Haswell News

Roger that, I missed that pdf. I'm too anxious right now I guess. I'm dreaming of issues that my current system is having so I can explain to my wife when I need to plunk down some cash for a new mobo and cpu (and ram keke).
 
I have no idea what that has to do with the post I made or the one I replied to.
Well when DDR5 basically didn't have enough bandwidth (and more importantly random access), why should DDR4 be better?

The official inter roadmap is the source,

Actually I looked at that document, and the relevant part was this:
Product Introduction Date 27 May – 7 June 2013 (Day/Time TBD)Sales and AdvertisingDates (to end users) Advertising: NO forms of advertising or promotions to end users allowed (includes print/web)until 2 JuneSales, Shipments: NO sales/shipments to end users until 2 June

If what I see on the document is true, Intel loses desktop big time.

BTW I seen already Z87 Asrock MBs (listed as not yet available and with an image without heatsinks).
 
Last edited:
I have one little question: DDR5 memory? When was that announced or introduced?

I have not seen specs for it. I know OF GDDR5 for video cards though. DDR4 is the newest memory specification from JEDEC that I can remember that's been talked about the past few years. Where is DDR5 coming from?
 
There is no DDR5. There is only GDDR5 which is based off the DDR3 specification with a DDR4-ish feature or 2 (namely 8-bit prefetch buffer) and conforms to the GDDR5 specification as outlined by JEDEC.
 
There is no DDR5. There is only GDDR5 which is based off the DDR3 specification with a DDR4-ish feature or 2 (namely 8-bit prefetch buffer) and conforms to the GDDR5 specification as outlined by JEDEC.

Thank you for the reply. I was wondering where the other user was getting DDR5 from because I know it doesn't exist except on video cards as GDDR5. He kept mentioning that he was working with DDR5 memory. On graphics cards perhaps?

If desktop and server PCs, and even laptops were to get DDR5, it won't be until long after DDR4 has come to market. And, DDR4 is to be out in about a year for servers-- 2014-ish-- and 2015 to 2016 for consumers.
 
Yes I meant these Hynix chips from Kepler cards. When you ignore these bit larger latencies for quasirandom access (which wreak havoc with memory controller), it has not shabby 70 GB/s bandwidth.

I wouldn't bet on DDR5 appearing on the market. They might start to use Linux naming convention. Odd DDR numbers for GFX cards, even DDR numbers for desktop.

BTW from what I seen DDR 4 would have brutal CL15+ latencies. (Considering my current RAM is CL5, it's a lot.) Thus DDR 4 would stay lemon until they would drop latencies a bit.
 
Yes I meant these Hynix chips from Kepler cards. When you ignore these bit larger latencies for quasirandom access (which wreak havoc with memory controller), it has not shabby 70 GB/s bandwidth.

I wouldn't bet on DDR5 appearing on the market. They might start to use Linux naming convention. Odd DDR numbers for GFX cards, even DDR numbers for desktop.

BTW from what I seen DDR 4 would have brutal CL15+ latencies. (Considering my current RAM is CL5, it's a lot.) Thus DDR 4 would stay lemon until they would drop latencies a bit.
Didn't the same happen when we moved from DDR2 to DDR3? Latencies went up but speeds and bandwidth jumped a lot, and power consumption went down.

Looks like similar is happening with DDR4, then.
 
If you know about the actual latency speeds of RAM, you'll find that DDR2 800 with a CL of 5 has the same real-time latency of DDR3 1600 with a CL of 10. The Latency is measured in clock-cycles, not actual time. So when the clock speeds go up, but the latencies stay the same, the CL number seems to go up. If DDR4 3200 has a CL of 20, it will still have the same real-time latency of DDR3 1600 CL10.
 
Yes I meant these Hynix chips from Kepler cards. When you ignore these bit larger latencies for quasirandom access (which wreak havoc with memory controller), it has not shabby 70 GB/s bandwidth.

I wouldn't bet on DDR5 appearing on the market. They might start to use Linux naming convention. Odd DDR numbers for GFX cards, even DDR numbers for desktop.

BTW from what I seen DDR 4 would have brutal CL15+ latencies. (Considering my current RAM is CL5, it's a lot.) Thus DDR 4 would stay lemon until they would drop latencies a bit.
They already have a naming convention and each one is tied to a JEDEC specification also. GDDR and DDR
You are incorrectly calling GDDR5 DDR5.
 
