Monday, January 23rd 2012

Radeon HD 7950 Specs Confirmed in GPU-Z Screenshot

Here is the first GPU-Z screenshot of a Radeon HD 7950 graphics card. Although put into a screenshot with ASUS GPU Tweak tool and Republic of Gamers-themed GPU-Z, the card doesn't appear to be an ASUS-made one, and is more likely HIS or PowerColor. The device ID checks out with the one HD 7950 has been associated with. Most other features match expectations. The HD 7950 is carved out of the 28 nm "Tahiti" GPU, with 28 GCN compute units (CUs) active, totaling 1,792 stream processors and 112 TMUs. The ROP count is untouched at 32, so is the memory, that's 3 GB GDDR5 across a 384-bit wide memory interface. With a memory clock of 5.00 GHz effective, it's churning out 240 GB/s of memory bandwidth. It's just the core clock speed where we weren't on target (from the previously-expected 800 MHz), either 880 MHz is the reference core clock speed of HD 7950, or this particular card is a factory-OC variant (PowerColor HD 7950 PCS?).

Source: DonanimHaber
Add your own comment

22 Comments on Radeon HD 7950 Specs Confirmed in GPU-Z Screenshot

#1
Yellow&Nerdy?
Nice to see the more affordable cards appearing. I wonder if this thing can unlock to a 7970? If they do, they will sell like hot cakes. This should also overclock like hell. Me personally am waiting for the 7870, hopefully it will be sub-300$ and overclock just as well as Tahiti.
Posted on Reply
#2
radrok
If anything is untouched other than clocks and shaders then it might have a good chance to unlock, we'll know at the launch though :)
Posted on Reply
#3
Isenstaedt
Looks like it will perform very close to the 7970.
Posted on Reply
#4
H82LUZ73
why is the GPUZ reporting PCIE at X4 1.1?

Yeah I too think the 7950 will unlock to 7970:rockout:
Posted on Reply
#5
ebolamonkey3
Nvidia needs to come out w/ their GTX 6xx cards fast.
Posted on Reply
#6
radrok
by: H82LUZ73
why is the GPUZ reporting PCIE at X4 1.1?
The link width and speed downclocks itself, it's a feature implemented on newer cards, if you check close to the PCIe specification on GPUz (0.5.8 or higher) there is a question mark, that triggers a render test to show the real link speed :)
Posted on Reply
#8
afw
if perf is equal to GTX580 and if the price is $400 or less ... it would be just perfect ... :)
Posted on Reply
#9
Completely Bonkers
I dont understand how "Radeon HD 7950 Specs Confirmed by GPU-Z Screenshot"... isnt the data (ie specs, ROPs, shaders, DX etc.) in GPUZ driven by an internal database? So based on product number Device-ID, it looks this information up. So what you mean is that Wiz has already added the specs into the 5.8 database? Nothing more... unless you are referring to clock speeds ;)

++++++++
EDIT
W1zzard answers here: http://www.techpowerup.com/forums/showpost.php?p=2525710&postcount=18
Posted on Reply
#10
Dj-ElectriC
I just love how everything supposedly leak
#AnythingToKeepNDAContractSquickyClean
Posted on Reply
#11
ice_v
by: Completely Bonkers
I dont understand how "Radeon HD 7950 Specs Confirmed by GPU-Z Screenshot"... isnt the data (ie specs, ROPs, shaders, DX etc.) in GPUZ driven by an internal database? So based on product number Device-ID, it looks this information up. So what you mean is that Wiz has already added the specs into the 5.8 database? Nothing more... unless you are referring to clock speeds ;)
I was also wondering about that...but quite sure there's a logical explanation for that :o

Anywho, I would like to point out the fact that according to several tech fellows and IT news journalists these new boards, (but more accuratly the chips) will feature -among other goodies- a distinguishably overall high overclockability. Now is it just me, or is this beginning to sound like bullsh*t marketing aimed at those pro-overclocker wannabes consumers?

