Sapphire Radeon RX 7600 PULSE review
Gainward GeForce RTX 4060 Ti GHOST review
Radeon RX 7600 review
ASUS GeForce RTX 4060 Ti TUF Gaming review
MSI GeForce RTX 4060 Ti Gaming X TRIO review
GeForce RTX 4060 Ti 8GB (FE) review
Corsair 2000D RGB Airflow Mini-ITX - PC chassis review
ASUS PG27AQDM Review - 240Hz 1440p OLED monitor
MSI MAG X670E Tomahawk WiFi review
Mountain Makalu Max mouse review
AMD Hawaii R9 290X GPU Specifications ?
A set of AMD Hawaii R9 290X GPU Specifications have been revealed by the German website 3DCenter.org. It is unclear how valid the specs really are. Key points: 4 raster engines, up to 2816 shader units, little bit more than 900 MHz base clock, temperature-controlled boost clock, 28nm process, around 430 mm² die-size, infos about process and die-size from an AMD interview with Forbes.
The 3DCenter website itself is in German, so this is a very rough Google translation, make of the information what you want:
From the corresponding speculative thread of our forum come first reasonably reliable data for Hawaii graphics chip from AMD's Volcanic Islands-generation , which AMD 25 September in the Radeon R200 series will introduce. However, the data stated are not speculations or assumptions, but come from a seemingly informed user - however, there is the residual risk that this is deliberate disinformation by AMD or the information in the transmission may have been partially corrupted. Thus, the following specifications are hardly ever can be laid on the gold scale - on the other hand correspond to these specifications Hawaii chip quite the preliminary expectations , that are realistic in every case:
AMD Hawaii
- 28nm production at TSMC
- ~ 430mm ² chip area (18% on R1000/Tahiti)
- 2.0 GCN Architecture (R1000/Tahiti: 1.0 GCN)
- DirectX 11.2 (as R1000/Tahiti)
- 4 raster engines (+100% over R1000/Tahiti)
- max 2816 shader units aka 44 shader clusters - in any case more than the 2304 shader units, the GeForce GTX 780 (+37.5% compared to R1000/Tahiti)
- Chip clock (Hawaii XT): slightly more than 900 MHz (similar to the Radeon HD 7970, but lower than the Radeon HD 7970 "edition GHz")
- Temperature-controlled boost clock - the card is not able to play to their full performance due to limits on temperature and power consumption regularly
- probably only a 384-bit DDR memory interface (such R1000/Tahiti)
- Presentation: 25 September 2013
- For sale: probably mid / late October 2013
- Sell ??name: Radeon probably R9-290 (Hawaii Pro) & Radeon R9-290X (Hawaii XT)
| Technik | Taktraten | Perf.Index | |
|---|---|---|---|
| Radeon HD 7990 | AMD 2x R1000/Tahiti, 4 Raster-Engines, 4096 Shader-Einheiten, 256 TMUs, 64 ROPs, 2x 384 Bit DDR Speicherinterface | 950/1000/3000 MHz | 600% |
| nVidia GeForce 690 | nVidia 2x GK104, 8 Raster-Engines, 3072 Shader-Einheiten, 256 TMUs, 64 ROPs, 2x 256 Bit DDR Speicherinterface | 915/1019/3000 MHz | 580% |
| GeForce GTX Titan Ultra | nVidia GK110, 5 Raster-Engines, 2880 Shader-Einheiten, 240 TMUs, 48 ROPs, 384 Bit DDR Speicherinterface | ~950/3400 MHz | 530-550% |
| Radeon R9-290X (Annahme 1) |
AMD Hawaii, 4 Raster-Engines, 2816 Shader-Einheiten, 176 TMUs, 32-48 ROPs, 384 Bit DDR Speicherinterface | ~950/975/3500 MHz | 485-510% |
| Radeon R9-290X (Annahme 2) |
AMD Hawaii, 4 Raster-Engines, 2560 Shader-Einheiten, 160 TMUs, 32-48 ROPs, 384 Bit DDR Speicherinterface | ~950/975/3500 MHz | 465-490% |
| GeForce GTX Titan | nVidia GK110, 5 Raster-Engines, 2688 Shader-Einheiten, 224 TMUs, 48 ROPs, 384 Bit DDR Speicherinterface | 837/876/3000 MHz | 480% |
| nVidia GeForce 780 | nVidia GK110, 4 oder 5 Raster-Engines, 2304 Shader-Einheiten, 192 TMUs, 48 ROPs, 384 Bit DDR Speicherinterface | 863/902/3000 MHz | 440% |
| Radeon HD 7970 "GHz Edition" | AMD R1000/Tahiti, 2 Raster-Engines, 2048 Shader-Einheiten, 128 TMUs, 32 ROPs, 384 Bit DDR Speicherinterface | 1000/1050/3000 MHz | 390% |
AMD Hawaii is 30% smaller than Kepler GK110 - 09/17/2013 08:39 AM
Forbed interviewed and he Talks New Radeon Cards, Next-Gen Consoles, 7990 Criticism. According to matt Skynner their new GPU is smaller than TITAN. Considerably smaller. It should be around 423mm2. T...