Whether Fudzilla is to be believed or not, it DOES INDEED look like Haswell-E will carry over the DDR4 memory controller from its server variants Haswell-EP/EN (14 cores max) and Haswell-EX (16 to 20 core chip):

http://www.fudzilla.com/home/item/31064-haswell-e-supports-ddr4

So far, we're now looking at:
- Quad-channel DDR4 - Remember 1 DIMM slot = 1 channel. If there isn't enough pins for octa-channel (8 DIMM slots), then we're probably going to have 16GB DDR4 memory modules by then since current X79 boards can have a maximum of 64GB of RAM.

- Maximum DDR4 2133MHz RAM speeds supported - Still well below the supported 3200MHz of DDR4.

- Maximum 2.5MB L3 cache per core

- Socket 2011 - NOT pin-compatible with existing X79 Socket 2011 boards

- 130W TDP minimum for Haswell-E to 145W to 160W TDP for server versions EP/EN and EX.

- 12 to 20 core max - Most likely servers will get the 14 to 20 core versions, and high-end enthusiasts will only see up to 12 cores.

- Integrated PCI-e 3.0 controller.

- All features from Socket 1150 Haswell will be carried over-- transactional memory, TSX instructions, integrated voltage regulator, etc.

- And, obviously, all will have Hyperthreading.​
That means consumers (high-end enthusiasts and workstation folks) will see DDR4 before Skylake is out. This, however, does mean it's coming out mid- to late-2014. DDR4 memory isn't going to be in full mass production until 2014 to 2015 at the most. Servers are supposedly getting them first before regular consumers, so don't expect these modules to be cheap at all in 2014.
 
Last edited:
Opposite of the GPU world. NV puts out it's best at a premium before it's next gen. We enthusiasts have to wait a year after Intel's next gen.:rolleyes:
 
Opposite of the GPU world. NV puts out it's best at a premium before it's next gen. We enthusiasts have to wait a year after Intel's next gen.:rolleyes:

What's another year to wait for the enthusiast version of Haswell? /sarcasm *cough, cough*

Hence, I'll just get Ivy Bridge-E now, be happy for the next three to four years, and by the time Broadwell-E or Skylake-E (if it exists) is out, I'll upgrade by then. I usually upgrade every 3 to 5 years anyway.

Haswell-E is way too far out there to wait for an upgrade. That and I'm sure initial DDR4 RAM modules won't be cheap in 2014, and it'll drop slightly in price by 2015 when Skylake is supposed to be out.

It's weird how it's scheduled...
June 2013 - Haswell

2014-ish - Broadwell (14nm)

Late 2014 - Haswell-E

2015-ish - Skylake (14nm)

Late 2015-ish - Broadwell-E???

2016-ish - Skymont??? (10nm)

Late 2016-ish - Skylake-E???

2017-ish - Who the hell knows?​
 
Opposite of the GPU world. NV puts out it's best at a premium before it's next gen. We enthusiasts have to wait a year after Intel's next gen.:rolleyes:
The chips you're referring to are consumer versions of Xeon models made for 1-2P systems. It doesn't usually take a year, but 6-9 months between mainstream socket and high end enthusiast socket releases aren't unusual. Server chips go through longer validation cycles and the high end enthusiast market is a tiny niche of desktop processors sales. IOW, there really aren't very many people waiting.

The year+ lag is usually for the MP versions, made for 4+ socket systems.
 
Well when DDR5 basically didn't have enough bandwidth (and more importantly random access), why should DDR4 be better?
Apples and non-apples.

You don't mention many details, but the 60 core comment implies you may have been talking about Xeon Phi, which has processor cores not balanced like mainstream high performance processors. Completely separate architectures are not remotely comparable by a metric like DDR5 vs DDR4, especially when execution pipeline characteristics, cache/memory hit ratios and effective bandwidth per core (varied in different application patterns) are so different.

You could clarify, but I still don't see what that had to do with the topics being discussed.
 
Opposite of the GPU world. NV puts out it's best at a premium before it's next gen. We enthusiasts have to wait a year after Intel's next gen.:rolleyes:

Yeah its the opposite of how it was for Nehalem/Bloomfield. I love my i7 920 but its day has passed. I can't wait for Haswell-E and IB-E isn't compelling. Thus I will be buying a regular Haswell system and just dealing with it...

If AMD was competitive on the high end we'd have had Haswell already, possibly Haswell-E but Intel can take its time while we hunger for more power.
 
Yeah its the opposite of how it was for Nehalem/Bloomfield. I love my i7 920 but its day has passed. I can't wait for Haswell-E and IB-E isn't compelling. Thus I will be buying a regular Haswell system and just dealing with it...

If AMD was competitive on the high end we'd have had Haswell already, possibly Haswell-E but Intel can take its time while we hunger for more power.

*covers my i7 920 and i7 970's ears'*

:eek:

Me too though. Haswell-E looks to be my replacement, even if it is 1.5-2 years away.
 
Back
Top