Example:
Engineer: Sir, here are the final results. According to our benches I'd suggest we set the default core to 980/940Mhz (7970/7950), with at leats another 100MHz to maximum 150MHz margin for an overall stable OC we should be ok...

CEO: Yeah I guess that should...

Head Marketing: FUCK THAT! Cut it down by at least 50MHz! So manufacturers can stamp on their top models a "200MHz OverClock" Shinny Golden Logo and prepare to have an early Christmas :pimp:
(imho)
Posted on Reply
#12
R_1
Expect initial overclock to be slightly higher for hallo effect, and with mature drivers to fell down. Also what is reported is max overclock in a single game. That is not representative, or 24/7 stable, nor every game stable. There will be games , where overclock can be problematic. Same goes for seasonal changes in ambient temperature and it's effect on max overclock. When you go for high overclock , it is better to rump up the fan speed at 50% and above. Increase noise is associated with that. A typical fan will fail after 2-3 months, 24/7 operation at increased speed - ask bitcoin miners. As a user I will be glad to buy new GPU that super fast in every game and will stay that fast in next 12-24 months. That is not the case with HD7000. They are not that fast even at 1080p, so overclock is needed for higher res and Ultra in-game settings. Well, it will give few FPS , but wont make your 35-40FPS to much comfortable 60-80FPS. I will personally wait for faster GPU, preferably at low price too, something like SB 2500k vs. any Bulldozer.
Posted on Reply
#13
xenocide
by: ebolamonkey3
Nvidia needs to come out w/ their GTX 6xx cards fast.
Why? In terms of costperformance they still offer competative products. The HD7970 costs about $50-100 (10-20%) more than a GTX580 and performs 10-20% better (at least). Nvidia is fine waiting. If the HD7950 comes out and outperforms the GTX580 at a lower cost, all they have to do is drop the price--something AMD cannot do right away.
Posted on Reply
#14
Completely Bonkers
by: xenocide
Why?
I'm about to upgrade a workstation. A 560Ti is my target performance. But I would like this horsepower with a low power commitment. I want cool, silent, and I know it aint going to happen, but I would love single slot. My workstation PCIe slots are full due to a couple of nasty gigabyte iRAMs in RAID0.

So 6xx series with 30-40% lower watts/performance is where I'm looking.
Posted on Reply
#15
xenocide
by: Completely Bonkers
I'm about to upgrade a workstation. A 560Ti is my target performance. But I would like this horsepower with a low power commitment. I want cool, silent, and I know it aint going to happen, but I would love single slot. My workstation PCIe slots are full due to a couple of nasty gigabyte iRAMs in RAID0.

So 6xx series with 30-40% lower watts/performance is where I'm looking.
Fair enough, but I was under the impression the person I quoted was saying they needed to hurry up because otherwise they would fall behind. I would rather Nvidia took their time and offered the best product they could rather than rushing like they have in the past and giving us a GPU line that only runs at 75%.
Posted on Reply
#16
H82LUZ73
by: radrok
The link width and speed downclocks itself, it's a feature implemented on newer cards, if you check close to the PCIe specification on GPUz (0.5.8 or higher) there is a question mark, that triggers a render test to show the real link speed :)
Never knew the cards will down clock the pcie lanes too.I thought they would just down the mem-gpu volt and shut off the fans.
Posted on Reply
#17
W1zzard
by: Completely Bonkers
I dont understand how "Radeon HD 7950 Specs Confirmed by GPU-Z Screenshot"... isnt the data (ie specs, ROPs, shaders, DX etc.) in GPUZ driven by an internal database? So based on product number Device-ID, it looks this information up. So what you mean is that Wiz has already added the specs into the 5.8 database? Nothing more... unless you are referring to clock speeds ;)
everything is read from the card except for gpu name, process size, die size, release date, transistors and directx
Posted on Reply
#19
Xaser04
by: R_1
That is not the case with HD7000. They are not that fast even at 1080p, so overclock is needed for higher res and Ultra in-game settings. Well, it will give few FPS , but wont make your 35-40FPS to much comfortable 60-80FPS. I will personally wait for faster GPU, preferably at low price too, something like SB 2500k vs. any Bulldozer.
Errrr my HD7970 at a relatively mild 1075/6000 is 60% faster in BF3 at 3560x1920 "high" settings than my old HD6970 (Framerate wise I went from mid 30's to mid 50's and the 7970's minimums are higher than the 6790's average....). At 1250/6500 it will be roughly as fast as a HD6990 without any of the muilti card issues.

