz DDR4 is essentially nonexistent. CPUs do not need more memory bandwidth.
DDR4's key properties are lower power consumption and higher density. These things matter in the server world when you're dealing with dozens of racks. It can give you a big upgrade in total memory while bringing down power consumption significantly if you have enough racks. For the consumer it doesn't matter. Even the increased density has no practical benefit to the consumer unless you plan on making a huge RAMDisk. DRAM producers are hyping it up because they don't want to see their profits continuing to fall year over year. You shouldn't be excited about it at all.
>>46625042
Signaling. The spec was made around the use of an interposer with the memory placed very close to the chip its feeding. You could have it placed on a mobo, but it would then have to be very close to the socket, and the CPU package would need a thousand more pins. Mobo manufacturers would have to design their own interconnects and ensure signal integrity which is extremely unlikely since the HBM itself would have to change how it signals to work over the longer distance.
>>46624983
Nothing with an IGP is going to use DDR4 for VRAM.
>>46625123
i just googled skylake and that seems very good. thanks
>>46624936
>HBM isn't going to be used as system memory.
HBM allows manufacturers to make your RAM upgrades dependant on CPU upgrades, which is probably Intels wet dream (you need to upgrade memory? here, have this new cpu which has 2x ram but is 33% slower clocked. Oh, you want a cpu that has more ram and isn't slower? that'll be $499.).
You can bet your ass HBM will be used as system memory eventually; it is really just a matter of time. And by time I mean whenever they can get density high enough so you can pop in 64gb per cpu in servers.
Dwarf Fortress.
>>46625233
Only in prebuilts, and not if anyone buying prebuilts has that nerdy friend or relative as a consultant.
Shit like that doesn't normally do so well in the market, only apple can pull that crap because their fanbase is full of morons.
>>46625148
>Across the board DDR4 is not an improvement over DDR3.
This was the exact same thing they said with DDR vs DDR2, then with DDR2 vs DDR3.
Back then, DDR1 had something like CL2 timings, I don't even remember because it has been so long.
>The difference between 2400mhz DDR4 and 3200mhz DDR4 is essentially nonexistent. CPUs do not need more memory bandwidth.
Onboard GPUs however do.
>>46625255
This.
DDR4 is only a year old, and the timings have yet to be fully shortened. Carrizo will use DDR4 memory and it will almost be required if you want to XF with a PCIe Radeon.
All DDR3 bottlenecks AMD APU graphics performance
>>46625255
GPUs use a different type of memory, plus its dedicated so that has nothing to do with DDR4.
>>46625286
>Using AMD
Your fault.
So wait a year for DDR4 to become usable and buy DDR3 at reduced price?
>>46625233
HBM would have to be in an entirely different format for it to be feasible, like HMC's long distance modules.
It isn't going to be used for system memory, and AMD is nothing like intel.
>>46625255
When DDR2 came out it wasn't drastically faster than DDR1.
When DDR3 came out it wasn't drastically faster than DDR2.
A DDR4 3200mhz kit is drastically faster than a DDR3 2400mhz kit, and the difference in bandwidth does nothing. CPUs do not need the bandwidth, no false equivalency argument is going to change this.
No IGPs are going to be utilizing DDR4 for VRAM.
>>46625286
Carrizo does not use DDR4, it uses DDR3.
All of the engineering samples are test boards with DDR3 DIMMS or SODIMMS. AMD isn't releasing a new socket specifically for Carrizo, so there will be no desktop APU utilizing DDR4 either.
It isn't happening. Zen based chips will have their own new socket and they'll use DDR4, but they'll have on package HBM for VRAM.
>>46625248
>Only in prebuilts, and not if anyone buying prebuilts has that nerdy friend or relative as a consultant.
They managed to get away with completely gimping overclocking, first by limiting it to certain extremely overpriced CPU, then by using shit thermal paste on only the overclockable CPUs so you can't really overclock them anyway (unless you go to extreme lengths with delidding/watercooling, and even then you might end up with a bad bin that can't overclock past 10%).
Plus HBM is ridiculously fucking fast, something like 20x faster than DDR3 or DDR4. Are you really telling me that dumbshit consumers won't go for it when benchmarks will say "DDR3: 30MB/s, Intel HBM: 600MB/s"?
>Shit like that doesn't normally do so well in the market, only apple can pull that crap because their fanbase is full of morons.
Intel could do it since they have near monopoly. If they start releasing HBM only, OEMs won't have anything else to use - they won't be able to use AMD, either because they don't have SKUs they can use, or because doing so would void their Intel Candyland tickets.
>>46625344
>not crossfiring your A10-7850k and r9 270x
>>46625376
>Crossfire an APU
You are the biggest retard.
