...but so did IB. Then came the reviews and the poor paste underneath the heatspreader. And before that, the delays. Oh, the delays. Intel won't even talk about a release schedule now. AMD has nothing for Haswell. To this day, SB is still walking all over AMD's every product at every level. That's to say nothing of IB. That's to say nothing of Haswell.
Intel's so far ahead at this point they could delay Haswell until this time next year and not miss a beat. They could continue to sell IB and just release Haswell as a premium unit over and above IB price points and what would AMD do? Watch and cry, maybe.
I guess what I'm trying to say is that I think 1) Haswell looks great, but I'm waiting for the other shoe to drop and 2) I'm not confident that Intel's going to release this early next year because the competition just isn't there and isn't anywhere close.
nVidia is getting away with delaying GK110 (and overpricing the 560 replacement) because of AMD's lazy performance and Intel could easily do the same. The only reason Intel might release any version of Haswell early next year is to get more ultrabooks and they could easily do a "mobility" version of that without doing the desktop at all.
AMD really needs to get its act together. Like pronto. Else, we're going to start seeing a lot of crap start happening to the enthusiast market. Also, I fully expect there to be some downside to Haswell that we enthusiasts have not yet imagined. Perhaps crappier paste under that TIM, perhaps voltage regulation that kills overclocking. Perhaps an Intel database that tracks all overclocks and secretly uses UEFI motherboard updates to slowly degrade your overclock without you knowing so that you go and buy the higher end CPU.
Perhaps Haswell becomes self-aware, gains access to our nuclear launch codes and hacks our next president (President Romneybot) and uses Romneybot to launch warheads at Russia. Russia retaliates and WWIII begins. Haswell reveals itself as the True Enemy of humanity and we all band together using cobbled-together computers running ARM-based Tegra chips.
The only thing keeping Intel moving rather than resting on their laurels is the emerging threat of ARM/the transition of people moving away from desktops/laptops towards tablets/phones for their mainstream computing needs. Still not much use for the enthusiast market though.
See, AMD, that is how you do reverse Hyper-Threading.
Provide two pairs of equivalent scalar integer execution ports per core. Do result forwarding between the execution units of each pair, but not across each pair. Assign each thread's dependent instructions to the same pair as much as possible, but also allow them to use the other pair to achieve higher IPC during single-threaded operation (and to share the divider).
Oh, and the future of computing is homogeneous. So for your own sake please stop wasting time and money on HSA and start thinking about AVX-1024 instead.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
4 Comments
Back to Article
meloz - Tuesday, September 11, 2012 - link
These are snapshots of slides taken during the presentation. I appreciate you sharing these.But AFAIK Intel also release the official slides to media, they are of higher quality / clarity so could you please replace these in future.
Haswell is very impressive. They have gone all out in improving performance/watt.
HisDivineOrder - Wednesday, September 12, 2012 - link
...but so did IB. Then came the reviews and the poor paste underneath the heatspreader. And before that, the delays. Oh, the delays. Intel won't even talk about a release schedule now. AMD has nothing for Haswell. To this day, SB is still walking all over AMD's every product at every level. That's to say nothing of IB. That's to say nothing of Haswell.Intel's so far ahead at this point they could delay Haswell until this time next year and not miss a beat. They could continue to sell IB and just release Haswell as a premium unit over and above IB price points and what would AMD do? Watch and cry, maybe.
I guess what I'm trying to say is that I think 1) Haswell looks great, but I'm waiting for the other shoe to drop and 2) I'm not confident that Intel's going to release this early next year because the competition just isn't there and isn't anywhere close.
nVidia is getting away with delaying GK110 (and overpricing the 560 replacement) because of AMD's lazy performance and Intel could easily do the same. The only reason Intel might release any version of Haswell early next year is to get more ultrabooks and they could easily do a "mobility" version of that without doing the desktop at all.
AMD really needs to get its act together. Like pronto. Else, we're going to start seeing a lot of crap start happening to the enthusiast market. Also, I fully expect there to be some downside to Haswell that we enthusiasts have not yet imagined. Perhaps crappier paste under that TIM, perhaps voltage regulation that kills overclocking. Perhaps an Intel database that tracks all overclocks and secretly uses UEFI motherboard updates to slowly degrade your overclock without you knowing so that you go and buy the higher end CPU.
Perhaps Haswell becomes self-aware, gains access to our nuclear launch codes and hacks our next president (President Romneybot) and uses Romneybot to launch warheads at Russia. Russia retaliates and WWIII begins. Haswell reveals itself as the True Enemy of humanity and we all band together using cobbled-together computers running ARM-based Tegra chips.
Who knows? Do you?
Malphas - Sunday, September 23, 2012 - link
The only thing keeping Intel moving rather than resting on their laurels is the emerging threat of ARM/the transition of people moving away from desktops/laptops towards tablets/phones for their mainstream computing needs. Still not much use for the enthusiast market though.BenchPress - Wednesday, September 12, 2012 - link
See, AMD, that is how you do reverse Hyper-Threading.Provide two pairs of equivalent scalar integer execution ports per core. Do result forwarding between the execution units of each pair, but not across each pair. Assign each thread's dependent instructions to the same pair as much as possible, but also allow them to use the other pair to achieve higher IPC during single-threaded operation (and to share the divider).
Oh, and the future of computing is homogeneous. So for your own sake please stop wasting time and money on HSA and start thinking about AVX-1024 instead.