Bulldozer and Windows 7

[red1776]

This story is getting rather interesting and starting to be picked up by mainstream sites. There is (was) a website by a Quin Etiam fellow that from day one claims to have identified the scheduling problem in Windows 7 and his group referred to only as 'we' have a working patch that significantly increased Bulldozers performance. Either he did a very good job and cracked into Passmark verification, or he actually submitted a BD 8150 score with the fix applied showing a score that was north of the i72600K. he then reported that Passmark removed it as "irregular" and it was gone from the site.
The website (quinetiam,com) at the time posted that he would be releasing a registry fix within hours. I went back a few hours later and the site had been taken down and in it's place was a 'wallpaper' covered with "seized by FBI". a few days later I located his new site that spelled out in detail the scheduling/registry problem and that it was fixable and a patch was on the way...then that site disappeared abruptly. Now it appears that another site is going up. http://quinetiam.com/?p=66
I figured it was hoax of some kind until Hexus and TechReport picked up the BD/Win7 scheduling issue. i am assuming that a 'Patch' by anyone would fall into the 'reverse engineering' category, but who knows. it's fun to follow. Some of the percentages and language are the same as the Quin Etiam stories so they are obviously linked. I am not holding my breath, but some of the blue team fanboys are already getting bet out of shape.
I am not endorsing the validity of this, I have no idea what to make of it, but it is interesting to follow now that these guys are picking up on it.
here are the stories so far...
http://techreport.com/articles.x/21865
http://hexus.net/tech/news/cpu/32394-bulldozer-benchmarks-correct-definitive/

 

[dividebyzero]

I think the main problem -perception wise- is that the TR article goes against AMD's official stance. AMD's literature and J.Fruehe's forum posting marathon all claimed that filling modules with the most demanding threads was the way to increase throughput. The TR article amongst others implies that parking heavy computation threads across the modules is the way to go...so the conclusion is AMD expected higher clocks (base and turbo) to get the required performance from loading up individual modules while leaving other modules lightly tasked. I think that much is a given at this stage.
What then becomes odd is that AMD would need to acknowledge that BD acts best as a one main thread per core processor with the second core utilised for light tasking or left idle...kind of sounds like four cores+hyperthreading situation, no?

The other side of the conundrum is that AMD's fanbase seem to think that MS should somehow be implementing a scheduling workaround/hack into Win7...which is basically pie-in-the-sky thinking. When has an OS ever had such a major code overhaul -much less one that is effectively more than halfway through it's product cycle? Service packs don't roll like that.
Maybe someone introduces a hack for Win7, but I don't see it as anything MS would sanction (especially if it detracts from Win8 sales), nor would I see it as the epitomy of stability if someone implemented it.
Hopefully anyone running it gets the extra performance, but to my mind- and that of most people I've spoken to about the issue ( a couple of friends of friends of mine work for MS in this area) are of the opinion that hardware should pretty much work out of the box without the software enviroment having to change to suit one product. AMD seem to be paying the price for a product line that fell short in a couple of key area's and made no contigencies for plan B once clockspeed and power requirement failed to meet target...such as TR noted:

Trouble is, right now, Intel has much better OS and application support for Hyper-Threading than AMD does for Bulldozer. In fact, we're a little surprised AMD hasn't attempted to piggyback on Intel's Hyper-Threading infrastructure by making Bulldozer processors present themselves to the OS as four physical cores with eight logical threads

[Source]

On a side note, I think most enthusiasts would like to see a component work at it's optimum regardless of what company it originated from. The non-enthusiasts are the ones who tend to point, snigger and wail without providing anything of substance. This small episode of ignorance should pale into insignificance once team Green hit the forums on the 14th November (SB-E launch day by all accounts)

 

[red1776]

I agree, from reading all this, if a registry patch/schedule fix actually works (and at this point I think it probably would to some extent) it would point out that it is nothing more than hyper-threading. It will be interesting to see what someone comes up with, but all of this is kernel level stuff, and I don't see it not having major repercussions even should it speed up some things. nevertheless i am going to follow it. the strange thing is that it's not all around slow. running the right programs its very fast. if someone comes up with a patch that seems initially credible, I might set up a machine and give it a go. ..I need a new boat anchor anyway.

 

[dividebyzero]

I agree, from reading all this, if a registry patch/schedule fix actually works (and at this point I think it probably would to some extent) it would point out that it is nothing more than hyper-threading.

Problem being I think that MS will certainly break the patch at every opportunity, and any widespread use paints AMD as a bunch of incompetents. If it comes to fruition I'm hoping Vegas opens a book on the number of times

"Real men use real cores" - John Fruehe

gets referenced in relation to the thread discussions.

 

[red1776]

Problem being I think that MS will certainly break the patch at every opportunity, and any widespread use paints AMD as a bunch of incompetents. If it comes to fruition I'm hoping Vegas opens a book on the number of times

I'll take the over.

"Real men use real cores" - John Fruehe
gets referenced in relation to the thread discussions.

well I have 8 of them....now pass the quiche

 

[EXCellR8]

i haven't been able to try out a bulldozer CPU, but i hear they are decent for the money. i'm probably gonna wait for ivybridge but i'm sure those aren't gonna be budget friendly units, especially at launch.

so what's the bottom line on bulldozer? is it worth the money to upgrade from say a six core or even a fast quad? i just build a good friend of mine a new system and she want's to know if she should just pay a little more for something if it's worth it. her board is AM3+ but i don't really have any advice since i haven't worked with any of the new AMD's myself...

 

[dividebyzero]

i haven't been able to try out a bulldozer CPU, but i hear they are decent for the money. i'm probably gonna wait for ivybridge but i'm sure those aren't gonna be budget friendly units, especially at launch.

