Overclock.net › Forums › Industry News › Hardware News › [Guru3D] AMD faces Lawsuit over Core Count on Bulldozer
New Posts  All Forums:Forum Nav:

[Guru3D] AMD faces Lawsuit over Core Count on Bulldozer - Page 17

post #161 of 333
Quote:
Originally Posted by tajoh111 View Post


It does come with 4gb of memory however. The only thing advertised on the box is the amount of memory it has. And more importantly, most people bought the gtx 970 for the performance characteristics. It had unrivaled value upon launch and collapsed the pricing of videocards around it. It was a good product and cratered pricing of all cards 550 and less.

If you think you got ripped off on the gtx 970 and should have bought a gtx 980 in the beginning, then you might as well not buy a new videocard ever again(particularly with the cost of 16nm cards). The gtx 970 offered outstanding value that you see once every few generations.

More people are likely going to be effected from the fx8150 not behaving like a true 8 core processor than the last 0.5 gb partition of memory on the gtx 970. When the performance characteristics of bulldozer not behaving and performing like a true octocore chip, because shown in reviews, the reviews showed it in a negative light and the price of the chips dropped 30% shortly after. The gtx 970 still holds its value because it is still a good value, even with the last 0.5 gb partition handicapped. fx 8150, you got a processor that only behaved like one in a few scenario's like encryption and zipping files. For gaming, general office use and productivity, it didn't perform as one would expect.

People still got the performance they paid for with the gtx 970. No one was bragging they got the gtx 970 for the memory, but alot of people were gloating that AMD was offering an 8 core processors for the price of Intels, 4 core processors.

AMD was hardly being transparent as they were talking up the real core angle alot. They were also saying stuff like IPC had gone up. They were throwing out a bunch of garbage to get people to hold off on their CPU purchases until they could lay their cards. Reality showed everything was a lie, so all AMD could do is play up the 8 core angle.

And people still got the performance they paid for with Bulldozer chips. Surely you read the reviews and knew about its performance before you bought the product? I mean I've heard the "you read reviews before you bought the product, the performance hasn't changed, so why are you complaining" line way too many times, so I don't see how that doesn't equally apply in Bulldozer's case as well.
post #162 of 333
Modules are exactly how they sound, modules. Modular system. AMD's design was a way to be cheap and escalate production by having modules with two cores that can be connected to additional modules. Zen will be a four core module capable of hyperthreading. I think what killed the FX series was the fact it also didn't hyperthread.
Power Tower
(22 items)
 
SteamBox
(9 items)
 
Doge Miner
(7 items)
 
CPUMotherboardGraphicsRAM
Ryzen 1700X AX370-Gaming 5 AMD Radeon R9 200 Series G.Skill DDR4-2400 
RAMRAMRAMHard Drive
G.Skill DDR4-2400 G.Skill DDR4-2400 G.Skill DDR4-2400 Samsung 840 Pro 
Hard DriveHard DriveHard DriveHard Drive
CX300 Crucial 480GB Toshiba 4TB Toshbia 4TB Western Digital Black 1TB 
CoolingOSMonitorMonitor
h110i Windows 10 42" LG TV 20" Digitizer ASUS 
KeyboardPowerCaseMouse
Corsair Vengeance Mechanical Keyboard  850watt Vampire Gold Rated NZXT S340 Elite Corsair RGB FPS Mouse 
Mouse PadAudio
Borderlands Mousepad Realtek HD 
  hide details  
Reply
Power Tower
(22 items)
 
SteamBox
(9 items)
 
Doge Miner
(7 items)
 
CPUMotherboardGraphicsRAM
Ryzen 1700X AX370-Gaming 5 AMD Radeon R9 200 Series G.Skill DDR4-2400 
RAMRAMRAMHard Drive
G.Skill DDR4-2400 G.Skill DDR4-2400 G.Skill DDR4-2400 Samsung 840 Pro 
Hard DriveHard DriveHard DriveHard Drive
CX300 Crucial 480GB Toshiba 4TB Toshbia 4TB Western Digital Black 1TB 
CoolingOSMonitorMonitor
h110i Windows 10 42" LG TV 20" Digitizer ASUS 
KeyboardPowerCaseMouse
Corsair Vengeance Mechanical Keyboard  850watt Vampire Gold Rated NZXT S340 Elite Corsair RGB FPS Mouse 
Mouse PadAudio
Borderlands Mousepad Realtek HD 
  hide details  
Reply
post #163 of 333
"People still got the performance they paid for with the gtx 970."

