AMD - Piledriver chitchat

[DanceswithUnix]

I don't know why I've only just spotted this, but supposedly there are a ton of different Haswell die variants.
http://wccftech.com/haswell-die-conf...idge-revealed/

Is the blacked out part on those die shots the bit the NSA made them put into the design?

Edit to add: Smaller memory controller can also mean lower speed. It takes a large area of silicon to drive the extra energy into wires to make them go faster. For laptop use perhaps they don't want fast ram anyway so save power and cost at both ends of the memory bus. I could check up on this guess with some googling, but am kind of busy for that sorry.

[scaryjim]

... It could be a photoshop as it does not look like a proper AMD slide.

I have some doubts over its validity,as I doubt if AMD would bother with AM3+ that long ...

Frankly it looks like a complete hash-up to me. Some of the earlier stuff is obviously based off the latest desktop roadmap [url="http://www.extremetech.com/gaming/171193-amds-2014-roadmap-a-hazy-future-for-amds-desktop-x86-chips-doubling-down-on-arm"here[/url], with some added rumours at the end just to spice it up. But it looks so far removed from a genuine AMD slide that I'd just ignore it, personally.

Edit to add: Smaller memory controller can also mean lower speed. ....

That's the way AMD went with the Hawaii memory controller of course, and would make a lot of sense. It's more likely than single channel (IMNSHO ) as even Bay Trail has a dual channel memory architecture now. Why on earth would Intel give their small-core offering DC but restrict any of the big-core offerings to SC...?

[watercooled]

It does look a bit unprofessional for an official slide IMHO; I know it's nitpicking but there are little boxes in the corners of each column which look out of place, the alignment and spacing is all over the place, it just doesn't look like AMD's style, and the font looks wrong.

Edit: Missed a page there. Yeah you could be right about the memory controller. 8 still seems quite a lot for one line, but based on what I found about die shapes of various products, it might have some truth in it. I can't locate the source used on the Wiki page claiming the die sizes (possibly Anandtech, with a bit of speculation thown in), which I assumed to be correct a while back in the thread for my first ramble on the subject (about Pentium/Celeron).

[CountBartok]

Some details about the clockspeeds of the desktop Kaveri CPUs:

http://prohardver.hu/teszt/mit_tudha...eri_igp-t.html

It seems the A10 7850K has a maximum clockspeed of 4GHZ. For the 20% projected improvement in performance over Richland,that would mean a decent boost in IPC(if you compare it to the A10 6800K),since the A10 7850K has 3.7GHZ to 4GHZ clockspeeds and the A10 6800K,4.1GHZ to 4.4GHZ clockspeeds. Even if we look at the earlier AMD claims of 10% to 15% performance improvement per generation, IPC probably has improved around 20% to 30% it seems.

A shorten version in English http://vr-zone.com/articles/new-kave...ked/65727.html
As google translate is not that good with Hungarian.

[CAT-THE-FIFTH]

Sigh!! TR is the next to get onto the Nvidia prodded bandwagon about the R9 290X and the "performance variability" between samples.

Its a sad state of reviewing in the sector,when the competitors cards have not undergone similar "retail testing" when issues have been highlighted with Nvidia Boost 1 and 2 downclocking on reference review sample GTX660TI,GTX760 and Geforce Titan cards,and some of these drops were huge.

AMD's marketing needed to get their arse out of gear and try to be more proactive in highlighting the Nvidia boost problems. They are foolish IMHO and needed to play the same game.

Edit!!

Flipping heck. Was reading the TR article again and it seems Nvidia was shipping R9 290X cards to reviewers to test.

That explains it!!!

[CAT-THE-FIFTH]

Some benchmarks of a Kaveri ES:

http://browser.primatelabs.com/geekb...aseline=223722

The 3.5GHZ Kaveri ES sample is matching a 4.1GHZ to 4.4GHZ Richland A10. I suspect that Turbo might not be enable on the ES sample,which might mean a decent IPC increase,and since the A10 7850K runs at between 3.7GHZ to 4GHZ,that means a reasonable increase in performance.

[CAT-THE-FIFTH]

Added to previous post.

[CountBartok]

Sigh!! TR is the next to get onto the Nvidia prodded bandwagon about the R9 290X and the "performance variability" between samples.

Its a sad state of reviewing in the sector,when the competitors cards have not undergone similar "retail testing" when issues have been highlighted with Nvidia Boost 1 and 2 downclocking on reference review sample GTX660TI,GTX760 and Geforce Titan cards,and some of these drops were huge.

AMD's marketing needed to get their arse out of gear and try to be more proactive in highlighting the Nvidia boost problems. They are foolish IMHO and needed to play the same game.

Edit!!

Flipping heck. Was reading the TR article again and it seems Nvidia was shipping R9 290X cards to reviewers to test.

That explains it!!!

And these reviewers used these cards?
They didn't think....erm hold on Nvidia are AMD's largest competitor in this market why would they send me there card for review?
I am kind of lost for words.

Anyway AMD should ask for these cards back and check them out.

[CountBartok]

Some benchmarks of a Kaveri ES:

http://browser.primatelabs.com/geekb...aseline=223722

The 3.5GHZ Kaveri ES sample is matching a 4.1GHZ to 4.4GHZ Richland A10. I suspect that Turbo might not be enable on the ES sample,which might mean a decent IPC increase,and since the A10 7850K runs at between 3.7GHZ to 4GHZ,that means a reasonable increase in performance.

http://browser.primatelabs.com/geekb...aseline=223722

Same cpu vs i5-2500k. i5 has double the ram and different OS used so not sure how much you can read into it.

[scaryjim]

Flipping heck. Was reading the TR article again and it seems Nvidia was shipping R9 290X cards to reviewers to test.

