Discussion in 'Article Discussion' started by Gareth Halfacree, 21 Nov 2013.
128MB eDRAM for next-gen GPUs.
Always good to see Intel making things backwards compatible.
Ho wait you still need a new Mobo to get a 9 series chipset.
Anyone think Intel is ripping people off here.
They have no decent competition in the consumer market, so they are able to pull any shenanigans they like now.
This seems ass-backwards. Integrated graphics is great for i3 and i5 builds where you can get away without a discrete GPU (e.g. media-centre, basic office Pc, etc), but if you're plumping for a higher-end CPU you'll likely be buying a discrete GPU too making that integrated GPU just a waste of die space (and money).
Iris is expensive.
Not at all. Remember that the gaming market isn't Intel's core demographic. Intel's Iris Pro IGP includes features like Quick Sync, designed to speed up video rendering - meaning that someone doing video-heavy work could easily not bother with a discrete graphics card in their system, saving cash, power draw, heat, noise and space. Even for 3D work, Iris Pro is good enough - especially if you're not rendering directly on the workstation but instead shipping the work out to a renderfarm, or if you're doing ray-tracing with something like one of Imagine's Caustic render cards.
Sure, a gamer will always pair a Core i7 with a discrete GPU or six - but gamers aren't the whole, or even the majority, market.
You speak such blaspheme
Quick grab the pitchforks
They're going to be the few who'll actually overclock their chips though. Anyone doing workstation stuff as you've mentioned wouldn't dare using a chip beyond its spec if it's time/business critical work they're doing. Hence there's no reason to have it on a K-series chip.
Unless they're buying pre-overclocked workstations, which aren't exactly uncommon. Add to that, I know several businesses that mildly overclock their workstations as standard as a means of getting more bang for their buck.
While i agree gamers are probably the few who will OC the chip, the K series is likely to come some 6 months or more after the standard chips. Overclocking and gamers are an after thought to their main market, the cost of totally redesigning a chip would be out of the question.
As much as people are talking about the eDRAM being used for IGP, you're all missing one cool point: we now have a 128MiB block of fast L4 cache, well, faster than RAM, slower than L3 and lower, as is the case with cache levels.
I for one am looking forward to programmers fitting core gameplay entirely into the cache and leaving the RAM out to be used for large resources, like textures and whatnot.
Are the variants with less L3 cache only the Iris Pro variants, or all the chips in the (what I assume will be named the 5000) -series will have less cache than their 4000-series equivalents?
No idea, and Intel ain't saying. Guess we'll be waiting for a more detailed leak, or Intel's official announcement next year, to find out.
If it's like Intel's current Haswell SKUs, you'll get a minor loss of L3 in exchange for a much larger L4/eDRAM block. ( http://ark.intel.com/compare/75131,76088,76642,75122 )
EDIT: of course, being K-seires, we may well see full cache allocations...
Separate names with a comma.