How can this be considered "not that fast"?

The HD7950 does look like an interesting card. Unlock ability is still up in the air though. Gibbo (purchasing manager over on OCUK forums) flashed a HD7950 with a HD7970 bios and GPU-z still read 1792 shaders. It also didn't perform the same as a HD7970 at the same clocks.

The potential is still there I think but it could be a case that the HD6950/HD6970 is the exception rather than the norm. The HD5850 didn't unlock after all, so who knows?!
Posted on Reply
#20
R_1
I am playing BF3 too, and can tell you that there is a significant visual difference between High and Ultra ingame settings. Also there are two independent AA modes AA deffered and AA post, with high impact on performance. Ambient occlusion are pretty tough to GPU too. So you are playing on console like quality and that suits you. Nice for you.
Posted on Reply
#21
davele
It is OC for a reason

by: ice_v
I was also wondering about that...but quite sure there's a logical explanation for that :o

Anywho, I would like to point out the fact that according to several tech fellows and IT news journalists these new boards, (but more accuratly the chips) will feature -among other goodies- a distinguishably overall high overclockability. Now is it just me, or is this beginning to sound like bullsh*t marketing aimed at those pro-overclocker wannabes consumers? ...
Yes there is likely to be an element of Marketing spin. But also in any design there are trade-off's. Most Engineers design items to withstand more stress than they are rated for. This permits extreme transient loads that do not risk permanently damaging / weakening the item.
eg: Bridges handle 4X the load ever expected of them, this reduces the potential to be weakened by fatigue, corrosion or extreme loads. Eg: Race Car engines, are better balanced, but even then the higher revs dramatically decrease their life.
Increasing the Clock Speed exponentially increases the Power Consumption & Heat Output of CPU’s & other electrical devices. Ie: The Failure Rates of 15,000RPM Hard Drives is much higher than those running at 5,400rpm.
So they ship at a spec they believe will be economical to run, to cool & with a lifespan long enough to run without error till it becomes obsolete.
If you are happy paying 2X for Cooling & 2X electricity for 20% perf increase & dramatically shortened component life. Then you will OC to the level you feel comfortable.
The trick is knowing what & how much you are giving up.
Posted on Reply
#22
Xaser04
by: R_1
I am playing BF3 too, and can tell you that there is a significant visual difference between High and Ultra ingame settings. Also there are two independent AA modes AA deffered and AA post, with high impact on performance. Ambient occlusion are pretty tough to GPU too. So you are playing on console like quality and that suits you. Nice for you.
Just LOL.

Console like quality? Since when is 3560x1920, high settings, Post AA and SSAO console like quality?

Considering I have extensively tested BF3 (I even have a thread about it FFS) I think I am quite versed in the relavent quality settings and there performance impact. The difference between High and Ultra in game is relatively minimal unless you like standing still and getting shot at.

Oh and for reference my HD7970 clocked at 1125/6600 pulls 58 FPS average at 1080p, Ultra Settings with 4x deffered AA. This is nearly twice as fast as my old HD6970....

Yes this damn HD7970 sucks....

Back to playing BF3 on my console quality Eyefinity setup :rolleyes:
Posted on Reply
Add your own comment