The ROP count, cache size, and VRAM speed were all lower than what they paid for.

Nvidia claimed (and still does, falsely, on their website) that the card has 4 GB of 224 GB/s VRAM.

The card, in reality, has a 196 GB/s partition of 3.5 GB and a 28 GB/s partition of .5 GB. There is XOR contention between them.
Quote:
Originally Posted by Ryan Smith 
it is not possible to actually get that much bandwidth at once when doing a pure read or a pure write. In the case of pure reads for example, GTX 970 can read the 3.5GB segment at 196GB/sec (7GHz * 7 ports * 32-bits), or it can read the 512MB segment at 28GB/sec, but it cannot read from both at once; it is a true XOR situation. The same is also true for writes, as only one segment can be written to at a time.

Unfortunately what this means is that accessing the weaker 512MB segment blocks access to the stronger 3.5GB segment if both memory operations are identical; or put another way, using the 512MB segment can harm the performance of the 3.5GB segment. For example, if we want to issue reads to both segments at once, reading the 512MB segment blocks any other reads to the 3.5GB segment for that cycle. If the 3.5GB segment is blocked in this fashion and doesn't have a non-blocking write to work on instead, it would have to go idle for that cycle, which would reduce the effective memory bandwidth of the 3.5GB segment.

This means that taken over time in our example, the larger the percentage of the time the crossbar is reading the 512MB segment, the lower the effective read memory bandwidth would be from the 3.5GB segment.
Quote:
Originally Posted by Ryan Smith 
the 512MB segment essentially serves as an additional layer of memory between the main VRAM and system memory
A kind way of saying it's not really VRAM because its performance is too slow to be practical in that capacity.

Even if one takes his edge case scenario (striping when over 3.5 GB of VRAM is in use) to argue that 224 GB/s is possible, that still can't hand-wave the ROP count and cache size.
Edited by superstition222 - 11/7/15 at 9:36pm
post #164 of 333
One partition can read while the other writes. If they alternate, then they do reach the full 224GB/s. The effective memory bandwidth approaches 224GB/s the more evenly distributed the read and write operations are.

Cache and ROPs are a different matter altogether.

Quote:
Originally Posted by Faithh View Post

An ALU cluster isn't a processor nor a core. They're execution resources in the end. Like I've said before, ALU's can only execute/store while fetching decoding (theyre also units) is also a part of the processing cycle which is the front-ends job. You don't have a CPU just with a execution resources only. There has to be something to drive them.

Then it's clear you never read my response.

Quote:
Originally Posted by Zero989 View Post

Right, AMD said years ago that their true "native" design was a true quad core design which would lead to better scaling The terms were used interchangeably. Phenom didn't really scale well but this was ironed out. Now they screwed themselves AGAIN by designing a shared FPU per 2 cores that sometimes splits, resulting in crappy scaling. Now they have native sometimes non true, sometimes true 8 core...

Correct, four cores on one silicon die vs Intel's dual-die solution.

Correct, eight cores on one silicon die vs Intel's SMT-based solution.
Triumvirate
(20 items)
 
Osmium
(8 items)
 
For Sale: [FS] Z97 system: Xeon and RAM
$160 (USD) or best offer
CPUMotherboardGraphicsGraphics
i7-5775C ASUS Sabertooth Z97 Mark 2 Sapphire RX 480 (reference) MSI Low-Profile 750Ti 
RAMHard DriveHard DriveHard Drive
Corsair Dominator Platinum - 2x8GB Crucial M500 - 960GB Samsung 840 - 250GB WD Scorpio Blue - 1TB 
Hard DriveCoolingCoolingOS
ADATA SP900 - 64GB Scythe BIG Shuriken 2 Rev. B PNY Quadro 600 blower Windows 10 
MonitorMonitorMonitorKeyboard
ASUS VE247H - 1080p Gateway FPD1960 - 1280x1024 Samsung S20D300 - 900p, portrait Rosewill RK9000I - Cherry MX Blue 
PowerCaseMouseOther
Rosewill HIVE 650W Riotoro CR1080 SteelSeries Rival 100 CyberPower 1500PFCLCD 
CPUMotherboardRAMHard Drive
AMD A10-7870K ASRock FM2A88X-ITX+ G.Skill Ripjaws X - 2x4GB @ 2400MT/s Samsung 840 120GB 
CoolingOSPowerCase
Evercool EC-HPS-810CP Windows 7 Pro 64-bit PicoPSU-120-WI-25 Mini-Box M350 
  hide details  
Reply
Triumvirate
(20 items)
 
