- Status
AlienIsGOD
Distinguished
- May 3, 2008
- 255
- 0
- 18,810
[citation][nom]f-14[/nom]i'm with this guy, why the thumbs down? he is dead on right why bother with RIM rambus just like why bother with DDR4 when DDR5 is already here and a finished product on video cards for the last year.look how long DDR2 lasted, you will be stuck in the same boat with DDR4 because the switch to DDR5 will be a year after.do you guys really believe DDR4 is going to keep up with quad, septa, octa cpu's, PCIE 3.0 and usb 3.0 at the same time? i sure don't, but that's due to past experience.DDR4 is not even out of development and it's already fossilized dinosaur poo.this is like win 8 just starting to implement touch interface when voice command has been out for almost a decade before siri arrived.you shouldn't be playing catch up, you should skip over it for the next great thing.anything less is just milking the consumers for nickle and dimes and is liable to ruin your reputation for being a leader instead of the follower you really are.[/citation]
i don't even know where to start correcting you...
i don't even know where to start correcting you...
what does nobody ever talk about wait states when talking about ram. Who gives a frack if it has 2133MT/S if it's 20-20-20-10??
Lol, if we have boards supporting 64~128 Gb RAM about 2014, thyen MS, Intel and AMD should start developing a 96~128 bits OS, because it'll overrun the 64 bits specs by 2018 ...
alextheblue
Distinguished
- Apr 3, 2001
- 3,078
- 106
- 20,970
[citation][nom]NoXtress[/nom]Lol, if we have boards supporting 64~128 Gb RAM about 2014, thyen MS, Intel and AMD should start developing a 96~128 bits OS, because it'll overrun the 64 bits specs by 2018 ...[/citation]Err, some 64-bit editions of Windows Server 2008 R2 support 2TB of memory already. So there's no reason that newer consumer 64-bit editions of Windows won't have their memory ceilings raised. So by then we'll have Windows 10 supporting plenty of memory on 64-bit, there's no dire need for 128-bit memory addressing yet.
[citation][nom]Bob Washere[/nom]what does nobody ever talk about wait states when talking about ram. Who gives a frack if it has 2133MT/S if it's 20-20-20-10??[/citation]
My friend tried to pull something like that before. He claimed to have a ridiculously high clock rate for his RAM, but the timing was like in the 20s or 30s.
I lol'ed.
My friend tried to pull something like that before. He claimed to have a ridiculously high clock rate for his RAM, but the timing was like in the 20s or 30s.
I lol'ed.
blazorthon
Glorious
- Sep 24, 2010
- 14,022
- 21
- 43,265
[citation][nom]demonhorde665[/nom]they quote a lot of figures while saying nothing about expected latency . keep in mind ddr3 was slow to adopt for several reasons 1. many peopel had jsut shelled out money of ddr2 based systems before ddr3 was annouced or available . 2. DRR3 did and still does cost drastically more than ddr 23. ddr3 has drastically higher latency compared to ddr 2 , if you take a lower speed ddr 3 module and match it to the same speed ddr2 the ddr 2 will perform better because it has lower latency number 3 being the Biggest reasons budget minded enthusiast were not enthusiastic about ddr3. we need latency figures along side the speed figures .[/citation]
1. DDR3 is far cheaper than DDR2 ever was.
2. DDR3 has lower latency than DDR2.
DDR3 has higher TIMINGS, not latency. Timings are how many clock cycles a latency takes. The clock frequency of DDR3 improved enough that the higher timings are offset and the latency is actually lower.
GDDR3 is based on DDR2, it is NOT DDR3. GDDR3 performs similarly to DDR3, but seems to be able to go to higher clock frequencies more stably, but graphics RAM has much looser timings than comparable system memory and can have higher voltages. GDDR4 is also based on DDR2/GDDR3. GDDR5 is based on DDR3. It is so much faster than DDR3 because GDDR5 isn't really DDR, it is QDR. That means quad data rate instead of dual data rate.
DDR4 is a whole new technology that is an evolution of DDR3 and also takes some improvements from graphics memory that system memory has thus far lacked. However, DDR4 is still dual data rate, unlike GDDR5's quad data rate. GDDR5 is not the same as DDR4, but they have some similarities. However, all memory has SOME similarities. DDR4 has a completely different topology from previous memory technologies too.
