Wednesday, May 15, 2013

Haswell dream shattered even before the release? Haswell is a flop?

Yesterday I came across a very disturbing news. A Chinese site had posted benchmark scores of the flagship Intel Core i7 4770K (@stock as well as at @4.5GHz) and the results were pretty bad. They were comparing it with the flagship model from the 3rd generation, namely the Core i7 3770K. They were not just posting the scores. They were also showing the screenshots of CPU-Z, AIDA and other benchmark programs. 

Let's be honest now. Nobody likes those Chinese sites. They are ugly, plus, nobody can read them. Google translate can do a decent job, but still the quality of content in those sites are BAD. Luckily, Xbitlabs recently extracted those results and posted on their website. Check it out by linking on the following link.
Web-Site Publishes Fully-Fledged Intel Core i7-4770K “Haswell” Review 

The following is an extract of the results posted on that site.

That really sucks! The only thing that might sound interesting is the last test, but that is all due to the better iGPU in Haswell. It has nothing to do with the CPU raw power.

But we knew that the performance improvement from the CPU side was not going to be fantastic. Intel was mostly concerned about two things. Improving the power efficiency (for tablets and ultrabooks) and improving the performance of the integrated graphics potion (for non-enthusiast crowd). They seems to have done that alright.

But we expected a bigger than a single digit percentage boost, did we not? In some tests, Haswell is even slower than its predecessor. That's unacceptable. The instructions per clock (or IPC) hasn't improved much, from even SandyBridge. For example, the 4770K@4.5GHz does 1M Super Pi calculations in 8.018s and my 2600K at same clocks does it in 8.344s. Just a mere 4% improvement at the same clocks. That's almost no architectural change IMO. No idea what those Intel engineers were doing for more than a year.

However, what every geek was waiting to see was how well the Haswell CPU overclocked compared to the previous generations. There were leaked overclocking results showing the CPU running at 6GHz with just 1.2V and 7GHz with a whopping 2.56V. It is possible that CPU-Z wasn't reporting the correct amount of Voltage because the Voltage Regulator Modules (VRMs) are moved from the motherboard on to the CPU itself. Anyways, 6GHz @1.2V seems like it can even do that on air, right? IvyBridge could do OK with 1.2V when it comes to the temperature. There is no need to doubt that would not be the case with Haswell because Intel did not change the process node from 22nm. 

But what that Chinese website reports is very troubling. They are saying that their 4770K CPU @4.5GHz could not complete Cinebench test because the internal temperature of the internals rose up too rapidly that their water cooler couldn't keep up with that rate, and ultimately crashing the PC. We don't really know if this is true or if they applied too much voltage to the CPU that it overheated too fast or if the reason for crash was something completely unrelated to it being a Haswell CPU. 4.5GHz seems to low to require a water cooler and that water cooler too being insufficient. Besides, this was a engineering sample. I don't think there would be significant difference between the retail product and this, but we shall see.

The CPUs are coming out in early June so it's only one month to go before we see the real things in action. But so far, things are not looking great for the enthusiast desktop user. If these results are true, I'll probably stick with my trusty Sandybridge CPU. Heck, I might even consider the IvyBridge-E platform to upgrade to. While I'm not happy that I won't be able to waste some of my hard earned cash this time round if Haswell turns out to be crap, I'm glad that this might actually give AMD a fighting chance when the Steamroller comes out later this year. (AFAIK, it's coming out in September so not so far away from now.)

No comments:

Post a Comment

Related Posts Plugin for WordPress, Blogger...