Not quite - nvidia paid for two retail 290Xes to be shipped directly from newegg to Tech Report. Which means that, in an effort to highlight a difference between press sample and retail cards, nvidia gave AMD two sales of their highest end, highest margin, graphics card. Somehow, I doubt AMD are too bothered about that

What they should be bothered about is that fact that there is clearly a difference in behaviour between TechReport's press sample and the retail cards. That could be random variance - AMD shipped out press samples and TR happened to get a good one - but without testing every single 290X press sample using the same methodology and looking in great detail there's no way of knowing. AMD were clearly right about the fans on retail samples not spinning at the correct RPM when set by duty cycle, as this is fixed in the driver update which sets the RPM directly. But it looks from TR's figures like that's not the full story - at least with their original press sample.

My only real concern with TR's testing is that they use quiet mode rather than uber mode - seems moot to complain about a card having lower performance when you're choosing to run it in the mode that deliberately sacrifices performance. If the performance of retail cards was lower in uber mode, then you've got more of an issue. But they didn't test that

[CountBartok]

Thanks scaryjim for clarifying that was feeling a bit cross.

[CountBartok]

Interview with Rory Read from Washington post http://www.washingtonpost.com/busine...2e1_story.html
and AMD reports resurgence in the UK system builder market http://www.microscope.co.uk/news/224...builder-market.

[CAT-THE-FIFTH]

Not quite - nvidia paid for two retail 290Xes to be shipped directly from newegg to Tech Report. Which means that, in an effort to highlight a difference between press sample and retail cards, nvidia gave AMD two sales of their highest end, highest margin, graphics card. Somehow, I doubt AMD are too bothered about that

What they should be bothered about is that fact that there is clearly a difference in behaviour between TechReport's press sample and the retail cards. That could be random variance - AMD shipped out press samples and TR happened to get a good one - but without testing every single 290X press sample using the same methodology and looking in great detail there's no way of knowing. AMD were clearly right about the fans on retail samples not spinning at the correct RPM when set by duty cycle, as this is fixed in the driver update which sets the RPM directly. But it looks from TR's figures like that's not the full story - at least with their original press sample.

My only real concern with TR's testing is that they use quiet mode rather than uber mode - seems moot to complain about a card having lower performance when you're choosing to run it in the mode that deliberately sacrifices performance. If the performance of retail cards was lower in uber mode, then you've got more of an issue. But they didn't test that

It goes back to what the HardOCP editor said about Nvidia during NDA meetings trying to make reviewers look at the R9 290X only in quiet mode,and there is a chart to show this from Nvidia. Guru3D said exactly the same about TH. Any review site using quiet mode only is doing so because of the urging of Nvidia. You would test both modes not one.

If anything it is still of much concern that even the GTX760 with reference coolers has throttling issues. We are talking 100MHZ+ throttling on a review sample. That makes 3 Nvidia reference cards with large throttling issues with Nvidia supplied samples. All these cards had issues over longer test runs,with degraded performance. Most test runs in reviews are lucky to go past 30 seconds to one minute. The tech world is crying crocodile tears over the R9 290X performance issues but gives multiple Kepler based cards a pass,when used in a computer case or run longer than 5 minutes? LOLWTF?

Sorry,but does it not make most of the benchmarks run by many websites,useless??

When four different review sites have shown this,one has to ask what the heck is the English language computer parts review world doing by only investigating issues when it is an AMD card,especially at the urging of their competitor?? Three of those who looked at the Kepler problems were not English language websites.

Knowing how Nvidia Turbo boost works with no real upper limit,why have most of the sites bothered to test multiple Nvidia retail cards too?? Knowing Nvidia past history,it seems too many sites are so terrified that Nvidia will cut them off(its happened before),that they don't want to push too far. Remember the bumps issue was pushed by CD,and how many of the tech world bothered to look into that?? That was nearly a quarter of a billion dollar problem.

I really think it is time AMD plays the same tactic against Nvidia and they deserve it,and I only think the computer hardware review world has themselves to blame if it happens. Nvidia have managed to piss off Microsoft,Sony and Intel in the past,and yet AMD is the only one who seems to put up with this nonsense from them. The same company behind astroturfers,ie, Rollo and the Nvidia Focus Group.People say that AMD must be already doing it,but yet its funny how we hear nothing about them doing such things themselves. Its time they push hard back.

[kompukare]

Remember the bumps issue was pushed by CD,and how many of the tech world bothered to look into that?? That was nearly a quarter of a billion dollar problem.

$250 million was only the least that Nvidia managed to get away with. They basically denied the problem and then only settled for as little as possible after a court case (and a bunch of irate OEMs, I guess). But basically the whole thing was totally secretive and the far from generous (miserly is a better word); the total opposite of Intel and SB chipset problem.

And the problem was a lot more widespread than G84 and G86 since I personally saw G92 as well and nForce 7150 chipset all dying of the same symptoms. No, if Nvidia behaved like Intel (that is stood by their products) it would have cost them probably four times as much or more.

[CAT-THE-FIFTH]

It seems I was right about the slide:

http://www.gamersnexus.net/news/1240...pus-fx-not-eol

[Noxvayl]

Good to see confirmation of that. I hope that DDR4 comes soon because I would like to see what that would do for APU's. If they are waiting for DDR4 to be released to update their top tier platform then I can understand, I am thinking the same with regards to upgrading. I don't want to spend on a new platform only for a significant technology update to come around the corner reducing the availability of my current system components. I had problems with DDR2 and that, still got a few machines using DDR2 and can't get them more RAM because it costs more than the system is worth.

I look forward to what we get next year, I will most likely be upgrading my older systems with Kaveri to save power and get all my systems using DDR3 so that I have some RAM to share when I upgrade my main PC in 2015.