[citation][nom]computerfarmer[/nom]Now that we have cleared up DDR4 vs GDDR5. I have another question. What is DDR4's bandwidth example: DDR3 1000mhz = 2000mhz effective bandwidth - GDDR5 1000mhz = 4000mhz effective bandwidth DDR4 1000mhz = ????mhz effective bandwidthLast time there was a memory change similar to this, Intel went one direction(Ram bus DR) and AMD went with DDR2. Motherboards are only at a max of 32gb of ram and windows 7 can handle 6x this amount.It is a good thing there will be lots of time before we see this change.[/citation]
DDR4 will be 1000MHz=2000MHz effective, but can have 2000MHz=4000MHz effective, unlike DDR3. DDR has higher clock frequencies than DDR3 instead of higher effective clock frequencies.
[citation][nom]A Bad Day[/nom]My friend tried to pull something like that before. He claimed to have a ridiculously high clock rate for his RAM, but the timing was like in the 20s or 30s.I lol'ed.[/citation]
We already have DDR3-2400MHz at 10-10-10 and similar latencies. DDR4 is a superior tech that can go to much higher bandwidths without huge latencies. It is simply more energy efficient, allowing higher frequencies. Remember, it is a balance of timings and frequencies that gives you latency, not just timings. 30-30-30 timings at like 10GHz would have almost four times lower latency than 1600MHz at 9-9-9 (compare 11.25 nanoseconds for the DDR3 to 3 nanoseconds for the hypothetical memory).
[citation][nom]NoXtress[/nom]Lol, if we have boards supporting 64~128 Gb RAM about 2014, thyen MS, Intel and AMD should start developing a 96~128 bits OS, because it'll overrun the 64 bits specs by 2018 ...[/citation]
64 bit addressing is good up until like 16 exabytes or so. All memory limits in an OS are artificial limits set by the creator of the OS and can be increased with each new OS. We shouldn't need to go beyond the limits of 64 bit addressing until 40 years from now for desktop machines, maybe earlier for servers. However, memory needs aren't increasing too quickly right now. In fact, going up from Windows Vista, each new OS is even lighter, getting closer to where XP was in memory usage. Because of that, I'm not sure about regular people EVER needing to go beyond 64 bit, or at least for a very long time, uneless there becomes another reason to advance past 64 bit.
1. DDR3 is far cheaper than DDR2 ever was.
2. DDR3 has lower latency than DDR2.
DDR3 has higher TIMINGS, not latency. Timings are how many clock cycles a latency takes. The clock frequency of DDR3 improved enough that the higher timings are offset and the latency is actually lower.
GDDR3 is based on DDR2, it is NOT DDR3. GDDR3 performs similarly to DDR3, but seems to be able to go to higher clock frequencies more stably, but graphics RAM has much looser timings than comparable system memory and can have higher voltages. GDDR4 is also based on DDR2/GDDR3. GDDR5 is based on DDR3. It is so much faster than DDR3 because GDDR5 isn't really DDR, it is QDR. That means quad data rate instead of dual data rate.
DDR4 is a whole new technology that is an evolution of DDR3 and also takes some improvements from graphics memory that system memory has thus far lacked. However, DDR4 is still dual data rate, unlike GDDR5's quad data rate. GDDR5 is not the same as DDR4, but they have some similarities. However, all memory has SOME similarities. DDR4 has a completely different topology from previous memory technologies too.
[citation][nom]computerfarmer[/nom]Now that we have cleared up DDR4 vs GDDR5. I have another question. What is DDR4's bandwidth example: DDR3 1000mhz = 2000mhz effective bandwidth - GDDR5 1000mhz = 4000mhz effective bandwidth DDR4 1000mhz = ????mhz effective bandwidthLast time there was a memory change similar to this, Intel went one direction(Ram bus DR) and AMD went with DDR2. Motherboards are only at a max of 32gb of ram and windows 7 can handle 6x this amount.It is a good thing there will be lots of time before we see this change.[/citation]
DDR4 will be 1000MHz=2000MHz effective, but can have 2000MHz=4000MHz effective, unlike DDR3. DDR has higher clock frequencies than DDR3 instead of higher effective clock frequencies.
[citation][nom]A Bad Day[/nom]My friend tried to pull something like that before. He claimed to have a ridiculously high clock rate for his RAM, but the timing was like in the 20s or 30s.I lol'ed.[/citation]
We already have DDR3-2400MHz at 10-10-10 and similar latencies. DDR4 is a superior tech that can go to much higher bandwidths without huge latencies. It is simply more energy efficient, allowing higher frequencies. Remember, it is a balance of timings and frequencies that gives you latency, not just timings. 30-30-30 timings at like 10GHz would have almost four times lower latency than 1600MHz at 9-9-9 (compare 11.25 nanoseconds for the DDR3 to 3 nanoseconds for the hypothetical memory).