>>46625351
>Carrizo
>FM2
>DDR4 + DDR3 compatible
AMD said it would be compatible, can you link me your sources?
>>46625148
>Signaling. The spec was made around the use of an interposer with the memory placed very close to the chip its feeding. You could have it placed on a mobo, but it would then have to be very close to the socket, and the CPU package would need a thousand more pins. Mobo manufacturers would have to design their own interconnects and ensure signal integrity which is extremely unlikely since the HBM itself would have to change how it signals to work over the longer distance.
See, my question is why AMD doesnt use some HBM as a huge on-package L3 cache, and let the GPU have access to it. It's certainly fast enough for the task, and it would help immensely with the memory bottleneck.
>>46625344
>GPUs use a different type of memory, plus its dedicated so that has nothing to do with DDR4.
ONBOARD gpus use part of the system memory. And it was already proven that those show a huge increase with faster memory (see any AMD APU benchmark).
Now imagine how they would behave when paired with memory that leaves GDDR5 in the dust.
>>46625351
>No IGPs are going to be utilizing DDR4 for VRAM.
So what the fuck are they going to utilize in a DDR4 system, a spare bank of dedicated DDR3 on the motherboard? The SSD? What?
>>46625286
>want to XF
Do people still actually want to SLI or Crossfire?
>>46625390
You can crossfire APU's though.
Wat u talkin bout
Not a 270x but you can with r7250 or 240
>>46625390
No John, you are the retard. It can be done and it's called Dual Graphics.
The fact that it still offers shitty performance is another matter altogether.
>>46625400
The odds of Carrizo coming on FM2+ are virtually zero, and FM2+ is not compatible with DDR4.
The socket does not have enough pins to support it, and no FM2+ board has 288pin DIMM slots.
The only variant of Carrizo that supported DDR4 was their server part that is on an entirely different platform. That server part will likely only exist on paper, as has been the case with a half dozen other server APUs.
>>46625412
Its applicability as an L3 depends on latency, not just its throughput. I've read that at present HBM1's total latency is in the same range as GDDR5 which is nowhere good enough for a CPU cache.
>>46625450
>So what the fuck are they going to utilize in a DDR4 system
System memory.
The HSA spec allows a system to have multiple memory pools. DDR4 will be system RAM. HBM on AMD's future APUs will be VRAM and a HUMA pool with flat address space for HSA compliant workloads.
>>46625461
Retards who want to spend twice as much to run coonsole games
>>46625390
mad cuz can't do that nvidifag?
>>46625148
>Across the board DDR4 is not an improvement over DDR3.
You're missing the main points of DDR4, which are as follows:
Lower power consumption,
Higher density,
Higher clock frequency beyond the current DDR3 modules, which are out of spec anyway,
One channel per module, resulting in higher bandwidth.
>Nothing with an IGP is going to use DDR4 for VRAM.
How can you say that with such certainty?
>>46625637
>You're missing the main points of DDR4,
Hilarious. Try actually reading the rest of that post.
>How can you say that with such certainty?
Intel has on package EDRAM. AMD will utilize HBM, they designed HBM specifically for their GPUs and APUs.
>>46625351
>When DDR2 came out it wasn't drastically faster than DDR1.
>When DDR3 came out it wasn't drastically faster than DDR2.
Yes they were.
>Waffling on about clock rates
Oh so you think the clock rate is the transfer speed? It's not, please stop talking.
This is what makes the change from DDR3 to DDR4 so remarkable. In a single module setup, the transfer rate per clock cycle hasn't increased. This is in stark contrast to all other improvements where the transfer rate per cycle per module increased notably.
>
A DDR4 3200mhz kit is drastically faster than a DDR3 2400mhz kit, and the difference in bandwidth does nothing. CPUs do not need the bandwidth, no false equivalency argument is going to change this.
Your and idiot, kill you're self.
Who the fuck cares? People don't make builds anymore.
>>46625662
Wait, seriously? You're saying
>Hurrdurr you won't use it for CPUs
But you're also saying
>Hurrdurr you'll use it for APUs
Man, you're a right laugh.
You also think that none of the mentioned points matter for consumers, which makes you a right muppet. Haven't laughed this hard in a while mate.
>>46625754
>Yes they were.
No they weren't. Don't try to talk out of your ass to someone who knows far more than you do.
Early DDR3 has the same transfer rate as mature DDR2, it took a long time for DDR3 to mature to the point where it was actually a direct upgrade. A simple wikipedia search shuts down your ignorance
http://en.wikipedia.org/wiki/DDR2_SDRAM#Specification_standards
http://en.wikipedia.org/wiki/DDR3_SDRAM#JEDEC_standard_modules
You're in over your head.