AMD Hawaii GPU might launch on September 25 - 08/08/2013 08:10 AM
AMD is planning to introduce its Hawaii GPU at the end of September according to a report on DSB. A press event is set to take place on September 23rd in Hawaii and the public launch is expected on Se...
Blackops_2
Senior Member
Posts: 319
Joined: 2011-07-12
Senior Member
Posts: 319
Joined: 2011-07-12
#4658987 Posted on: 09/21/2013 06:34 AM
Because it carries 1000$ mark or because it supports double precision?
Idk, why so negative, Titan's SP and DP is around the same as 7970Ghz, its only faster because it has more ROPS, higher fillrate and more raster engines; which makes it 15-25% faster then 7970Ghz. Now add all that to R9-290X? with more cores and its min 20% faster for sure.
To clarify the 290x is the single Hawaii solution right? Or is it the double?
Because it carries the 1000$ mark and all these released specs are speculation. 2800+ cores would undoubtedly beat Titan but that seems like a lot for the same process and the same power envelop/efficiency.
Titan is 30% faster than the 7970GE and 40% faster than the 7970/680. 20% Faster than the 7970GE sounds about right, but it still doesn't match Titan, though would theoretically compete with the 780. I just felt like it would be hard to achieve much more on the 28nm process.
I'm not meaning to be negative per say but when they come out and state they aren't wanting to compete with Titan i just don't think it's about to blow the lid off Titan.
I'm expecting competition against the 780 for 500-550, hopefully it beats the 780 giving us a price war. If it's bundled with BF4 i'll be getting them for sure.
Because it carries 1000$ mark or because it supports double precision?
Idk, why so negative, Titan's SP and DP is around the same as 7970Ghz, its only faster because it has more ROPS, higher fillrate and more raster engines; which makes it 15-25% faster then 7970Ghz. Now add all that to R9-290X? with more cores and its min 20% faster for sure.
To clarify the 290x is the single Hawaii solution right? Or is it the double?
Because it carries the 1000$ mark and all these released specs are speculation. 2800+ cores would undoubtedly beat Titan but that seems like a lot for the same process and the same power envelop/efficiency.
Titan is 30% faster than the 7970GE and 40% faster than the 7970/680. 20% Faster than the 7970GE sounds about right, but it still doesn't match Titan, though would theoretically compete with the 780. I just felt like it would be hard to achieve much more on the 28nm process.
I'm not meaning to be negative per say but when they come out and state they aren't wanting to compete with Titan i just don't think it's about to blow the lid off Titan.
I'm expecting competition against the 780 for 500-550, hopefully it beats the 780 giving us a price war. If it's bundled with BF4 i'll be getting them for sure.
thatguy91
Senior Member
Posts: 6640
Joined: 2010-08-27
Senior Member
Posts: 6640
Joined: 2010-08-27
#4658989 Posted on: 09/21/2013 06:47 AM
Just hope the first cards come out with good cooling solutions
.
Just hope the first cards come out with good cooling solutions
.
yasamoka
Senior Member
Posts: 4872
Joined: 2009-08-29
Senior Member
Posts: 4872
Joined: 2009-08-29
#4659006 Posted on: 09/21/2013 08:30 AM
They fixed it in software (still Phase 1 with no Eyefinity support though). It's practically as good as Nvidia's hardware solution. Anyways I cast doubt over the hardware frame-metering in Nvidia's hardware since according to them, it's been there since G80 (8800GTX) but we all know that Nvidia suffered from horrendous microstutter (that improved every generation) until Kepler which finally *eliminated* it in a general sense.