[citation][nom]NoXtress[/nom]Lol, if we have boards supporting 64~128 Gb RAM about 2014, thyen MS, Intel and AMD should start developing a 96~128 bits OS, because it'll overrun the 64 bits specs by 2018 ...[/citation]
64 bit addressing is good up until like 16 exabytes or so. All memory limits in an OS are artificial limits set by the creator of the OS and can be increased with each new OS. We shouldn't need to go beyond the limits of 64 bit addressing until 40 years from now for desktop machines, maybe earlier for servers. However, memory needs aren't increasing too quickly right now. In fact, going up from Windows Vista, each new OS is even lighter, getting closer to where XP was in memory usage. Because of that, I'm not sure about regular people EVER needing to go beyond 64 bit, or at least for a very long time, uneless there becomes another reason to advance past 64 bit.
DDR4standard
Honorable
- May 21, 2012
- 3
- 0
- 10,510
DDR4 technology already is available in Netlist (NLST) HyperCloud memory - now available from IBM (as HCDIMM) and HP (as HP Smart Memory HyperCloud or HDIMM).
NLST HyperCloud will supplant LRDIMMs - as LRDIMMs are based on NLST IP.
LRDIMMs had copied NLST IP - and been aggressive about it. However they have suffered a defeat in their challenge of Netlist IP - as NLST patents '537 and '274 have survived reexamination at the USPTO with all claims intact - which is a powerful signal of what is to come for Netlist vs. Inphi (basically Inphi will be unable to challenge NLST IP in court !).
LRDIMMs also underperform - as they have high latency and are unable to reach full speed at full memory loading (3 DPC at 1333MHz is not achievable by LRDIMMs on HP servers).
Therefore the market for LRDIMMs will be served by NLST HyperCloud - and if LRDIMMs are sold maybe NLST will get damages from that eventually (or LRDIMMs could face recall ?).
LRDIMMs are also at end-of-life because they have implemented NLST IP improperly - by choosing a centralized buffere which leads to asymmetrical line lengths and skew (thus the higher latency issues).
Netlist says they will be the first proprietary standard which will be adopted by the industry - for DDR4.
And DDR4 is looking more and more like NLST HyperCloud.
So expect JEDEC to license NLST IP sometime before finalization of the DDR4 standard.
Until DDR4 appears you will only be able to buy NLST HyperCloud to do what you thought LRDIMMs could (as LRDIMMs under a cloud) - however for DDR4 expect licensees to also be making memory along with NLST.
NLST HyperCloud will supplant LRDIMMs - as LRDIMMs are based on NLST IP.
LRDIMMs had copied NLST IP - and been aggressive about it. However they have suffered a defeat in their challenge of Netlist IP - as NLST patents '537 and '274 have survived reexamination at the USPTO with all claims intact - which is a powerful signal of what is to come for Netlist vs. Inphi (basically Inphi will be unable to challenge NLST IP in court !).
LRDIMMs also underperform - as they have high latency and are unable to reach full speed at full memory loading (3 DPC at 1333MHz is not achievable by LRDIMMs on HP servers).
Therefore the market for LRDIMMs will be served by NLST HyperCloud - and if LRDIMMs are sold maybe NLST will get damages from that eventually (or LRDIMMs could face recall ?).
LRDIMMs are also at end-of-life because they have implemented NLST IP improperly - by choosing a centralized buffere which leads to asymmetrical line lengths and skew (thus the higher latency issues).
Netlist says they will be the first proprietary standard which will be adopted by the industry - for DDR4.
And DDR4 is looking more and more like NLST HyperCloud.
So expect JEDEC to license NLST IP sometime before finalization of the DDR4 standard.
Until DDR4 appears you will only be able to buy NLST HyperCloud to do what you thought LRDIMMs could (as LRDIMMs under a cloud) - however for DDR4 expect licensees to also be making memory along with NLST.
- Status
TRENDING THREADS
-
Question No POST on new AM5 build - - - and the CPU & DRAM lights are on ?
- Started by Uknownflowet
- Replies: 13
-
-
-
-
Discussion Feasibility & feedback on a compact, high-density 20-NVMe JBOF Enclosure ?- Started by NOR1NCO
- Replies: 1
-
Latest posts
-
-
-
-
-
-
Question RTX 5070 (12GB) vs RX 9070 (16GB) - Which is truly best for 1440P gaming?- Latest: Dark Lord of Tech
Facebook
Twitter
Instagram