>>46625804
You're honest to god retarded.
>>Hurrdurr you won't use it for CPUs
This isn't what I said at all, not even remotely close. Your childish bullshit has no place in a factual discussion. I pointed out that DDR4's features don't matter to the consumer, because they don't. I never stated or even implied that it wouldn't be used.
>>Hurrdurr you'll use it for APUs
Because AMD's future platforms will use DDR4, its a JEDEC standard for system memory. I explicitly stated that they would not be using it for VRAM, because they won't. AMD created HBM specifically for their GPUs and APUs.
Shitposters belong in /b/.
>>46625480
No, it's not another matter. We know it can be done but the fact that it offers shitty performance and comes with all the normal downsides of having dual graphics is why it's retarded to do it.
Britfag got BTFO
>>46624809
Bah, why did they move the notch closer to the middle? I hated this on DDR2.
Quad channel DDR4 not faster than DDR3.
Keep on believing poorfags.
>>46625662
>Intel has on package EDRAM.
wouldn't they want to use HBM instead though since that EDRAM die is huge, runs hot as fuck and gulps down the watts?
>>46626160
intel is also using that EDRAM as an L4 cache so they're getting some more use of it. How much power it sucks down ultimately depends on their fab, they could feasibly get it to the point where it'd no longer eat up a ton of thermal headroom. I think the mobile chips like the i7 4950HQ were just a test run for them to get experience fabbing it.
>>46626134
No one said this.
Lot's of good information ITT
>>46626264
Oh, an advertisement that says the latest is the fastest. What news!
Also: as if 1866MHz CL9 Memory would be fast for todays standards. They are comparing mid end old gen that is 10$ cheaper than high end old gen against latest expensive new gen stuff.
>>46626892
>Oh, an advertisement that says the latest is the fastest. What news!
Not even, look at the CPU clock speeds.
>>46626906
not only that, it's an 8 core chip vs a 6 core chip
ddr4 already is the standard.
>>46626264
>Average FPS
>Lower is better
u wot m8
>>46626264
>>46626946
AHAHHHH
DDR4 blows 2133MHz Out of the fucking water.
Think of that speed with lower latencies!
>>46626946
The moron from tweaktown didn't proof read his own chart.
>>46626052
That's not the worst part.
Look closely at the gold contacts.
The edge have smaller metal contacts than the center. It is a MASSIVE pain in the ass to install ddr4. Too much force and you're heading go snap city with your motherboard. Too little and it won't make contact at all forcing you to check every single stick. The rention clips will snap and secure the ram regardless if it's making contact or not.
I don't know which asshole designed it but I hate it. They made the simplest part to install one of the most painful now.
>>46626973
The only performance gains shown there are from CPU clock speeds. Try actually reading.
>>46624936
I think you'll find that AMD has been pushing hUMA (unified memory) so they'll be looking to use one form of memory as the system and video memory for their heterogeneous software architecture, that memory is HBM.
>>46627041
No. They've been highlight HUMA for its benefits in compute, not for everything in general.
The HSA spec specifically has exceptions for system memory. HBM is for VRAM and HSA compliant workloads, their APUs are still going to use normal DIMMs for system memory.
>>46627027
> snap city
make sure not to round your back and use full range of motion when performing a DDR4 super set. It's like you don't even want to make it.
>tips superior knowledge and breeding
>>46627069
You don't have any clue what you're saying do you?
>>46627480
I know more than you ever will.
http://semiaccurate.com/forums/showpost.php?p=219746&postcount=2
>>46625344
Integrated GPUs use regular system RAM you stupid fuck, get off my /g/
>>46627489
>Trusting PTmD "2^10" or Charlie Bongdankian
I should have known you would do this.
HBM confirmed for Main Memory 2017
>>46627535
>>46627557
http://www.hsafoundation.com/standards/
The moar you know
>>46627563
No ones talking about HSA. The tech in question is HBM, which is to say main memory with Through Silicon Vias, which are already in production and use in servers as main memory. So HSA isn't really on topic at all.
The more you know...
>>46627622
Looks like you're too stupid to coherently follow the discussion.
The HSA spec has exceptions that allow a totally HSA compliant device to have multiple pools of memory.
HBM is being used by AMD's APUs for VRAM, and for HSA compliant workloads. They are still going to use standard DIMMs for system memory, this is why those exceptions exist. They are not sacrificing expandable system memory.
Shitposting like a little retard isn't going to help you save face. Everything quoted in pTmd's post is pulled straight from the HSA provisional spec, so trying to write it off just shows how childish you are.