Frame-metering is an algorithm that paces not when the frames are displayed, but when they are rendered (else you'd have frames being repeated). This tells me that it takes effect before the GPU gets to rendering the second, pushed frame. So it's earlier than when the GPU receives what it has to render.
The 7970 did have excellent DP performance compared to the GTX680 and it didn't sacrifice much, if any, gaming performance against the 680. However, it does consume more power.
Why would there be an artificial driver limit on either if they can cripple DP performance straight on the hardware? AMD's GCN is 1/4 DP:SP and it performs as expected, even with the FirePros thrown in.
The workstation cards have specialized tasks that they excel in, but that is not necessarily tied directly to DP performance.
Let hope they fix their horrendous crossfire issues in hardware this time. I believe I read somewhere that was their goal but I'll believe it when I see it. The 7970 was the last straw for me which pushed my back to nvidia.
They fixed it in software (still Phase 1 with no Eyefinity support though). It's practically as good as Nvidia's hardware solution. Anyways I cast doubt over the hardware frame-metering in Nvidia's hardware since according to them, it's been there since G80 (8800GTX) but we all know that Nvidia suffered from horrendous microstutter (that improved every generation) until Kepler which finally *eliminated* it in a general sense.
Frame-metering is an algorithm that paces not when the frames are displayed, but when they are rendered (else you'd have frames being repeated). This tells me that it takes effect before the GPU gets to rendering the second, pushed frame. So it's earlier than when the GPU receives what it has to render.
DP performance is completely irrelevant in the context of gaming. If AMD puts silicone towards DP performance then it'll be sacrificing gaming performance for DP compute performance ( completely irrelevant to 99.99% of all computer users). Not to mention that DP performance is artificially limited at driver level for consumer level cards so they can sell Quadros/FirePros/Teslas at much higher margins.
The 7970 did have excellent DP performance compared to the GTX680 and it didn't sacrifice much, if any, gaming performance against the 680. However, it does consume more power.
Why would there be an artificial driver limit on either if they can cripple DP performance straight on the hardware? AMD's GCN is 1/4 DP:SP and it performs as expected, even with the FirePros thrown in.
The workstation cards have specialized tasks that they excel in, but that is not necessarily tied directly to DP performance.
warlord
Senior Member
Posts: 2760
Joined: 2012-10-22
Senior Member
Posts: 2760
Joined: 2012-10-22
#4659018 Posted on: 09/21/2013 08:56 AM
^AMD never caught up to Nvidia's smoothness since forever, what SLI microstutter i am hearing about? LOL even 13.10beta is not making things smooth enough. They should learn something from nvidia's superiority in mgpu solutions. Maybe hardware pacing is better to be done than driver level.
^AMD never caught up to Nvidia's smoothness since forever, what SLI microstutter i am hearing about? LOL
even 13.10beta is not making things smooth enough. They should learn something from nvidia's superiority in mgpu solutions. Maybe hardware pacing is better to be done than driver level.
Click here to post a comment for this news story on the message forum.
Senior Member
Posts: 3202
Joined: 2005-10-24
Let hope they fix their horrendous crossfire issues in hardware this time. I believe I read somewhere that was their goal but I'll believe it when I see it. The 7970 was the last straw for me which pushed my back to nvidia.
Because it carries 1000$ mark or because it supports double precision?
Idk, why so negative, Titan's SP and DP is around the same as 7970Ghz, its only faster because it has more ROPS, higher fillrate and more raster engines; which makes it 15-25% faster then 7970Ghz. Now add all that to R9-290X? with more cores and its min 20% faster for sure.
DP performance is completely irrelevant in the context of gaming. If AMD puts silicone towards DP performance then it'll be sacrificing gaming performance for DP compute performance ( completely irrelevant to 99.99% of all computer users). Not to mention that DP performance is artificially limited at driver level for consumer level cards so they can sell Quadros/FirePros/Teslas at much higher margins.