Osmium
(8 items)
 
For Sale: [FS] Z97 system: Xeon and RAM
$160 (USD) or best offer
CPUMotherboardGraphicsGraphics
i7-5775C ASUS Sabertooth Z97 Mark 2 Sapphire RX 480 (reference) MSI Low-Profile 750Ti 
RAMHard DriveHard DriveHard Drive
Corsair Dominator Platinum - 2x8GB Crucial M500 - 960GB Samsung 840 - 250GB WD Scorpio Blue - 1TB 
Hard DriveCoolingCoolingOS
ADATA SP900 - 64GB Scythe BIG Shuriken 2 Rev. B PNY Quadro 600 blower Windows 10 
MonitorMonitorMonitorKeyboard
ASUS VE247H - 1080p Gateway FPD1960 - 1280x1024 Samsung S20D300 - 900p, portrait Rosewill RK9000I - Cherry MX Blue 
PowerCaseMouseOther
Rosewill HIVE 650W Riotoro CR1080 SteelSeries Rival 100 CyberPower 1500PFCLCD 
CPUMotherboardRAMHard Drive
AMD A10-7870K ASRock FM2A88X-ITX+ G.Skill Ripjaws X - 2x4GB @ 2400MT/s Samsung 840 120GB 
CoolingOSPowerCase
Evercool EC-HPS-810CP Windows 7 Pro 64-bit PicoPSU-120-WI-25 Mini-Box M350 
  hide details  
Reply
post #165 of 333
Quote:
Originally Posted by superstition222 View Post

"People still got the performance they paid for with the gtx 970."

False.

The ROP count, cache size, and VRAM speed were all lower than what they paid for. The specs were false.

Nvidia claimed (and still does, falsely, on their website) that the card has 4 GB of 224 GB/s VRAM.

The card, in reality, has a 196 GB/s partition of 3.5 GB and a 28 GB/s partition of .5 GB. There is XOR contention between them.

Don't forget the micro stuttering and 3.5GB vRAM limits in some games.
post #166 of 333
Quote:
Originally Posted by Zero989 View Post

Don't forget the micro stuttering and 3.5GB vRAM limits in some games.
It's my understanding that the enthusiast community's microstutter findings are what forced Nvidia to out the true specs of the 970.

It is amazing, frankly, that people feel that 28 GB/s VRAM performance meets the reasonable expectation of performance for an enthusiast card given that it's half the speed of a relatively midrange card from 2007 using a clearly lesser form of VRAM.

I could see the excuses being taken seriously for a low or lower-midrange card. But, for a card that was sold in large quantities for SLI, precisely because it allegedly had the same amount of high-speed VRAM as the top-end card... it defies credibility.
Quote:
Originally Posted by Cynical Unicorn 
One partition can read while the other writes. If they alternate, then they do reach the full 224GB/s.
Ryan Smith said the only time striping will do this is when the VRAM in use is above 3.5 GB. It is clearly an edge case at best.

The more the slow partition is used the more it hampers performance. And, yet... if you want to ever reach 224 GB/s you have to be sure that there is plenty of data in it. rolleyes.gif

Buyers had the right to know that the VRAM scheme in the 970 had such problems reaching the advertised spec. Being able to possibly reach 224 GB/s, only when VRAM in use is above 3.5 GB (which, in turn, carries a general performance penalty) and only when there are no pure reads or pure writes happening, isn't good enough.

The VW vehicles can reach their claimed pollution specs when they think they're being tested by the EPA.rolleyes.gif
Edited by superstition222 - 11/7/15 at 9:52pm
post #167 of 333
Quote:
Originally Posted by CynicalUnicorn View Post

One partition can read while the other writes. If they alternate, then they do reach the full 224GB/s. The effective memory bandwidth approaches 224GB/s the more evenly distributed the read and write operations are.

Correct, four cores on one silicon die vs Intel's dual-die solution.

Correct, eight cores on one silicon die vs Intel's SMT-based solution.

And thus the fact that it's shared makes it impossible to even get a valid court case. It's not possible to create a winnable court case while not knowing how the processor works since its tricky. And if he know how it worked, the case wouldn't exist. It's actually quite humorous.
post #168 of 333

Bottom line is that AMD stated it before launch.it can handle 8 Fp instruction per cycle if they're 128 bit or Four 256 bit fp.this lawsuit says AMD bulldozer can't handle eight 256 bit Fp.