Should hit the same price points as Sandy Bridge. Whether you consider SB budget friendly or not is another matter. The die will be smaller even taking into account a larger GPU, and whilst the process is new and Intel need to recoup their investment, they do have customers other than themselves at their fabs for 22nm this time around.
Motherboards should also be at the same price point. Theoretically, any P67 or Z68 board with dual (or more) BIOS chips should be IB upgradeable. Z77 boards, aside from native USB3.0 support (as opposed to third party controllers) are supposed to have pretty much the same feature set as Z68.

IB isn't expected to radically overhaul Sandy Bridge's present performance - more a case of lower power usage, cooler running, higher potential overclock (6.3GHz + ~5% BCLK as opposed to 5.7GHz + ~5% BCLK), and maybe a slightly higher base/turbo clock...other than that it will business as usual. IB is more about getting closer to what Sandy Bridge was designed to be - i.e. an all purpose ( desktop and portable computing) series of parts.

I'll leave red to wax lyrical about the merits of BD...If you were prepared to wait for IB then maybe it could be also worthwhile seeing what Piledriver brings to the table.

 

[red1776]

Hi EX,
Unless you want one for 'enthusiast' purposes, I would recommend waiting for the next round of IB and Piledriver/Vishera. For heavily threaded stuff it's very fast, but good god does it use a lot of juice. If your friend already has an PII x 6, i would wait. I have mine at 4.9Ghz....and its fast...and yet, you will want to replace it when the next gen comes down, if that makes any sense. on top of that, I do a lot of heavily threaded work. So it does better for me than the majority of users.

 

[dividebyzero]

Those rascals over at Donanimhaber are reporting that the AMD's next chipset is in the works.

The good news is that the Piledriver chipset is socket AM3+ compatible.
The bad news is that the Piledriver chipset is socket AM3+ compatible.

The 1090FX (how's that for confusing) and 1070 don't seem anything more than a rehash of the 990/970:
2 x PCIex16 2.0* @ x16 ( or 4 x PCIe x16 @ x8) for the 1090FX, 1 x PCIe x16 @ x16 for the 1070.

At least the (SB1050) southbridge seems to have connectivity to burn ( 8 x SATA 6GB, native USB 3.0)

* Going to be kind of awkward selling a series of PCI-E Gen3.0 compliant graphics cards - the HD 7000 series- whilst having to note in the fine print that the cards would need an Intel CPU and chipset to utilise the PCI-E 3.0 specification.

According to TPU:

Some industry sources we spoke to commented that PCI-Express 3.0 chipset support is unlikely till Piledriver processors start selling

...which sounds very equivocal to me. Looks like AMD will try to squeeze as much mileage out of existing hardware as possible and hoping that that will be sufficient to keep people buying AMD until they junk the AM3 socket and move over to APU's.

 

[red1776]

...umm...wait FX 1090 I think you mean PII 1090....STOP THE MADNESS!!!!

Are you surmising that the next gen GPU's 7000 & 600 are going to actually need the Bandwidth of the 3.0 spec?

With everyone on every board giving the standard canned response to inquiries about the need for PCIE 3.0 width, i did dome reading and it seems that 3.0 may have some reduced latencies and such that make it faster even without saturation being an issue. i will try to find the article.

 

[Leeky]

The more I read about BD, the more I wonder if I'm just better off flogging off my AM3+ board and just going SB, or even SB-E, which is fast approaching.

I'm sure my outlook would be much different if I was you though Red.

 

[dividebyzero]

Are you surmising that the next gen GPU's 7000 & 600 are going to actually need the Bandwidth of the 3.0 spec?

Nope. But rest assured that PCIe 3.0 will be a marketing bulletpoint for both AMD and Nvidia, so how do AMD make a marketing case for the HD7000 series without highlighting that their CPU's and chipsets don't support the spec?

Since the PR and marketing is likely to be outsourced (as AMD seems to have fired all their PR people) it will be a good test for the new contractors.

With everyone on every board giving the standard canned response to inquiries about the need for PCIE 3.0 width, i did dome reading and it seems that 3.0 may have some reduced latencies and such that make it faster even without saturation being an issue. i will try to find the article.

True enough. Where it comes into play significantly is PCI-E based SSD's at are already constrained by PCIe x1 and x4, and more importantly PCI-E based RAID controller cards that are SATA 6GB compliant and designed or use with multiple (i.e up to triple figures) HDD's/SSD's...afaik LSI already has a 12GB/sec controller ready to roll in the enterprise sector.

EDIT: The reduced latency issue is actually detailed in the PCI Express 3.0 specification. Both reduced memory and atomic operations (resource allocation) latency. PDF here

The more I read about BD, the more I wonder if I'm just better off flogging off my AM3+ board and just going SB, or even SB-E, which is fast approaching.

Unless you're worried about depreciation and are considering an upgrade anyway, I wouldn't worry too much about the need to move to SB/SB-E. Most of the hoopla concerns barely discernable differences in most instances. I tend to take a longer view simply because the tech interests me and my recommendations usually translate into 20-30 customer builds and maybe half that number again for upgrades every year. I've got a small reputation for making those recommendations the right choices and that brings me repeat or referred custom.
If a new build is on the horizon, then, unless you plan on a full-on enthusiast rig with its attendant lack of cost/performance metric I would probably wait for Ivy Bridge and Z77 (not to mention that an SB 2700K would be half the price it is now when IB launches) at least. If upgrading isn't too much of a high priority then Haswell ( early 2013) is shaping up to be something special by all accounts ( 8 cores, 32MB L3 cache).