Question : Why the F* did you silence over 4 years? 4 Years??? It was stated Before Launch.You knew it before.

post #169 of 333
Quote:
Originally Posted by Defoler View Post

Source
So a new "Gate"?

I wonder how this was found, or lawsuit was up only now, and not much earlier.

Please. For the sake of the internet --- No more gate crap! Lol

AMD go caught in the act and someone wanted people to know about it haha.
    
CPUMotherboardGraphicsRAM
Intel®Core™ i7-6920HQ  MSI GT80S 6QF-072 MXM GTX 980 SLI Kingston HyperXImpact CL14 Dual Channel 8GB X4 
Hard DriveHard DriveOptical DriveOS
Samsung 950 Pro 512 GB X2 Samsung 850 Pro 2TB ASUS Blu-Ray/ Asus CD Burner Windows 10 Home 64-Bit (Came with laptop) 
MonitorKeyboardPowerCase
Acer HN274H BMIIID 3D 27-Inch LED Monitor - Black SteelSeries 6Gv2 Mechanical Gaming Keyboard Bui... 330W Power Brick MSI GT80S Titan 
MouseMouse PadAudioOther
Logitec Proteus Core Whatever works! Nahimic Sound Wireless Killer 1535 
  hide details  
Reply
    
CPUMotherboardGraphicsRAM
Intel®Core™ i7-6920HQ  MSI GT80S 6QF-072 MXM GTX 980 SLI Kingston HyperXImpact CL14 Dual Channel 8GB X4 
Hard DriveHard DriveOptical DriveOS
Samsung 950 Pro 512 GB X2 Samsung 850 Pro 2TB ASUS Blu-Ray/ Asus CD Burner Windows 10 Home 64-Bit (Came with laptop) 
MonitorKeyboardPowerCase
Acer HN274H BMIIID 3D 27-Inch LED Monitor - Black SteelSeries 6Gv2 Mechanical Gaming Keyboard Bui... 330W Power Brick MSI GT80S Titan 
MouseMouse PadAudioOther
Logitec Proteus Core Whatever works! Nahimic Sound Wireless Killer 1535 
  hide details  
Reply
post #170 of 333
Quote:
Originally Posted by Xuper View Post

Bottom line is that AMD stated it before launch.it can handle 8 Fp instruction per cycle if they're 128 bit or Four 256 bit fp.this lawsuit says AMD bulldozer can't handle eight 256 bit Fp.
Question : Why the F* did you silence over 4 years? 4 Years??? It was stated Before Launch.You knew it before.

Lawsuit process takes an enternity to get this far... Maybe its all click bait?
    
CPUMotherboardGraphicsRAM
Intel®Core™ i7-6920HQ  MSI GT80S 6QF-072 MXM GTX 980 SLI Kingston HyperXImpact CL14 Dual Channel 8GB X4 
Hard DriveHard DriveOptical DriveOS
Samsung 950 Pro 512 GB X2 Samsung 850 Pro 2TB ASUS Blu-Ray/ Asus CD Burner Windows 10 Home 64-Bit (Came with laptop) 
MonitorKeyboardPowerCase
Acer HN274H BMIIID 3D 27-Inch LED Monitor - Black SteelSeries 6Gv2 Mechanical Gaming Keyboard Bui... 330W Power Brick MSI GT80S Titan 
MouseMouse PadAudioOther
Logitec Proteus Core Whatever works! Nahimic Sound Wireless Killer 1535 
  hide details  
Reply
    
CPUMotherboardGraphicsRAM
Intel®Core™ i7-6920HQ  MSI GT80S 6QF-072 MXM GTX 980 SLI Kingston HyperXImpact CL14 Dual Channel 8GB X4 
Hard DriveHard DriveOptical DriveOS
Samsung 950 Pro 512 GB X2 Samsung 850 Pro 2TB ASUS Blu-Ray/ Asus CD Burner Windows 10 Home 64-Bit (Came with laptop) 
MonitorKeyboardPowerCase
Acer HN274H BMIIID 3D 27-Inch LED Monitor - Black SteelSeries 6Gv2 Mechanical Gaming Keyboard Bui... 330W Power Brick MSI GT80S Titan 
MouseMouse PadAudioOther
Logitec Proteus Core Whatever works! Nahimic Sound Wireless Killer 1535 
  hide details  
Reply
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Hardware News
Overclock.net › Forums › Industry News › Hardware News › [Guru3D] AMD faces Lawsuit over Core Count on Bulldozer